📰 News — Awesome Speech Audio

Rebecca Bellan · AI News & Artificial Intelligence | TechCrunch

Until now, voice mode has only been available on Claude Haiku, Anthropic's faster but less powerful model. Now the company is making its Opus and Sonnet models available in voice mode, and extending its reach into apps like Gmail, Slack, and Canva. When Anthropic launched voice mode last year, it was primarily focused on delivering […]

NEWSJul 23, 2026

Runway launches AI model router as generative media gets crowded

The Media Router is a tool that automatically selects the best image, video, or audio generation model for a request based on whether a developer prioritizes quality, speed or cost.

Wednesday, July 22, 2026

NEWSJul 22, 2026

Google justifies its massive AI spending with a booming cloud business

Lucas Ropek · AI News & Artificial Intelligence | TechCrunch

Google's cloud business is thriving, as companies adopting its AI and AI infrastructure services help the tech giant to report record profits.

NEWSJul 22, 2026

OpenAI’s AI spending spree has ballooned to $750B

Tim De Chant · AI News & Artificial Intelligence | TechCrunch

OpenAI will spend the equivalent of Sweden's GDP on infrastructure through 2030.

NEWSJul 22, 2026

AMD commits up to $5 billion to Anthropic

Emma Roth · AI | The Verge

AMD says it's going to invest up to $5 billion in Anthropic, while helping to expand the AI company's computing power, according to an announcement on Wednesday. As part of the new partnership, Anthropic will deploy up to 2 gigawatts of AMD's Instinct MI450 AI GPUs using the chipmaker's new Helios rack-scale system, as reported […]

Tuesday, July 21, 2026

NEWSJul 21, 2026

AI and the rise of the universal entertainment app

Sarah Perez · AI News & Artificial Intelligence | TechCrunch

Over the past decade, streaming platforms competed by dominating individual formats like music, video, podcasts, or audiobooks. Now, as AI makes it easier to create, organize, and recommend content, those distinctions are fading, pushing companies like Spotify, Netflix, YouTube, and TikTok to become all-purpose entertainment destinations instead.

NEWSJul 21, 2026

Google launches a cheaper alternative to large AI security models like Mythos

Emma Roth · AI | The Verge

Google is launching Gemini 3.6 Flash alongside a new security model dedicated to quickly finding and patching security vulnerabilities. In a blog post on Tuesday, Google describes Gemini 3.5 Flash Cyber as a "cost-efficient and highly capable alternative" to larger, more expensive AI systems, such as the one offered by Anthropic's Mythos. The cybersecurity model […]

Sunday, July 19, 2026

NEWSJul 19, 2026

I hate that I don’t hate this song made with Suno

Kate Park · AI News & Artificial Intelligence | TechCrunch

I would never go so far as to say there's no place for AI in music (I'm a fan of Holly Herndon, after all). But I generally find music made with generative AI to be offensively boring, especially the outputs of Suno. So I'm having a bit of a tough time processing the fact that […]

NEWSJul 19, 2026

Nonprofit Current AI is racing to build the World Wide Web of AI, free for all

Current AI, a non-profit building AI that leaves no one culture behind, has made remarkable progress across devices, AI chat and more.

Saturday, July 18, 2026

NEWSJul 18, 2026

Kimi: Threat or menace?

Anthony Ha · AI News & Artificial Intelligence | TechCrunch

Chinese company Moonshot AI released a new version of its Kimi model this week, prompting concern about "full AI communism."

Friday, July 17, 2026

NEWSJul 17, 2026

Why the first GPU financiers are turning to inference chips in a $400 million deal

Tim Fernholz · AI News & Artificial Intelligence | TechCrunch

A $400 million chip-backed loan points to the next wave of AI infrastructure deals.

Thursday, July 16, 2026

NEWSJul 16, 2026

The AI compute gap: Enterprises are buying infrastructure faster than they can measure what it costs

AI | VentureBeat

Across 107 enterprises, AI infrastructure spending is accelerating well ahead of the ability to see or steer its economics. Most organizations run their AI on a familiar base of hyperscalers and model-provider APIs, yet the next dollar is aimed at specialized compute almost none of them use today; a majority intend to switch or add providers within the year, many within a quarter. Buying decisions turn on integration and total cost of ownership rather than headline token price — which is fortunate, because most enterprises cannot yet see their unit economics clearly: GPUs sit at half utilization or less, and fewer than half rigorously track what their compute actually costs. The result is a compute gap — heavy, fast-moving investment running ahead of the visibility needed to control it.This wave of VentureBeat Pulse Research examines enterprise AI infrastructure and compute: where organizations are in their deployment journey, what they run AI on today, how satisfied they are, what would make them switch, where they plan to evaluate their investments, and — most revealingly — how well they can measure and control the economics of the compute underneath it all.The central finding is a compute gap — the distance between how aggressively enterprises are investing in AI infrastructure and how little of its economics they can see. Only about one in five (21%) run AI in production at scale, yet spending intentions are outrunning that maturity: the single largest planned area enterprises plan to evaluate over the next year is AI-specialized clouds (45%), a layer almost none of these enterprises use today. Meanwhile the compute already in place runs cold — 83% report GPU utilization of 50% or less — and fewer than half (44%) can rigorously track what their AI compute costs. Enterprises are buying more infrastructure faster than they can account for what they already own.Enterprises are not settled on their infrastructure vendors, either: A clear majority (64%) plan to switch or add an infrastructure provider within twelve months, and 38% within the next quarter — unusually high churn intent for a category this foundational. When they choose, they choose on integration with the existing stack (41%) and total cost of ownership (35%), not on headline price: cost per million tokens is the deciding factor for just 8%. And the frontier constraint that will shape the next round of decisions — the shift from GPU compute to memory bandwidth as inference scales — is barely on the radar, with roughly one in five enterprises either unaware of it or yet to address it.MethodologyVentureBeat fielded this survey as part of its ongoing Pulse Research series, this survey focused on enterprise AI infrastructure, compute, and inference economics. Responses are filtered to organizations with more than 100 employees (n=107; the survey’s smallest size band, 1–100 employees, is excluded), drawn from a single Q2 2026 (June) wave. Because this is one wave rather than a pooled multi-month sample, the report reads cross-sectionally and does not infer month-over-month trends. Several questions were multiple-select, so those shares can sum to more than 100%.By organization size the sample concentrates in the mid-market: 101–250 employees (36%) and 251–1,000 (27%) lead, with 1,001–5,000 (22%), 5,001–10,000 (8%), and 10,001+ (7%) above them. By role it spans managers (38%), individual contributors (28%), VPs and directors (19%), and the C-suite (13%); on purchasing authority it is buyer-credible, with 45% final decision-makers and another 30% recommenders or influencers for AI solutions. Technology/Software is the largest industry at 26%, followed by Healthcare/Life Sciences (15%), Financial Services (13%), and Retail/E-commerce (12%).At 107 respondents the sample is large enough to read directionally but should be treated as a directional signal rather than a precise measurement; it is self-selected and is not a probability sample. It also skews toward the mid-market and toward earlier-stage adopters, so it is best read as the view from organizations actively building out AI infrastructure rather than from the largest hyperscale operators.Finding 1: Ambition outpaces productionOnly one in five run AI in production at scaleWe asked where organizations sit in their AI deployment journey. Most are still building toward production rather than operating at scale.The maturity curve is front-loaded. Three-quarters of enterprises (76%) are either experimenting or running only some workloads in production, and just 21% describe AI in production at scale. This matters for everything that follows: the infrastructure decisions in this report are being made largely by organizations still early in deployment, whose compute footprint — and whose costs — are about to grow. The evaluation and switching intentions in Findings 3 and 4 are the leading edge of that build-out, not the settled preferences of operators who have already found what works.Finding 2: Enterprises run on hyperscalers and model APIsThe specialized GPU clouds barely register — todayWe asked which providers and platforms enterprises currently use to run their AI. The answer is a familiar one: the incumbents.The current stack is hyperscaler-and-API. Google Cloud leads at 48%, and the general-purpose clouds (Google, Microsoft, AWS, Oracle) together with the major model APIs (Gemini, OpenAI, Anthropic) account for essentially all current deployment. The specialized “neocloud” GPU providers that dominate AI-infrastructure headlines — CoreWeave, Lambda, Crusoe, Nebius and peers — register at or near zero among these enterprises today. Only 6% run their own on-prem GPU clusters and 4% a custom open-source stack. Enterprises are, for now, running AI on the providers they already buy from — which makes the evaluation intentions in Finding 3 all the more striking.(A note on reading these shares. As described in the methodology section, this sample is self-selected and skews mid-market, and this question counted every provider a respondent uses — an average of 2.1 selections each — so the figures measure presence in the stack rather than spending or primary status. A sample built this way will show a different provider mix than a spend-weighted census of the broader market; Google's strength here, for example, is consistent with its long-standing position among smaller enterprises building on AI. Read these shares as a portrait of what this AI-active cohort runs today, and treat gaps between these figures and industry-wide market share estimates as a property of the sample rather than a contradiction of either.)Finding 3: The next dollar goes to infrastructure they don’t yet runAI-specialized clouds top the evaluations listWe asked where enterprises planned to evaluate AI infrastructure over the next 12 months. Their answers point away from the stack they run today.Here is the report’s sharpest tension. The single most-cited planned evaluation area — AI-specialized clouds, at 45% — is the very category almost none of these enterprises use today (Finding 2). Nearly a third (32%) intend to evaluate non-Nvidia accelerators, and 28% in next-generation Nvidia silicon; even decentralized compute networks (16%) and sovereign compute (11%) draw meaningful interest. Read against current usage, this is not incremental — it is the leading edge of a re-platforming. The direction-of-travel question tells the same story: every infrastructure approach is net-expanding, but specialized AI clouds carry the highest net momentum (+24), edging out even the hyperscalers (+22). Enterprises are preparing to move a meaningful share of AI compute off the general-purpose cloud.This continues a trend we saw in our April-May survey wave. Back then, usage of the AI-specialized clouds was equally marginal — CoreWeave at 3%, Lambda at 4%, Crusoe at 2% of enterprises. When we asked enterprises what change they planned in their AI infrastructure strategy over the next twelve months, the most-cited answer was moving workloads to specialized AI clouds, at 33%. Asked in April-May which emerging compute option they were most likely to evaluate AI-specialized clouds again drew the most responses. Two waves, two differently worded questions, one consistent picture: the type of cloud enterprises are most eager to assess is the type they have barely begun to use.Finding 4: A switching wave is buildingSix in 10 plan to change providers within a year — many within a quarterWe asked whether and when enterprises plan to switch or add an infrastructure provider. Very few intend to stand still.For a category as foundational as compute, this is a remarkable amount of intended movement. Only 36% have no plans to change, meaning a clear majority (64%) intend to switch or add a provider within twelve months — and 38% within the next quarter alone. Where that interest points is telling: the providers drawing the most switching consideration are again the incumbents — Microsoft Azure and Google Cloud (33% each), OpenAI (30%), and Gemini (22%) — which suggests much of the near-term movement is reshuffling among the majors and consolidating spend rather than defecting to new entrants. The neocloud interest in Finding 3 is a 12-month evaluation thesis; the switching in the next quarter is mostly incumbents trading share.(Method note: Respondents who selected both "no plans to change" and a specific switching window are counted as switchers, on the logic that naming a timeframe is the more specific answer; three respondents were reclassified under this rule.)Finding 5: Nobody buys on token priceIntegration and total cost of ownership decide — not sticker priceWe asked what matters most when enterprises select an AI infrastructure provider. Headline price finished last.Enterprises do not buy AI infrastructure on pricing, which is the place vendors compete on hardest. Integration with the existing stack (41%) and total cost of ownership (35%) dominate, while the headline metric — cost per million tokens — is the deciding factor for just 8%, dead last. The pattern is coherent: buyers are optimizing for how a provider fits and what it truly costs to operate, not for the advertised unit rate. It also foreshadows Finding 7 — enterprises say TCO matters most, yet most cannot yet measure it rigorously. The stated priority and the measured capability are out of step.Finding 6: Expensive GPUs, idle most of the time83% report GPU utilization of 50% or lessWe asked what share of their GPU capacity enterprises actually utilize. The answer is a well-known but rarely quantified inefficiency.Disclosure: Band percentages count every selection against all 107 qualified respondents; 14 respondents selected more than one band, so bands overlap. At the respondent level, 83 of the 100 GPU-operating enterprises reported utilization at or below 50%The compute already in place runs cold. Adding the bands at or below half capacity, 83% of enterprises that operate GPUs report utilization of 50% or less, and nearly half (49%) run at 25% or below. Only 12% clear the 50% mark, and a further 8% do not measure utilization at all. Idle accelerators are expensive accelerators, and this is the clearest single measure of the compute gap: enterprises are planning to buy more GPUs and specialized compute (Finding 3) while the capacity they already own sits substantially unused. The efficiency headroom in the current fleet is large — and largely unmeasured.Finding 7: Spending fast, measuring slowlyFewer than half rigorously track what their compute costsWe asked whether enterprises can quantify the cost and return of their AI infrastructure spend, and how satisfied they are with what they run. Confidence in the ledger lags the spending.Measurement trails money. Fewer than half of enterprises (44%) rigorously track the cost and return of their AI compute; the majority track only partially (39%), cannot quantify it yet (20%), or have not prioritized it (6%). That gap is consequential given Finding 5, where total cost of ownership was the second-ranked buying criterion — enterprises are choosing providers on an economic basis they mostly cannot yet measure. Satisfaction with current infrastructure is moderately positive but not enthusiastic: on a five-point scale, overall satisfaction averages 4.0, with ease of implementation (3.8) and value for money (3.9) trailing slightly — the softness landing, tellingly, on cost. Enterprises are spending quickly and accounting slowly.Finding 8: The next bottleneck few are watchingAs inference shifts from compute to memory, the field scattersFinally, we asked how enterprises would address the emerging constraint in large-scale inference — the shift from GPU compute to memory, specifically KV-cache capacity. The responses reveal a frontier that is not yet a priority.The memory frontier is real but barely governed. Asked which approach they would rely on as the binding constraint in inference shifts from compute to memory bandwidth, enterprises scatter: Dell leads at 31%, Nvidia follows at 16%, and the rest fragments across storage vendors, open-source tooling, and model-level efficiency techniques. Most telling is that roughly one in five (18%) either do not recognize the constraint or have not begun to address it. For a shift that will reshape inference cost and architecture, this is an early and unsettled market — and, consistent with the measurement gap in Finding 7, one where many enterprises simply do not yet have a view. It is the next chapter of the compute gap, arriving before most have closed the current one.The bottom line: A compute gap that faster spending will widen, not closeOrganizations with more than 100 employees are investing in AI infrastructure faster than they can measure it. Most are still early in deployment, yet their spending intentions point past their current stack — toward specialized clouds and alternative accelerators almost none of them run today — and a clear majority intend to change providers within the year. They buy on integration and total cost of ownership rather than headline price, which is rational; the difficulty is that most cannot yet see those economics clearly.The visibility gap is concrete. The GPUs enterprises already own run at half utilization or less for the overwhelming majority, and fewer than half can rigorously track what their compute costs or returns. Satisfaction is decent but unenthusiastic, softest on value for money — the dimension hardest to judge without measurement. And the next constraint, the shift from compute to memory in large-scale inference, is arriving while most enterprises are still unaware of it. At 107 respondents in a single Q2 wave this is a directional read, skewed toward the mid-market and earlier-stage adopters — but the direction is consistent: the appetite to spend is running well ahead of the instrumentation to spend well. The compute gap is not a capacity problem that more hardware will solve on its own; it is, first, a problem of seeing what the hardware already costs. The open question for later waves is whether enterprises build that visibility before the re-platforming arrives — or buy the next layer of infrastructure as blind to its economics as the last.Based on survey responses from 107 qualified enterprise respondents (100+ employees), drawn from a single Q2 2026 (June) wave. Because this is one wave rather than a pooled multi-month sample, the results read cross-sectionally rather than as a month-over-month trend, and at 107 respondents this is a directional signal rather than a precise measurement — the sample is self-selected, skews mid-market, and leans toward earlier-stage adopters rather than the largest hyperscale operators. Respondents include managers, individual contributors, VPs/directors, and the C-suite, with buyer-credible purchasing authority, across Technology/Software, Healthcare/Life Sciences, Financial Services, Retail/E-commerce, and other industries.

Wednesday, July 15, 2026

Connie Loizos · AI News & Artificial Intelligence | TechCrunch

Thinking Machines amps up its bet against one-size-fits-all AI with its first open model, Inkling

It's the company's first public proof point after a year and a half spent building AI infrastructure largely out of public view.

Jess Weatherbed · AI | The Verge

Suno snatched millions of songs from YouTube, Genius, and Deezer

Suno data obtained in a hacking incident has exposed that the AI music generator was trained by scraping millions of songs and lyrics from online audio platforms, including YouTube Music, Deezer, and Genius, 404 Media reports. Given that Suno has avoided revealing what's in its training datasets and how they were acquired, this a rare […]

Will Douglas Heaven · MIT Technology Review

Meet GPT-Red: an LLM super-hacker OpenAI built to make its models safer

OpenAI has built an LLM super-hacker called GPT-Red that it uses as a sparring partner to help its other models boost their defenses against cyberattacks. Last week the company released the latest version of its flagship LLM, GPT-5.6. OpenAI says that training it against GPT-Red made the model its most robust release yet. GPT-Red automates…

Amanda Silberling · AI News & Artificial Intelligence | TechCrunch

Hack suggests AI music generator Suno scraped YouTube for training data

The hacker used an employee's credentials to access source code, which revealed how Suno scraped decades of audio.

Tuesday, July 14, 2026

NEWSJul 14, 2026

Apple opens its new Siri AI to everyone with the iOS 27 public beta

Sarah Perez · AI News & Artificial Intelligence | TechCrunch

If you’ve been waiting to try Apple’s revamped Siri without installing a developer beta, you now can. The company on Tuesday released the iOS 27 public beta, giving iPhone owners early access to its AI-powered assistant and other new features before the software’s official launch this fall.

NEWSJul 14, 2026

Spotify is now an AI chatbot, too

Jess Weatherbed · AI | The Verge

Spotify is experimenting with a new AI feature that allows Premium subscribers to play and explore music, audiobooks, and podcasts by having conversations with a chatbot. The "Talk to Spotify" feature appears across the Home and Now Playing view on Spotify's mobile app. You can interact with the chatbot by typing your request in the […]

Monday, July 13, 2026

NEWSJul 13, 2026

Building a Foundation Stack for General-Purpose Robots

X Square Robot · IEEE Spectrum

This article is brought to you by X Square Robot.Large language models gave artificial intelligence a working recipe. Pretrain a large model on broad data, and general capability follows. Robotics has no such recipe. Robotics systems have long been assembled from separate perception, planning, and control parts that rarely add up to intelligence a robot can carry from one task to another, or one machine to another. The central problem in embodied AI is to find the equivalent recipe, and the field does not yet agree on what it is.X Square Robot, a Chinese embodied-AI company, has made an unusually explicit bet. It argues that the recipe is an integrated stack, spanning the data a robot learns from, a world model for predicting changes in the physical world, and an action model that brings together perception, planning, reasoning, and decision-making to generate executable robot behavior. The company also believes that the stack should be built and released in the open. X Square Robot shares its vision of bringing robots into real homes.X Square RobotX Square Robot’s embodied AI stackWhat holds the stack together is a small set of principles rather than a single overarching model.The first is that the basic unit of robot data is an interaction, not a trajectory; a demonstration is successful only if it changes the world as intended, not simply because the joints moved. The second is that pretraining should yield usable capability, not just an initialization for later fine-tuning. The third is that behavior should be modeled around physical events rather than fixed slices of time. These principles make the layers interdependent, since the same robot-free data that trains the action model is also structured to feed the world model. It is worth being precise, though. The company describes the world model and the action model as complementary but independent model families that share a code base. Both sit within its broader World Unified Model, which it has presented as an architecture for training vision, language, action, and physical prediction together.Robot learning data: Engineering for quality and cost, not scaleFor the X Square Robot team, one of the biggest constraints on general-purpose robots is the cost and quality of interaction data, not the number of parameters. To address that, the company built its Universal Manipulation Interface (UMI) data collection system, QUANXTA Zero Series. It works by collecting demonstrations from people wearing a rig with dual grippers rather than teleoperating a robot. This approach is not itself new, and builds on established methods for robot-free data capture. What sets it apart are two engineering choices. X Square Robot emphasizes data quality control, recording trajectories and replaying them on a real robot, with only those that actually complete the task counted as valid.X Square RobotThe first is quality control, and it is the most distinctive part. Rather than accepting recorded trajectories as they are, the system runs a closed inspection loop, and its notable step is physical playback. A sample of trajectories is replayed on the real robot, and only those that actually complete the task count as valid. That makes the validity rate a measured quantity rather than an assumption. For example, a gripper that closes a fraction of a second too early still looks like a grasp in the data, yet it has pushed the object away, so it shouldn’t be classified as valid. A smaller clean dataset can be worth more than a larger noisy one.The second choice is how lower-cost human data and scarce robot data are combined. The company pretrains on a large volume of robot-free demonstrations to build general representations, then adds a small amount of real-robot data as an anchor to the specific machine’s dynamics. It reports that this reaches performance comparable to an all-robot dataset at roughly a 20-fold lower cost of collection, driven mainly by how much cheaper the wearable rig is than a teleoperation setup. The resulting dataset is deliberately model-agnostic, formatted to feed both action models and world models. The caveat is that the strongest results are measured on the company’s own robots and data-collection pipelines. Broader independent testing will help confirm and extend these promising results across a wider range of settings.A world model organized around eventsIn developing its world model, called WALL-WM, X Square Robot took a differentiated approach. Most action models predict a fixed-length chunk of motion from the current image and instruction. That is convenient, but it segments behavior into fixed-duration windows, so the boundaries fall where elapsed time dictates rather than where one action ends and the next begins. WALL-WM instead treats an action-grounded semantic event as its unit: a coherent piece of behavior such as reaching, grasping, or placing, something that can be named in language, seen in video, and executed as motion. X Square Robot’s world model, called WALL-WM, treats an action-grounded semantic event as its unit: a coherent piece of behavior such as reaching, grasping, or placing, something that can be named in language, seen in video, and executed as motion.X Square RobotWALL-WM’s design reflects a specific concern about not discarding what large video models already know. To achieve that, a text-to-video model is coupled to a freshly initialized action network that reads from the video features without overwriting them, which preserves the visual prior. From that one process, it offers two modes. An event mode runs in variable-length segments and suits reasoning over long horizons, while a fixed-length mode produces the steady, real-time output a controller needs. That places WALL-WM between mainstream chunk-based action models and pure video world models, keeping the predictive character of a world model while still yielding executable control.In a series of experiments, the company relied on a generalization test that is more specific than most. A model trained on a limited dataset was evaluated on long-horizon tasks in unseen settings and, on the company’s real-robot benchmark, reportedly outscored baselines that had been fine-tuned on related data. That is a meaningful result if it holds. For now, it is measured on the company’s own benchmark. With the code now being released, the broader community will have the opportunity to test, reproduce, and build on them across more settings.A policy that runs before fine-tuning, and action tokens with meaningThe action layer carries two connected ideas. The first is a requirement the company sets for itself with Wall-OSS-0.5, its vision-language-action model: The pretrained model should run on a real robot before any task-specific fine-tuning. The interest is less in the scores than in the design behind them. The model trains three objectives together, namely discrete action tokens, language grounding, and continuous action generation. And it keeps gradients flowing through all of them rather than freezing parts of the network as some rival designs do. It’s also a more strict method, since it reports untuned behavior such as approaching, grasping, and recovering, including on a deformable task held out of training. As part of X Square Robot’s Wall-OSS-0.5 vision-language-action model design, the pretrained model should run on a real robot before any task-specific fine-tuning. X Square RobotThe second idea is the action interface itself, called X-Tokenizer. Most systems that turn continuous motion into discrete tokens produce codes that the language model cannot interpret. X-Tokenizer reframes tokenization as learning a semantic interface, so that the top-level code stands for the intent of a motion while lower-level codes carry finer detail, all aligned with the language model’s own features. A useful consequence is stability. Adding noise to an action barely moves the intent code, which is what lets one tokenizer to be reused across robots without re-tuning. The tokenizer inside the production action model is a related variant of this approach. Together, the two ideas give the action layer something rather powerful: capability that transfers.The future of embodied AI stacksX Square Robot is betting that its unique approach combining three layers, each specialized in solving a key part of the problem, will stand out from other embodied AI stacks. The physical-playback step that grounds data quality is uncommon and sensible. The reframing of world modeling around events, with one backbone serving both reasoning and control, is a genuinely distinct approach. And the pairing of a deployable pretraining standard with a tokenizer designed as a semantic interface gives the action layer unusual coherence. X Square Robot’s valuation has climbed above 20 billion yuan (about US $2.9 billion), suggesting that investors increasingly view data infrastructure, foundation models, and scalable training systems as long-term differentiators in embodied AI.The next phase will bring broader validation. Much of the current evidence comes from X Square’s own robots and benchmarks. With the world model code now being made public, and as the community begins to test, reproduce, and build on the work, the reported capabilities will be tested across more robots, tasks, and settings.X Square Robot’s recent funding rounds reflect similar confidence. The company’s valuation has climbed above 20 billion yuan (about US $2.9 billion), suggesting that investors increasingly view data infrastructure, foundation models, and scalable training systems as long-term differentiators in embodied AI.What’s next for X Square RobotTo learn more about its future plans, the following Q&A with the X Square Robot team further explores the company’s technology, strategy, and vision.What made now the right moment, technically, to commit to this stack? What recently became possible that wasn’t possible a couple of years ago?It is not one breakthrough but several trends maturing together. Foundation models gave us a shared representation across vision, language, and action, so we can model what a robot sees, what it is asked to do, and how its actions change the world in one framework, rather than as separate perception, planning, and control modules. Compute and infrastructure are finally sufficient for large-scale pretraining over long-horizon, multi-embodiment data. Just as importantly, we realized that data, not model size, is the real bottleneck for general robots—what is scarce is diverse, high-quality, reproducible interaction data. And world modeling has become practical. The useful question is no longer how to predict a few seconds of video, but how to understand the ways actions change objects, contacts, and task states. Two years ago these ingredients existed separately. Today they are mature enough to work as one system.“We realized that data, not model size, is the real bottleneck for general robots—what is scarce is diverse, high-quality, reproducible interaction data. And world modeling has become practical.”Your data system captures demonstrations with a wearable VR rig and custom grippers rather than teleoperating robots. What was wrong with standard teleoperation?Teleoperation is built around controlling the robot. It forces the operator to work within the machine’s kinematics, latency, and viewpoint, and the resulting demonstrations are slower, stiffer, and less diverse. We built our system around capturing human skill instead. Manipulation is really about contact, timing, finger coordination, and recovery, not just the path the hand takes, and a wearable rig records those before the behavior is compressed onto one particular robot. It also breaks teleoperation’s expensive scaling law, in which every demonstration needs a robot. People can generate rich data independently of any robot, and the crucial property is that those demonstrations can still be replayed and executed on a physical robot through the model. Mobility is convenient, but that replay is the real point, because it is what lets the same data be reused across different platforms. In X Square Robot’s approach, demonstrations can be replayed and executed on a physical robot through the AI model, allowing the same data to be reused across different platforms.X Square RobotX Square Robot reports that its pipeline has roughly an 85 percent data-validity rate. Why is quality control such an underrated bottleneck?Because errors in robot data are far more expensive than in language data. A small timing or contact error can change what a demonstration means. If a gripper closes a fraction of a second too early, the motion still looks like a grasp, but physically it has pushed the object away. A dataset that mixes failures and accidental successes teaches ambiguity, not skill, because the real unit is the interaction, not the trajectory. So we run automated inspection, kinematic checks, and physical replay, where we play a sample of trajectories back on the real robot and count only the ones that actually complete the task. Data quality sets the ceiling on how good a policy can be. In our experience a smaller, cleaner dataset often beats a much larger, noisier one, which is why we treat quality control as part of the model, not a preprocessing afterthought.The model runs in both “event mode” and “chunk mode.” When does each matter?Both matter, for different reasons. The physical world changes through events—when contact occurs, a grasp forms, or an object slips—not in fixed-frame windows. Event mode concentrates the model’s attention on those moments, and it matters most for long-horizon tasks, like clearing a table, where progress is a sequence of semantic events rather than a smooth stream. It runs in variable-length segments that follow the task rather than a clock. Chunk mode matters for deployment. Real controllers need a stable, real-time interface, and fixed-length chunks integrate cleanly with existing control systems. We organize learning around events in the first place because a fixed window can split one motion in half or merge two together, which turns training into short-horizon pattern matching and weakens the model on long tasks. So the world model’s job is to connect event-level understanding, which is where the reasoning happens, with a fixed-length output a real robot can actually run.Why make “deployable before fine-tuning” the criterion?Pretraining should produce capability, not just a good starting point. If a model is only useful after heavy fine-tuning, then most of the intelligence still lives in the downstream supervision, not in the foundation model. Deployable before fine-tuning is a more honest test of what pretraining actually learned. A well-pretrained robot should already know how to approach, grasp, move, avoid obstacles, and correct itself. Fine-tuning should adapt it to a specific task or robot, not create the ability from nothing. It is also a practical requirement. A robot in a home or a workplace shouldn’t need a brand-new dataset and a new policy every time the task changes, so a foundation model that already carries general skill, and some ability to recover, is the minimum bar for something genuinely useful in the real world.What is the most challenging part of cross-embodiment learning?Robots differ in control frequency, delay, compliance, sensing precision, and contact dynamics, so the same instruction can require different action decompositions and recovery strategies, and a behavior that works on one arm cannot simply be copied to another. Cross-embodiment learning needs an intermediate abstraction, lower than language but higher than joint angles: how you approach an object, how you make contact, how you apply force, and how you recover from a mistake. When we say cross-embodiment, the main capability we mean is multi-embodiment generalization: transferring across robots, training on many embodiments at once, and adapting to different kinematics. Human-to-robot transfer and other techniques are specific approaches to that goal.“A robot in a home or workplace shouldn’t need a new dataset and policy every time the task changes. A useful foundation model should already carry general skills and the ability to recover.”What would you most like to see other researchers attempt to reproduce or stress-test?Three things, above all. Whether event-level representations really generalize beyond our own datasets, across more tasks, scenes, objects, embodiments, and failure conditions. Whether pretraining stays effective on robots the model never saw during training, or whether its capability is still too tightly coupled to what it has already seen. And whether real-robot evaluation can become a shared language for the field, so that we compare not just success rates but the reasons systems fail, where an instruction was misread, where perception broke down, or where recovery fell short. Robotics has been driven too often by impressive demonstrations, and real progress comes from results that are reproducible and diagnosable.What capability is still missing before robots become dependable in homes?Benchmarks measure competence, like whether a model can finish a task. Homes demand reliability, safe and consistent operation over time in a place that changes every day, with objects moving, instructions that are vague, and people interrupting. The missing piece is not a higher one-time success rate: it is robust recovery. A dependable home robot has to know when it is uncertain, when to slow down, when to ask for help, and how to bring the world back to a safe state after it drops something or misunderstands a request. In a real home, failure recovery matters more than raw success, because the home does not reset itself. Homes also demand careful personalization, learning a household’s routines and preferences over time, with safety and trust as first principles. That combination, not any single skill, separates a capable demonstration from a robot people can live with. X Square Robot’s approach is that, in a real home, failure recovery matters more than raw success, because the home does not reset itself and it demands careful personalization, with safety and trust as first principles. X Square RobotHow do the open-source components fit into X Square Robot’s World Unified Model direction?We see these releases as layers of the World Unified Model direction rather than isolated projects. Wall-OSS-0.5, the action model, asks whether an open vision-language-action model can gain directly measurable capability from large-scale pretraining, so it is the capability layer. WALL-WM, the world model, asks how a robot should understand change in the world, shifting from fixed windows to event-level modeling, so it is the representation layer. The data system supplies the interaction data that both of them learn from. Together they form a loop in which models produce capability, world models organize understanding, and the open-source community drives reproduction and improvement. World Unified Model is the broader architecture those layers support, bringing vision, language, action, and physical prediction together. We are releasing these pieces openly because embodied intelligence cannot be solved by one organization; it needs many embodiments, many real tasks, and broad feedback, and the long-term goal is a stack that keeps learning and ultimately moves robots from laboratory demonstrations toward reliable everyday use.

Friday, July 10, 2026

NEWSJul 10, 2026

Open source AI matters more than ever, according to Hugging Face’s Clem Delangue

Rebecca Bellan, Theresa Loconsolo · AI News & Artificial Intelligence | TechCrunch

Open source AI is booming, according to Hugging Face CEO Clem Delangue. The company has grown into something like a GitHub for AI in recent years, where AI builders can share and download open models and datasets, now used by roughly half the Fortune 500. Delangue has seen the same story play out again and again: companies start […]

NEWSJul 10, 2026

SK Hynix raises $26.5B in the biggest foreign IPO in US history, is urged to build new US fabs

Kate Park · AI News & Artificial Intelligence | TechCrunch

The AI chip boom just produced its biggest Wall Street moment yet. Now SK Hynix and Samsung are being asked to build U.S. factories.

Thursday, July 9, 2026

NEWSJul 9, 2026

Paris-based AI voice startup Gradium raises $100M seed, backed by Nvidia

Julie Bort · AI News & Artificial Intelligence | TechCrunch

The company is using the cash to open an office in the Bay Area and compete for talent there, "strengthening its position at the heart of the world's leading AI ecosystem."

NEWSJul 9, 2026

Meta’s new AI chips will begin production in September

Ram Iyer · AI News & Artificial Intelligence | TechCrunch

The company is taking a modular approach to designing these chips, anticipating that their needs will change as AI evolves rapidly by the time the chips are in production.

NEWSJul 9, 2026

Large Tabular Models Excel Where LLMs Fail

Benjamin Skuse · IEEE Spectrum

The large language models (LLMs) that form the basis of generative AI chatbots such as ChatGPT, Claude, and Gemini can generate uncannily human-like text and images. But these models still struggle with a skill that, ironically, looks at face value to be right in their wheelhouse: analyzing structured data. A new type of generative AI is set to change this situation.Although you can get your favorite chatbot to solve intractable math problems, review dense legal documents, compose a catchy pop song, or put together some slick PowerPoint slides, give it anything more than a small table and it doesn’t have a clue what to do.For most companies and organizations, the most important data sits in spreadsheets. Whether it’s a bank’s transaction logs, a marketing agency’s website metrics, clinical trial participants’ vital signs, or the vast amount of proton collision information produced at atom smashers like the Large Hadron Collider, structured, row-and-column data runs the world, and LLMs can’t deal with it.AI startup Fundamental is pioneering a new type of AI foundation model, known as a large tabular model (LTM), to fill the gap. Fundamental came out of stealth mode on 5 February 2026 with US $275 million in funding and a model called NEXUS, purpose-built for tabular data. Now, the model is being adopted by companies such as Amazon Web Services, while others race to build their own LTMs. Why LLMs struggle with spreadsheetsPart of why structured data has garnered less attention is a very human bias, argues Boris van Breugel, a senior AI researcher based in Amsterdam. “People like to see images, videos, and ChatGPT responses,” he says. “But tabular data really lags behind because it’s not fun to look at numbers.” Different tabular datasets are also difficult to compare, explains van Breugel, who co-wrote a prescient position paper on this topic in 2024. Whereas most language has similar semantics, making LLMs well-suited to being trained on vast amounts of text data, van Breugel argues that it is much harder to train a single tabular model on tables with very different variables. Additionally, language is sequential by nature (as are music, images, and video). Changing the order of words in a sentence may change or completely destroy its meaning. But the structured data you find in spreadsheets isn’t sequential. You can swap the order of columns or play around with rows, but the underlying factual meaning of the data remains the same.This independence from linear order is incompatible with an LLM’s fundamental purpose of predicting the next value in a linear sequence. “With LLMs, even slightly changing the input, you get a different output,” says Jeremy Fraenkel, CEO of Fundamental. “That’s fine, and actually often desirable for LLMs, but when you’re making a prediction of whether a transaction is fraudulent or not, you want to make sure that the prediction is the same, or deterministic, no matter what.”Developing Fundamental’s LTMCurrent tabular data solutions are limited to machine learning algorithms, such as XGBoost, that have been around for more than 15 years and are used by organizations globally. These algorithms—called gradient-boosted decision trees—have to be trained and optimized by data scientists over the course of months for each and every use case. In contrast, NEXUS and other emerging LTMs are foundational, leveraging learning amassed from pre-training on diverse databases so that they can be applied across a range of different predictive tasks with minimal bespoke feature engineering or task-specific model building.And unlike LLMs, which primarily model sequences of tokens, LTMs model the structure of tabular data directly. They jointly learn from each entry’s numerical value, what it represents, and how it relates to other entries. For example, imagine an entry in a grocery stock inventory table for bananas: The LTM can take in not just the magnitude—say, 500—but the fact that the entry represents the current banana stock quantity, its category (produce), and the statistical properties that link the entry with the rest of the column. This contextual understanding enables more accurate reasoning and prediction over structured data.According to Fraenkel, one of Fundamental’s biggest challenges in developing NEXUS was obtaining the right training data. Unlike natural language, which is abundant and broadly uniform in structure, tabular data is relatively hard to find—much of the data is sensitive or proprietary—and diverse. There are very few similarities between, for instance, a biology dataset and a financial one. That combination of factors meant Fundamental needed to invest in building a huge training set.“We pre-trained NEXUS on billions of tables using a combination of proprietary datasets acquired through partnerships and licensing, high-quality public and open-source datasets, and data augmentation techniques that expanded the diversity and coverage of our training corpus,” Fraenkel says, though he is keen to point out that NEXUS is not trained on customer data. In fact, it is a confidential computing platform, which means that Fundamental physically cannot access customer data, let alone train on it.This feature was most likely a key consideration when in June, Amazon Web Services (AWS) embedded NEXUS in Amazon SageMaker, widely considered the default operating system for secure machine learning. This brings NEXUS to many customer’s often sensitive data—a contrasting approach to LLMs, where the data has to be imported to the model.“With Amazon, we have a first-party partnership, which means that our model exists as if it’s a native AWS solution,” says Fraenkel. “And over time, the goal is to expand these types of relationships to allow [ens-users] to really access their data wherever they do their predictions.”The future of data analysisThough Fundamental has taken the lead, at least in enterprise applications, the company is not alone in pursuing foundational LTMs. In March, Feedzai, which provides fraud and financial crime prevention services, and credit card company Mastercard separately launched similar proprietary technologies focused on finance. Then, in late June, Google launched its own foundational competitor TabFM, trained entirely on hundreds of millions of synthetic datasets. And machine learning researchers are not far behind either. FlexTab, TabICL, and iLTM are just three of a raft of LTMs developed by the research community in the past year, all in the pursuit of bringing the success of LLMs to the tabular domain.For all involved, the direction of travel is clear. “I would be very surprised if most data processing and analysis is not done through an automated system in the future, whether that’s an LLM, an LTM, or some combination,” says van Breugel. “Most people don’t necessarily like to do data analysis, and these systems will be able to do it a lot better.”Fraenkel agrees. “I see the relationship between LLMs and LTMs as being a bit like the human brain: The left side is good at reasoning and understanding and summarizing text, and the right side is really good at understanding numbers and statistics and patterns,” he says. “But it’s when you combine both of those that you really get something much more powerful.”

Wednesday, July 8, 2026

NEWSJul 8, 2026

AI Models Overthink Problems—and It’s a Security Risk

Edd Gent · IEEE Spectrum

Large language models (LLMs) that can think through problems step-by-step have significantly increased the scope of tasks that AI can tackle. But new research suggests these reasoning capabilities also introduce a critical vulnerability that could allow attackers to slow these systems to a crawl.While earlier generations of LLMs would immediately produce a response to a user’s request, today’s most advanced models generate an internal monologue where they break down the problem into steps and reason about the best way to tackle it before providing an answer. This has allowed AI to tackle increasingly complex problems, particularly in areas like coding and math.However, previous research has shown that these models are susceptible to sometimes producing excessively long streams of reasoning that do little to boost performance, a phenomenon known as “overthinking.” In research presented this week at the International Conference on Machine Learning 2026, in Seoul, researchers from Zhejiang University and e-commerce giant Alibaba in China demonstrate that they can deliberately induce overthinking by subjecting models to logically inconsistent prompts. The result is a form of denial-of-service attack on commercial AI models.Evolutionary Prompt Attack on LLMsThe team has developed an evolutionary algorithm that corrupts the logical structure of prompts, causing models to spiral into overthinking as they attempt to reason through fundamentally unsolvable problems. Generating longer responses costs more and increases the load on a model provider’s servers, so if done at scale, the researchers say, this could significantly degrade the experience of legitimate users. The attack was effective against reasoning models from leading AI companies including DeepSeek-R1, Alibaba’s Qwen3-Thinking, OpenAI’s GPT-o3, and Google’s Gemini 2.5 Flash, and resulted in outputs up to 26 times as long as standard responses on a standard math benchmark.“Across multiple datasets and reasoning models, our method substantially amplifies the output length,” Wei Cao, a master’s student at Zhejiang University, wrote in an email to IEEE Spectrum. “Our results suggest that overthinking is not an isolated phenomenon specific to individual models, but rather a shared vulnerability among modern reasoning models.”The team’s approach builds on previous research from another group of researchers that showed reasoning models tend to overthink when faced with a question in which a key premise has been removed—such as asking how far someone who walks 10 miles a day covers in total without specifying how many days they walked for. Rather than identifying that the problem is unsolvable, models often engage in extended but ultimately fruitless reasoning loops in an attempt to answer the question.Taking the idea a step further, the authors took 940 problems from three math benchmark datasets and used an LLM to break down their logical structure into a set of premises and a final question. The genetic algorithm then jumbled these up using a variety of “mutations,” including swapping premises between problems, adding extra premises to problems, deleting existing premises from problems, and swapping the final questions between two sets of premises.After each round of mutations, the problems are scored on how many words they cause a target model to output and also whether they increase the frequency of specific linguistic markers of overthinking—words like “but,” “wait,” “maybe,” or “alternatively.” The problems that scored highest on both measures are retained, and the remaining ones are jumbled up again, and this process is repeated for five generations. Crucially, the approach doesn’t require access to the internals of a model and can generate malicious prompts by simply querying the target, which makes it possible to attack closed-source commercial services, says Cao.Overthinking Vulnerability in AI ModelsThe researchers found that the approach consistently led to outputs several times longer than those generated by the unmodified questions for the reasoning models they tested it on. The biggest jump came from DeepSeek-R1 on the MATH dataset, which is made up of problems from high school math competitions, where the maximum output was 26.1 times as long as the longest response the model provided to unaltered questions. While the main thrust of the research was focused on math problems, the authors also tested it on coding, scientific reasoning, and dialogue challenges, and observed significant jumps in output length in all three.One challenge for the approach is that developing the malicious prompts requires repeated queries to expensive reasoning models, which Cao admitted could limit its cost-effectiveness. However, the researchers also demonstrated that when they used a smaller, cheaper model to generate the malicious prompts, they were still able to induce the target models to produce outputs several times longer than normal. This ability to transfer malicious prompts between models significantly increases the attack’s feasibility, Cao wrote.However, he pointed out that the goal of the research is not to develop a practical DoS attack on reasoning models. Factors like the providers’ pricing model, rate limiting policies, context window size, and existing defenses could all impact how effective the approach is. The intention is instead to highlight these models’ vulnerability to logically inconsistent prompts so that providers can attempt to mitigate the problem.“Our objective is not to demonstrate that large-scale attacks can be launched at negligible cost, but rather to establish that this attack surface exists,” he wrote. “Our results indicate that the vulnerability represents a realistic security concern.”

Saturday, July 4, 2026

NEWSJul 4, 2026

What is Mistral AI? Everything to know about the OpenAI competitor

Anna Heim · AI News & Artificial Intelligence | TechCrunch

Mistral AI, which offers some open source AI models, has raised significant funding since its creation in 2023, with the ambition to “put frontier AI in the hands of everyone.”

Friday, July 3, 2026

NEWSJul 3, 2026

Anthropic wants to develop its own drugs

Robert Hart · AI | The Verge

At the event "The Briefing: AI for Science" earlier this week, Anthropic announced Claude Science, a new "AI workbench for scientists" that pulls fragmented tools and datasets into one environment, and generates figures and visuals. Anthropic, already dominating the industry with its popular coding tools and powerful AI models, framed the launch around what it […]

Thursday, July 2, 2026

NEWSJul 2, 2026

Mark Zuckerberg tells staff that AI agents haven’t progressed as quickly as he’d hoped

Lucas Ropek · AI News & Artificial Intelligence | TechCrunch

At an internal meeting, the Meta CEO reportedly said that AI development efforts were not moving as quickly as anticipated.

NEWSJul 2, 2026

Anthropic is discussing a new custom chip with Samsung

Lucas Ropek · AI News & Artificial Intelligence | TechCrunch

The news comes about a week after OpenAI announced its own custom AI chip in a partnership with Broadcom.

Wednesday, July 1, 2026

Marina Temkin · AI News & Artificial Intelligence | TechCrunch

Ashton Kutcher leaving Sound Ventures to launch new VC firm with Morgan Beller

Sound built its reputation on concentrated, high-conviction bets in category-leading AI labs, while Kutcher's new fund appears to be chasing the layer underneath those companies — the infrastructure and energy that power them.

Sarah Perez · AI News & Artificial Intelligence | TechCrunch

Cloudflare’s new policy pushes AI companies to pay for publishers’ content

Cloudflare is giving AI companies until September 15 to separate web crawlers used for search from those used for AI training and agents, or risk being blocked by default on many publisher sites.

Sarah Perez · AI News & Artificial Intelligence | TechCrunch

Gemini Spark, Google’s agentic assistant, is now available on Mac

Google's 24/7 agentic assistant, Gemini Spark, comes to Mac alongside other improvements, like real-time tracking and support for more apps.

Rebecca Bellan · AI News & Artificial Intelligence | TechCrunch

Meta, like SpaceX, looks to turn excess AI compute into cash

Meta is developing plans for a cloud infrastructure business, selling access to AI compute power and models. The move would pit it against the big cloud providers like Amazon Web Services, Google Cloud, and Microsoft Azure.

Thomas Macaulay · MIT Technology Review

The Download: Anthropic launches Claude Science, and California’s carbon manure math

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. Claude Science is Anthropic’s newest flagship product At an event for pharmaceutical executives, biotech founders, and researchers yesterday, Anthropic announced Claude Science, a major new product intended to support scientific research…

Hayden Field · AI | The Verge

Anthropic’s long-sidelined Fable 5 is greenlit to return

After weeks of negotiating with the Trump administration, Anthropic is finally going to be able to bring Claude Fable 5 back online. In a post on X, Anthropic said it plans to begin restoring access Wednesday to users globally on Claude platforms, and that the company would re-enable access on AWS, Google Cloud, and Microsoft […]

Tuesday, June 30, 2026

Grace Huckins · MIT Technology Review

Claude Science is Anthropic’s newest flagship product

At an event for pharmaceutical executives, biotech founders, and researchers on Tuesday, Anthropic announced Claude Science, a major new product intended to support scientific research in the same way that Claude Code supports software engineering. Like Claude Code, Claude Science can autonomously carry out meaningful work when given concise, high-level instructions, and it has access…

Rebecca Bellan · AI News & Artificial Intelligence | TechCrunch

Anthropic launches Claude Sonnet 5 as a cheaper way to run agents

Anthropic’s Claude Sonnet 5 brings stronger agentic capabilities, lower pricing, and improved safety, positioning the model as a cheaper alternative to Opus, GPT-5.5, and Gemini Pro.

Rebecca Bellan · AI News & Artificial Intelligence | TechCrunch

Anthropic’s Claude Science bets on workflow, not a new model, to win over scientists

Anthropic's Claude Science is a workbench that gives scientists one environment to do computational research, saving them from the need to bounce between databases, pipelines, and tools.

Russell Brandom · AI News & Artificial Intelligence | TechCrunch

Amazon launches new $1 billion FDE org, following OpenAI and Anthropic

Engineers on the new team will embed within companies to deploy purpose-built agents, focusing on fast deployments and customer self-sufficiency.

Monday, June 29, 2026

NEWSJun 29, 2026

Tidal won’t pay royalties on AI-generated music, but isn’t banning it outright

Andrew Liszewski · AI | The Verge

Tidal shared its new policies regarding AI-generated music today and how the platform plans to "protect artists" and "inform listeners." Instead of banning it outright, starting on July 15th Tidal will label tracks it has identified as being 100 percent AI-generated with an icon. But starting today those tracks will no longer be monetizable. "Tidal's […]

Sunday, June 28, 2026

NEWSJun 28, 2026

China’s Z.ai claims it can match Mythos on cybersecurity

Kate Park · AI News & Artificial Intelligence | TechCrunch

China's Zhipu AI (Z.ai) released its open-weight GLM-5.2, and some researchers have claimed that it matches Mythos in certain bug-finding and cybersecurity scenarios. While GLM lags behind models from Anthropic and OpenAI in other, more general tasks, it seems that China has dramatically reduced the gap in the capabilities between its models and those of […]

Saturday, June 27, 2026

NEWSJun 27, 2026

Asian AI startups launch Mythos-like models as Anthropic’s export ban drags on

New models are launching in Asia that promise Mythos-like capabilities without fear of an export ban. U.S. AI labs may never recover this enormous market.

NEWSJun 27, 2026

Trump Admin releases Anthropic Mythos to be used by more than 100 US companies, agencies

Julie Bort · AI News & Artificial Intelligence | TechCrunch

Over 100 companies and government agencies are reportedly authorized to use Mythos 5, including their non-American employees.

Friday, June 26, 2026

NEWSJun 26, 2026

OpenAI limits GPT-5.6 rollout after government request, says restrictions shouldn’t be the norm

Rebecca Bellan · AI News & Artificial Intelligence | TechCrunch

“We don’t believe this kind of government access process should become the long-term default,” says OpenAI. “It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.”

NEWSJun 26, 2026

Why everyone from OpenAI to SpaceX is building their own chips (and turning up the heat on Nvidia)

Theresa Loconsolo · AI News & Artificial Intelligence | TechCrunch

Nvidia has dominated the AI chip market for years, but the era of total dependence might be ending.   OpenAI just shared its plans to spice things up with Jalapeño, its custom inference chip built with Broadcom, joining Google, Apple, and SpaceX in a growing list of companies building their way out of single-supplier risk. The goal is less of a […]

NEWSJun 26, 2026

OpenAI unveils GPT-5.6 amid US AI regulatory drama

Hayden Field · AI | The Verge

Less than 24 hours after news broke that OpenAI would stagger its next model release at the request of the Trump administration, that model, GPT-5.6, is here. On Friday, the company unveiled the limited preview of its new GPT 5.6 model suite: Sol, the flagship; Terra, a medium-tier model for "high-volume work"; and Luna, a […]

Thursday, June 25, 2026

NEWSJun 25, 2026

The White House is asking OpenAI to slow roll the release of its new model over safety concerns

Lucas Ropek · AI News & Artificial Intelligence | TechCrunch

OpenAI reportedly plans to share its newest model, GPT 5.6, with a select group of partners instead of with the broader public. The reason: the Trump administration told it to.

NEWSJun 25, 2026

Patronus AI lands $50M to build ‘digital worlds’ that stress-test AI agents

Marina Temkin · AI News & Artificial Intelligence | TechCrunch

Agent-testing startup Patronus AI, founded by former Meta AI researchers, is experiencing nearly insatiable demand, its investor says.

NEWSJun 25, 2026

Amazon ups India bet with fresh $13B AI infrastructure investment

Jagmeet Singh · AI News & Artificial Intelligence | TechCrunch

Amazon’s latest India investment comes as global tech companies race to expand AI infrastructure in the country.

Wednesday, June 24, 2026

Amanda Silberling, Lucas Ropek · AI News & Artificial Intelligence | TechCrunch

AI researchers continue to leave Google for its rivals

Top AI researchers Jonas Adler and Alexander Pritzel are leaving Google for Anthropic, following departures from top scientists Noam Shazeer and John Jumper.

Emma Roth · AI | The Verge

OpenAI reveals its first AI processor: Jalapeño

OpenAI has just revealed a new "intelligence processor" chip for AI servers made in partnership with Broadcom. The chip, called Jalapeño, is designed to power current and future large language models, according to an announcement on Wednesday. Jalapeño is an ASIC (Application-Specific Integrated Circuit), meaning it's designed for a specific purpose: AI inference. With AI […]

Kaushik Sengupta · IEEE Spectrum

AI Is Designing Radio Chips That Humans Couldn’t Even Imagine

SummaryRFIC design is a complex “dark art” that limits progress in wireless technologies like 5G, autonomous vehicles, and satellite communications.Princeton researchers use reinforcement learning and inverse design to rapidly create RFICs from scratch.Diffusion models rapidly generate novel or human-interpretable RF layouts, achieving record performance and drastically reducing design time.Future progress needs large, shared chip design datasets and open ecosystems so AI can learn universal electromagnetic and circuit behaviors.Take a moment and try to imagine your life without the wireless advances of the past three decades.Have you lost your luggage? What a shame AirTags have not been invented. The airline representative has promised to call with updates, so settle in for a long wait by the kitchen telephone, because there are no affordable cellphones. You’ll be stuck listening to whatever is on the radio while you wait, because there are no streaming services. That’s not even to speak of all the movie plots that would have been ruined.This is just a tiny sliver of how wireless technology makes itself felt in your day-to-day existence. The effects it has had on supply chains, infrastructure, and how the economy runs have been world-altering.None of it would be possible without the radio-frequency integrated circuits that allow all our devices to unobtrusively send and receive information.Now imagine what the further evolution of this technology will bring: Wide-spread autonomous vehicles, quantum communications, 6G mobile service and satellite communications. Continued momentum will depend on newer and more advanced versions of today’s RF chips.But there’s the rub. Whereas the design of most of the world’s computing chips has been standardized into its own science, RF design has remained stubbornly in the realm of art. A dark art, even, that is mastered only through years of experience. As any sorcerer will tell you, the dark arts keep their own schedule. And that schedule is impeding progress not just in RF chip design but in every other technology that depends on it.About seven years ago, in the wake of AlphaGo’s victory over world Go champion Lee Sedol, my students at Princeton and I began to wonder: Could AI be taught this art as well? Recent successes suggest that, to a large extent, it can. Over the last few years, our group and other leaders in the field have started to develop machine-learning-driven algorithmic methods for designing RFICs. Some of the resulting chips look more like modern art than circuit layouts. Yet in many cases, the physical prototypes bested state-of-the art circuits in terms of performance. The real achievement, however, is that it took the AI orders of magnitude less time to conceive a working design than it would a human designer.This is not about one or two RF chips. AI-enabled design could be the future of all RF design, and maybe much more.The Dark Art of RFIC DesignSo why do these chips all have to be crafted by hand? Why aren’t RFICs designed with an algorithmic synthesis process, much as CPUs and GPUs are?The design of RFICs is an exercise in engineering across multiple physical domains. Maxwell’s equations, operating across different spatial and temporal scales, govern how electromagnetic fields interact with active and passive devices that must be carefully codesigned for the chip to function. Alongside these are the laws of thermodynamics, which determine how heat is generated and removed during operation, as well as the mechanics of thermal expansion and contraction that dictate how reliably the chip and its packaging survive temperature changes.AI Could Short-Circuit RFIC Design The design of a radio-frequency integrated circuit requires human intuition and multiple, often-repeated optimization steps. The hope is that through an understanding of Maxwell’s Equations, an AI can be taught to short-circuit this process and quickly produce a design.Simultaneously accounting for all the physical constraints these impose makes the design space almost impossibly large. Every decision involves complex priorities that often compete with one another, preventing the optimization of any of them.To better understand the issue, let’s walk through the steps involved, after which you’ll better understand why a single new chip design takes years and tens to hundreds of millions of dollars. Most of the area of radio-frequency integrated circuits is dominated by complex electromagnetic structures. Human-designed RFICs, like this broadband power amplifier [1], start with templates and follow a symmetric, understandable pattern. But freed from the constraints of human-designed templates and the need for humans to even understand the rationale of electromagnetic structures, power amplifier ICs [2–5] and low-noise amplifiers [6] can take on truly wild-looking yet efficient designs. SENGUPTA LABLet’s say you’re an engineer assigned to design a new 28-gigahertz power amplifier for a 5G-millimeter-wave handset. (This is the type of RFIC that boosts the 5G signals on your phone and transmits them to the antenna where they can be picked up by a distant base station). Where do you start?RFIC design has some features in common with house building. Just as the blueprint for a house dictates the number of bedrooms and bathrooms to be built and the hallways connecting them, the blueprint for an RFIC—called the architecture—establishes the kinds of elements the RFIC needs to fulfill its intended function. Instead of rooms, the architecture includes, for example, the number of stages of amplification your power amplifier needs. Instead of hallways, it shows the paths that signals must take to get through those stages.The blueprint for RFICs is actually mostly hallway; passive elements, like inductors and transmission lines, take up far more real estate than active elements like transistors.Here’s why. As you have probably experienced yourself, a typical CPU’s transistors overheat when faced with operating frequencies of just a few gigahertz. The frequencies RFICs can operate at are higher by an order of magnitude—28 and 39 GHz for 5G signals, 26.5 to 40 GHz and even higher for satellite communications, and 77 GHz for automotive radar. Under this onslaught, a CPU’s transistors would fail.RFIC transistors avoid this fate because these chips cleverly manage the signal’s energy with careful electromagnetic design. This takes the form of byzantine networks of metal elements that dominate the chip’s real estate. These structures are geometrically regular, often symmetrical, and so intricately constructed they sometimes resemble lacelike filigree. But while they may look decorative, they are essential to the chip’s functioning.Electrically speaking, these “hallways” work more like the chip’s plumbing. Like plumbing, this extensive labyrinth of passives confines electromagnetic energy only to the places it should be traveling around the chip.The major challenge in RFIC design is putting all these elements together to ensure they work, just as constructing a house from its blueprints demands exact specs for load-bearing beams, pipes, and external walls. On an RFIC, the architecture needs to be realized with physically fabricable transistors and passive components that are connected just so, to permit the signal to travel through the chip and be processed. The way these devices are connected locally is what we call the circuit’s topology.The RFIC Design ProcessTo make that power amplifier, then, your first step is to identify a candidate circuit template: The combination of structures that will meet the goals of a particular architecture with a specific circuit topology. Over the years, researchers have eased your burden by developing reusable design templates for specific functions. For example, templates suggest how many amplification stages a circuit needs (because sometimes, combining the output of two smaller amplifiers will result in better bandwidth and efficiency than you would get from a single larger one). And they suggest what the general configuration of the passive structures should be. Today there is an extensive library of such templates.However, these can’t simply be used off-the-shelf, because each comes with trade-offs. Some have better gain at the expense of stability; some better bandwidth at the expense of efficiency; still others are more energy efficient at the expense of output power, and so on. There is rarely a clear best choice.To arrive at the “sweet spot” where all these different parameters are balanced into optimal harmony, designers will typically lay out several different versions of the circuit, using intuitions and methods they have picked up in their years of training.The challenge is that the decision around the architecture, circuit topology, or the electromagnetic passives cannot be done separately. One decision influences the others. So, designing an RF circuit can often feel like trying to fit an oversized carpet into too small a room—press down one corner, and another pops up.At microwave and millimeter-wave frequencies, even the smallest misstep is the difference between a chip that works and one that doesn’t, and any number of things can go wrong. For example, when an electromagnetic wave encounters a transistor—or any other component —the path it travels must be properly “matched” to what comes next. If it isn’t, some of the energy reflects backward instead of flowing forward. Imagine trying to connect a high-pressure fire hose directly to a narrow garden hose. Without the right adapter, water will splash backward at the junction. Very little will make it through. In electronics, this is called the impedance-matching problem.To prevent those reflections, engineers design special transitions, essentially microscopic adapters, that smooth the handoff between components. On a chip, these adapters can be surprisingly intricate. They don’t just pass the signal along; they can also split it, combine it, or distribute it across multiple paths with carefully controlled timing and strength.Once you’ve done the architecture, plumbing, and everything in between comes the moment of truth. Have all the choices you have navigated through the enormous design space resulted in an RFIC that meets its specifications? If the specifications are not met, you will have to go back, either redoing the topology or the entire architecture, and repeat the whole process. So get ready for months of time- and resource-heavy simulation and iteration. Perhaps you now see why, for decades, a core belief has persisted in the RFIC community: “RF design is an art.” It was said that only an experienced designer—with an artisanal understanding of how the pieces make up the whole—could master the subtleties of analog and RF design. Unfortunately, this entrenched notion has long held back algorithmic innovations in the field just when we need them most. Traditional, artisanal RFIC design is hitting its limits as the complexity of these systems inexorably grows.AI for RFIC DesignWhile RFIC designers continued their battle against their “oversized carpet” problem, a series of interesting developments emerged in allied disciplines. Across a range of other previously intractable problems like protein folding and climate modeling, AI has been able to successfully navigate multidimensional complex spaces. This gave us the incentive to look deeper into AI for RF. After all, the combinatorial complexity of protein folding is not that different from the nature of the design space in our domain.We were not the first to think of using artificial intelligence to speed up parts of RFIC design. Researchers had previously trained machine learning algorithms on circuit templates in the hope of speeding up the normal optimization processes. While this approach was undoubtedly faster than humans at optimizing templates, it still relied fundamentally on libraries of existing designs invented by humans.Training an AI to Design a Chip A machine learning system learns to do end-to-end RFIC design like other AIs learned to play such games as Go. Essentially, it turns the process into a game, learning from the results of its own efforts.We didn’t want that. We wanted to break free from the restrictions of prefabricated topologies. Because while a designer’s experience and hard-won heuristics are crucial to building a working design, they also place fundamental limits on it. Furthermore, such an approach would necessarily require simulation steps as part of the optimization cycle, and even the fastest simulations use a lot of computing resources. Worse still, in many advanced cases, such as for broadband designs, there are no existing templates.But if we didn’t start with templates, where could we start?The goal here was to allow algorithms to determine—entirely from scratch—every parameter for architecture, constituent circuits, and electromagnetic passives. This approach differs fundamentally from conventional optimization, which is limited to determining the parameters—like transistor dimensions and passive component geometries—that optimize structures originally devised by humans.In our new approach, the architecture begins essentially from nothing and is progressively assembled through successive iterations. The system explores the design space by generating myriad candidate circuit combinations and mapping the resulting performance trade-offs as it navigates this landscape. Because the process is not biased by prior human design choices, it can produce completely novel circuit topologies that look markedly different from those created by human designers.In some ways, the approach echoes AI systems such as AlphaGo Zero, which achieved superhuman performance not because it was trained on games played by humans but because it explored the rules by playing against itself. Similarly, our algorithm develops new circuit architectures by exploring and evaluating its own design strategies. In so doing, it learns to understand circuits, electromagnetics, and the close codesign they need to achieve the end-to-end design of RFIC.Inverse Design for RFICsTo realize this capability, we proceeded in two stages. First, we developed a reinforcement-learning (RL) framework that determines the optimal system architecture, circuit topology, device parameters, and even the properties of the electromagnetic interfaces that connect different circuit elements. In this stage, the algorithm effectively defines how signals should propagate and interact across the system.The algorithm trains very similarly to how a computer learns to play a game. If you let it play enough times, it can learn to play better by observing the relationship between the actions it took and the score it achieves. In a similar way, the RL agent here learns to design effective circuits by playing with a set of combinations, and over time, it can map the space between the circuit performance to its architecture, topology, and parameters. This training takes a few days to a week, but once trained, the agent can design circuits very quicklyThe next step was to determine the physical structure of the IC’s electromagnetics—the plumbing—that can create the desired properties of the passive elements, which are characterized by a set of metrics called scattering parameters. These measure if a signal entering a component actually moves forward—or is reflecting backward, being wasted, as in our previous example with the fire hose and the garden hose.Deriving the structure from the desired scattering parameters is an example of an approach called inverse design, which appears across many areas of engineering. In structural engineering, for example, one might collaborate with an architect on a physical goal—such as creating large interior spaces with high ceilings—and then determine the arrangement of arches or buttresses that can support it.Generative AI for Electromagnetic NetworksBut RF integrated crcuits pose a particular challenge for inverse design: The process must account simultaneously for circuit behavior and the electromagnetic responses of the interconnects and passive elements that link them together. But it has to figure that out without doing a lot of artisanal iterating.So we replaced our RF circuit simulator with an AI-based emulator. This AI model can predict the behavior of electromagnetic fields going through any structure—even totally arbitrary two-dimensional shapes—without having to compute the underlying physics from scratch, as simulation tools do. It would predict the solution of Maxwell’s equations and tell you the scattering parameters for any structure you showed it, without actually doing the math. With such an AI in hand, what a time-consuming electromagnetic solver normally takes minutes or hours to accomplish is reduced to milliseconds.We chose to build our emulator around a convolutional neural network—a machine learning model that has been remarkably successful for image processing. Such networks can extract spatial features from any structure, and it turns out that the image of a structure contains a lot of spatial information that can accurately predict its electromagnetic performance. Then we trained it on a vast number of random pixelated structures whose scattering parameters had been labeled.Once we had our inverse-design RL and suitable AI emulator, we essentially had an end-to-end AI designer. So we asked it to design us a power amplifier.Unconventional RF ArchitecturesIn 2023, we published this proof of concept—a power amplifier targeting the millimeter-wave band, specifically spanning 30 to 100 GHz, which covers most of the relevant 5G and radar frequencies. The final design achieved the best combination of wide bandwidth, output power, and efficiency then reported for a silicon-based power amplifier—meaning it could amplify a large amount of data across a wide swath of frequencies—while maintaining record efficiency.The structure of the IC’s electromagnetic pathways was unlike anything any human would ever consider. Since the AI is not trained on human designs, the layout that emerged looked more like an arbitrary pattern or perhaps a QR code than the regular symmetrical structures we are used to seeing.One unexpected insight revealed by this prototype, and our research generally, is that there’s no evidence that the templates we’ve historically relied on are even close to optimal for modern design goals. It’s not that a human designer can never come up with a better design. But with the removal of the templates and the time to synthesize cycle upon cycle of optimized circuits, it is now clear that AI-driven synthesis could break traditional design barriers and push the limits of RFIC capabilities.Our 5G amplifier had only one input port and one output port. Adding more inputs and outputs to a design is not straightforward. Every port electromagnetically couples to every other port, so the scattering parameters quickly add up. Two ports give you four scattering parameters. Four ports, 16 scattering parameters. The math gets ugly fast. Could our model keep up?We next trained our model on larger classes of electromagnetic structures with many input and output ports. In 2024, we published work showing that multiport integrated circuits are no problem for these AI algorithms either. Where previously multiport electromagnetic simulation required days or weeks of toil, this model evolved new structures in minutes. Since then, a plethora of work in the space by research communities across the globe have demonstrated the power of inverse design in RFIC.Combining the reinforcement learning framework with the inverse design, we now had the ability to create an RFIC from specifications all the way to a fabrication-ready layout. We’ve so far shown this is true for RFICs ranging from low-noise amplifiers to subterahertz and broadband power amplifiers. The hope is that this will work just as well for other circuits.Making AI Designs InterpretableOur goal was to make RFIC design better and easier, but we didn’t want to make it beyond human understanding. Chip testing and debugging is a long, arduous process, sometimes even more so than design. Engineers often prefer ICs to have interpretable structures, so that if a problem crops up, they can understand how the chip works well enough to debug it.To create structures that are more interpretable, we turned to diffusion models, which you may know from their remarkable ability to generate realistic images from text prompts.AI-driven synthesis could break traditional design barriers and push the limits of RFIC capabilities. Imagine you go to your favorite image-generation engine and ask it to create a painting of the sky in the style of Picasso, Van Gogh, or Michelangelo. You will get images that capture the essence of their brushstrokes, their use of colors, and their framing. All are pictures of the sky nonetheless, but in different styles.Electromagnetic design is similar in that multiple structures can have very similar electromagnetic responses. Instead of using text input, we used scattering parameters as our input, and the electromagnetic structure of an RFIC chip as our output. As part of the inputs to the diffusion model, we created a dial that sets the spatial frequency of the final structure. By turning the dial, a designer can direct the model to synthesize structures with low (classical-looking and interpretable), medium (mazelike structures), or high (pixelated or arbitrarily-shaped) spatial frequency.From prompts to output, the entire process took about 6 minutes. With this diffusion model, algorithms can now both discover novel architectures and accelerate the creation of conventional, so-called classical ones.All an RFIC designer needs to do is specify virtually any valid set of scattering parameters. As long as they are physically realizable under Maxwell’s equations, the model pops out a corresponding structure as if it were a vending machine.The Future of AI-Driven RFIC DesignThe results of our investigations have drawn the attention of the RF community. The traditional bottom-up design process is clearly beginning to reverse.But there are still questions: How generalizable are these methods? Can they consistently deliver truly high performance? Can we get to a place where AI produces designs that maximize every conceivable trade-off, holistically optimizing every parameter to its most ideal physical state? We want to take this strategy beyond RFIC design and invent other kinds of circuits that are different from anything humans have ever done.These are exciting and ambitious prospects, but we are not there yet. AI can hallucinate a design that creates bad circuits that don’t work. This means verification methods need to remain under human oversight. And, while hallucinations are rare, it would still be good to reduce their occurrence.History suggests that meeting these dreams of the future will take much more data than we’ve been using. Before the creation of the ImageNet repository—a repository of 14 million varied, human-annotated images—image-recognition models didn’t function well in the real world. The datasets they had been trained on were too tiny to be effective. ImageNet’s massive amounts of training data ushered in a revolution that led to AI that can generalize and recognize images in the wild. The rest was history.If the goal for RFIC and analog design is a universal foundational model—something that learns the governing laws of electromagnetics and circuit behavior—then we also need data.The good news is that this data is plentiful. Around the world, countless engineers at companies and academic labs simulate nearly identical RF circuits and passive structures every day. The bad news is that it’s all locked away behind nondisclosure agreements.Open ecosystems have propelled other areas, and we think the RFIC community should do the same. There had been some movement toward this. Natcast, the operator of the U.S. CHIPS and Science Act’s R&D program, would have bolstered shared infrastructure and innovation for the next generation of wireless, sensing, and defense technologies. Unfortunately, both the organization and the program it ran specifically for machine learning and RFICs have been closed.But the momentum Natcast’s effort sparked hasn’t died out. Building on our early work, groups across the community have already demonstrated remarkable advances. AI-driven IC design is part of a much broader technological shift. From biology and materials science to automotive and aerospace engineering, AI is reshaping how complex systems are conceived and optimized. Deeper collaboration between AI researchers and chip designers will unlock the field’s full potential. It’s by no means a foregone conclusion, but if we get this right, this genie won’t stay in its bottle.

MIT Technology Review Insights · MIT Technology Review

The emergence of the web data infrastructure layer for AI

AI is booming. New use cases are emerging each day. To capitalize on the technology’s potential, enterprises require data at scale. In many cases, though, the relevant information is blocked or unstructured, which limits its use by AI models.  To understand this challenge, consider the foundation of the web itself. The web was not designed…

Saturday, June 20, 2026

NEWSJun 20, 2026

The Atlantic created a searchable database of the music used to train AI