Every published Top AI Stories item tagged with NVIDIA, newest first.
The inference company Wafer published benchmarks on July 31 putting Moonshot's Kimi K3, at 2.8 trillion parameters, at 952 tokens per second per node on AMD's MI355X against 1,568 on Nvidia's B300. The AMD part is slower in raw throughput but roughly 2.4-times cheaper per GPU, which works out to 48 tokens per second per dollar versus 33. Its 288 gigabytes of memory per GPU also lets the model fit in a single node where the Nvidia configuration needs two.
Anthropic published a position paper stating plainly that it "has never advocated for a ban on open-weights models," calling open models without dangerous capabilities a public good. The post follows an open letter organized by Nvidia, signed by Meta, Microsoft and Mistral, that Anthropic pointedly declined to join. What Anthropic does want is narrower: chip export controls on China, a crackdown on state-backed distillation, and mandatory pre-release safety testing for every sufficiently capable model, open or closed.
Nvidia is putting billions into Safe Superintelligence, the lab Ilya Sutskever founded after leaving OpenAI, in a deal Bloomberg sizes at $5 billion. The investment buys access to Nvidia's Vera Rubin platform and lifts the lab's compute by about an order of magnitude. Sutskever framed it as a threshold moment: "We have research that is worthy of scaling up, and having access to a big NVIDIA computer will let us do so." Safe Superintelligence has now raised $7 billion at a $32 billion valuation.
The Wall Street Journal reports that Nvidia is in talks to guarantee $250 billion in financing so OpenAI can lease a 10-gigawatt campus that SoftBank's energy arm is developing in Piketon, Ohio, on the site of a former uranium enrichment plant. The full project could cost more than $500 billion, and Nvidia is separately discussing another $350 billion tied to chip purchases. Investor Michael Burry, who is short Nvidia, called the structure circular — a chip vendor guaranteeing its own customer's spending on its own chips.
Nvidia and South Korea's SK Group announced an expanded partnership valued at over $500 billion, covering multi-gigawatt AI factories and a long-term memory supply agreement. SK Telecom will build a 2-gigawatt facility in Korea running Nvidia's next-generation Vera Rubin systems on SK hynix HBM4 memory, with the first phase due online in 2027. The memory half matters most: high-bandwidth memory has become the tightest constraint on shipping AI hardware.
Nvidia chief executive Jensen Huang published a letter — his first-ever post on the social platform X — urging Washington not to restrict downloadable AI models, and within a day the signatory list doubled from 25 companies to 50, including OpenAI, Google, Microsoft, AMD, Cisco, and GitHub. The conspicuous absences were Anthropic, the most vocal proponent of tighter controls, and Amazon, Anthropic's largest investor. With Google — itself an Anthropic backer — signing anyway, the split leaves the safety-focused lab increasingly isolated on the year's central AI-policy fight.
AMD used its Advancing AI event to launch Helios, its first full rack-scale system aimed squarely at Nvidia's grip on AI data centers. Each rack packs 72 of AMD's new Instinct MI455X GPUs with sixth-generation EPYC processors and Pensando networking, delivering up to 2.9 exaflops of inference performance. Microsoft will run Helios on Azure and Anthropic plans to install up to two gigawatts of the chips, with OpenAI and Meta also signed on. Systems ship in the second half of the year.
The White House's technology chief, Michael Kratsios, publicly accused Chinese lab Moonshot AI of running large-scale distillation against Anthropic's Fable model to help build its Kimi K3 system — and of training on export-controlled Nvidia GB300 servers accessed through Thailand. Treasury Secretary Scott Bessent said sanctions and Entity List designations are "on the table." Some researchers are skeptical distillation alone could explain Kimi K3, noting Anthropic only released Fable publicly on July 1.
Capping a Tokyo visit by chief executive Jensen Huang, Nvidia and the Japanese government unveiled a national Physical AI Initiative to build open foundation models for robots, autonomous machines, and factory digital twins — a bet on Japan's manufacturing base and shrinking workforce. Partners span Toyota, which is building assisted-driving systems and factory simulations on Nvidia platforms, robot makers Fanuc, Yaskawa, and Kawasaki, plus SoftBank, Fujitsu, and Mitsubishi Electric. Japan is pairing the effort with up to $6.2 billion in planned spending over five years toward homegrown physical AI. It is the clearest sign yet that the robotics race is consolidating around a handful of national, chip-anchored ecosystems.
Reflection, the American open-weight AI lab founded by two former DeepMind researchers, agreed to spend $1 billion for access to Nvidia's latest GB300 chips through 2029, hosted by European infrastructure provider Nebius. The deal follows a similar compute pact with SpaceX weeks earlier, as Reflection, valued at $8 billion, races to train open alternatives to OpenAI and Anthropic. Backers of the startup include Nvidia, Sequoia, and Lightspeed.
An internal memo reviewed by Reuters shows Meta plans to begin manufacturing Iris, the fourth generation of its custom MTIA accelerator, in September — designed with Broadcom and built by TSMC. Iris will not replace the Nvidia and AMD GPUs Meta already buys in bulk, but it is meant to shave the cost of scaling to seven gigawatts of compute by year-end and 14 gigawatts in 2027. Meta expects to spend as much as $145 billion on AI infrastructure in 2026 alone. The chip cleared bug testing in roughly six weeks with no major problems.
Reuters reports that DeepSeek has spent about a year secretly recruiting chip engineers and courting manufacturing partners to build an in-house processor for AI inference — the stage where a trained model answers user queries. The goal is to cut its dependence on Nvidia and Huawei as US export controls keep tightening. It follows OpenAI's Broadcom-built Jalapeno inference chip and Anthropic's own reported chip ambitions — a sign that frontier labs increasingly want to own their silicon.
Micron posted fiscal third-quarter revenue of $41.5 billion, up 346 percent from a year ago, with profit jumping to $28.2 billion — and its shares spiked enough on June 25 to briefly vault its market value past Meta and Tesla, near $1.4 trillion. The surge reflects a broader shift: high-bandwidth memory has become the scarcest part of an AI server, and Micron is one of only three suppliers that can make it at scale. Wall Street analysts are now openly asking whether the memory maker is the next Nvidia.
SpaceX agreed to lease AI computing capacity to Reflection AI, an open-source lab founded by former Google DeepMind researchers, for $150 million a month from July through 2029 — up to $6.3 billion in total. Reflection gets Nvidia GB300 chips at SpaceX's Colossus 2 data center near Memphis, Tennessee, the facility xAI built before folding into SpaceX. It joins much larger SpaceX compute deals with Anthropic and Google, turning the company's spare chip inventory into a fast-growing leasing business.
AI chipmaker Groq confirmed a $650 million funding round, led by the investment firm Disruptive, as it rebuilds after Nvidia's roughly $20 billion deal late last year to license Groq's technology and hire away its founder and much of its top engineering team. Groq is now doubling down on its "neocloud" business — renting out fast inference capacity — and restocking its executive bench. The startup last carried a valuation near $6.9 billion and did not disclose a new one.
Qualcomm is reportedly negotiating to acquire Tenstorrent, the AI-chip startup led by legendary designer Jim Keller, in a deal valued between $8 billion and $10 billion. Tenstorrent builds RISC-V-based processors it claims beat general-purpose GPUs on specific AI workloads, so the purchase would hand Qualcomm a credible data-center play against Nvidia and a path beyond smartphone chips. The talks are unconfirmed and could still fall apart, but they show how aggressively mobile-chip makers are now chasing AI silicon.
Amazon is in talks to sell its in-house Trainium chips directly to outside data centers for the first time, breaking from years of reserving them for AWS cloud customers. AI chief Peter DeSantis confirmed the discussions in Paris but named no buyers, pointing to rising demand — especially in Europe — to keep AI compute under local control. CEO Andy Jassy has called the homegrown chip business an opportunity worth 50 billion dollars a year; pulling it off would put Amazon in direct hardware competition with Nvidia for the first time.
Nvidia sold $25 billion of investment-grade notes across seven tranches maturing as far out as 2056 — its first corporate bond deal since a $5 billion raise in 2021. Demand reached more than $85 billion, over three times the offering, and the company upsized the sale from an initial $20 billion. For a firm sitting on a roughly $5 trillion market value and ample cash, tapping the debt market signals just how much it plans to spend building out AI data-center capacity.
Loft Orbital says its YAM-9 satellite became the first spacecraft to run a vision-language model in orbit, using Google DeepMind's Gemma 3 to answer plain-English queries — classifying where wilderness meets development, or spotting infrastructure near rail hubs — without beaming raw imagery to ground analysts first. The model ran on an Nvidia Jetson edge chip paired with NAVI-Orbital software from NASA's Jet Propulsion Laboratory. Onboard triage like this could sharply cut how much data satellites need to downlink, and Loft says 50 to 100 such craft would give near-real-time eyes on Earth.
Germany's Neura Robotics raised $1.4 billion in a Series C round backed by Amazon, NVIDIA, Qualcomm, Bosch, and the European Investment Bank, valuing the cognitive-robotics company at about $7 billion. It is the biggest single round ever raised by a full-stack robotics maker, and a sign that strategic players are betting hard on humanoids for warehouse and factory work. Neura already holds more than $1 billion in pre-orders for its two-armed, two-legged 4NE-1 robot, whose first units are due to ship late this year.
Nvidia told Chinese cloud customers that its new Vera data-center CPU is now open for orders and could ship as early as August, according to Reuters. The pitch is a workaround: after US export curbs choked Nvidia's GPU sales in China, the company is leaning on standalone CPUs — which face looser restrictions — to defend the market. Nvidia's finance chief has said Vera CPU revenue could approach $20 billion in the coming fiscal year.
Jensen Huang and Hyundai chair Chung Euisun used a meeting in Seoul to widen their partnership from research prototypes toward factory-ready robotics, spanning mobility, manufacturing, and autonomous driving. Hyundai will stand up an AI supercomputer built on 50,000 Nvidia Blackwell GPUs to train models for self-driving and smart factories, and the two companies will work with the Korean government on new AI technology centers. Huang called Hyundai one of the best robotics companies in the world. The deal is one of the largest national commitments yet to physical AI — the effort to move machine intelligence off the screen and into machines that move.
Bloomberg reports that Beijing is preparing to spend roughly 2 trillion yuan, about $295 billion, over five years on a network of interconnected AI computing hubs run mostly by state firms like China Mobile and China Telecom. The blueprint, drafted by agencies including the National Development and Reform Commission, would source at least 80 percent of the hardware, including AI chips, from domestic suppliers such as Huawei — effectively designing Nvidia and AMD out of the country's largest buildout. It is the clearest signal yet that China intends to win the AI compute race on homegrown silicon.
Google will pay SpaceX about $920 million a month from October 2026 through June 2029 — roughly $30 billion in total — for access to around 110,000 Nvidia chips, capacity SpaceX first built for its own xAI division. Google Cloud called it short-term "bridge capacity" for its Gemini Enterprise platform while its own data centers scale up. The deal lands days before SpaceX's IPO, where Google already holds about a 5 percent stake, and echoes an earlier SpaceX compute arrangement with Anthropic worth roughly $1.25 billion a month.
In this week's Stratechery analysis, Ben Thompson argues that AI agents shift the center of gravity from the device to the cloud — making Nvidia's GPU-heavy RTX Spark AI PC a poor fit, since agents want strong local CPUs that call out to cloud inference, while praising Microsoft's Project Solara, which treats the cloud as the hub and phones and PCs as interchangeable spokes. He frames Microsoft's in-house MAI models as a way for cautious enterprises to own custom agents without handing their workflows to frontier labs.
At Computex, Intel detailed Crescent Island, a data-center GPU built on its Arc Xe3P architecture and aimed at AI inference. Partner cards can carry up to 480 gigabytes of LPDDR5X memory — more than Nvidia's and AMD's flagships — while skipping the scarce, expensive high-bandwidth memory those chips rely on, which Intel says makes the card cheaper to produce and run on a 350-watt air cooler. Intel is targeting a second-half 2026 launch as it tries to chip away at Nvidia's data-center lead.
NVIDIA used its Computex keynote to enter the personal-computer market, unveiling the N1X — an Arm-based chip co-designed with MediaTek that pairs a 20-core processor with a Blackwell graphics processor and 6,144 CUDA cores. It will debut in Windows laptops from Microsoft, Dell, HP, ASUS, Lenovo, and MSI before the 2026 holidays, with performance models priced above $2,000 to challenge Apple's MacBook Pro and longtime chip leaders Intel and AMD. A lower-power N1 variant will start under $1,500.
Separately at the same Computex keynote, NVIDIA launched Nemotron 3 Ultra, the largest model in the open-weights family it first previewed in December. The hybrid Mamba-Transformer mixture-of-experts (MoE) model carries roughly 550 billion total parameters with about 50 billion active per token, and NVIDIA says it tops US open-weights rankings while running about 30 percent cheaper than leading alternatives. The weights and training recipes are free to download.
Groq is raising 650 million dollars to refocus the company on its inference neocloud — the on-demand cloud platform powered by Groq's own AI chips — after a December 2025 arrangement saw Nvidia pay 20 billion dollars for senior Groq engineering talent and a hardware license. Existing investors are leading the round, with Disruptive and Infinitium committed to fill any unsubscribed shares. Adam Winter is now interim CEO and Matt Eng interim CFO. The company argues inference is a much bigger market than training right now, and the wedge for the smaller team that remains.
At the groundbreaking for a new Taipei headquarters, Jensen Huang announced that Nvidia will spend $150 billion a year in Taiwan, up from $10 to $15 billion annually four to five years ago. The headquarters will employ 4,000 people and target a 2030 opening, anchoring Nvidia closer to TSMC, which fabricates its chips, and Foxconn, which assembles them into server racks. The pledge lands as Washington has spent two years pushing chipmakers to build US capacity, and it cements Taiwan rather than the United States as the structural center of advanced AI manufacturing for at least the rest of the decade.
In this week's Stratechery analysis, Ben Thompson argues that SpaceX's rumored IPO at a $2 trillion valuation only makes sense if Starship enables data centers in orbit. His core thesis: terrestrial data center expansion is now constrained more by community zoning opposition than by power generation, and the existing Starlink V2 Mini satellite form factor — about 7.4 meters by 2.7 meters — is comparable to NVIDIA's NVL72 rack. Combined with Starlink's laser interconnects, the constellation already has the network topology required for distributed orbital compute; power dissipation and radiation hardening become engineering problems rather than fundamental obstacles to agentic-inference workloads.
In this week's Stratechery analysis, Ben Thompson argues that Nvidia's revised reporting taxonomy — splitting hyperscaler revenue from everyone else — reveals the contours of an emerging commoditization fight at the top of the stack. Nvidia is "fighting commoditization" with its largest customers, where hyperscalers increasingly design their own silicon, while it "runs the whole stack" for AI clouds, sovereigns, and enterprises. The reframe lands one week after Nvidia's record $81.6 billion quarter on May 20, when hyperscaler revenue held at roughly half of the $75.2 billion data-center total.
At a press event in Taipei this week, NVIDIA CEO Jensen Huang told reporters the company has *"largely conceded"* China's AI accelerator market to Huawei, with NVIDIA's share now near zero after the US H200 export-clearance stalemate dragged into a second month. Huawei expects roughly $12 billion in 2026 revenue from its Ascend line — up from $7.5 billion in 2025 — on orders already placed by Alibaba, ByteDance, and Tencent, all of which deployed DeepSeek V4 services within hours of the model's Ascend-optimized release in April. Huang said he still expects Beijing to eventually allow H200 imports, but for now the homegrown stack is shipping while NVIDIA's clearance letters sit in customs.
NVIDIA Labs has quietly posted Nemotron Diffusion to Hugging Face — a 14-billion-parameter language model that switches between autoregressive decoding, parallel diffusion decoding, and a *"self-speculation"* mode that drafts with diffusion and verifies with autoregression, all without changing model weights. The accompanying technical report claims a 2.2-times throughput lift over the comparable Qwen 3 8-billion-parameter baseline at matched accuracy, scaling to 850 tokens per second on a GB200 (a 3.3-times lift). Base, instruct, and vision-language variants are all open-weight; the architecture is positioned as a path from memory-bound to compute-bound inference as GPUs keep outrunning memory bandwidth.
Nvidia reported $81.6 billion in revenue for the quarter ending April 26, a 20% sequential jump, with data-center revenue at a record $75.2 billion. The earnings disclosure also surfaced $43 billion in non-marketable startup equity — nearly double the prior quarter — including a previously-undisclosed []0 billion commitment to OpenAI. On the call, Jensen Huang positioned the new Vera CPU as a "brand new toolIds: 00 billion total addressable market" built for autonomous-acting agents.
NVIDIA Labs released SANA-WM, an open-source 2.6 billion parameter world model under Apache 2.0 that generates one-minute videos at 720p resolution with 6 degrees of camera-pose control. The team reports training on roughly 213,000 video clips in 15 days on 64 H100 GPUs, and says a distilled variant runs on a single consumer GPU. The paper positions SANA-WM as a baseline for embodied-AI and robotics research at a fraction of closed-model compute budgets.
In a TechCrunch interview, Menlo Ventures partner Deedy Das argues that roughly 10,000 founders and employees at OpenAI, Anthropic, NVIDIA, Meta, and xAI have crossed 20 million dollars in personal wealth over the past five years — while engineers earning under 500,000 dollars a year increasingly fear they cannot get there from here. Das calls the split in San Francisco "the worst I've ever seen" and ties the gap to a broader "deep malaise about work" pervading even well-paid technical roles in the AI era.
In this week's Stratechery analysis, Ben Thompson argues that AI compute is bifurcating into three distinct workload categories that need fundamentally different hardware. Training keeps high-bandwidth GPUs (NVIDIA's lock-in); answer inference rewards token speed (Cerebras's WSE-3 packs 44 gigabytes of on-chip SRAM at 21 petabytes per second of bandwidth versus NVIDIA H100's 80 gigabytes of HBM at 3.35 terabytes per second); agentic inference, where humans aren't in the loop, mostly cares about memory capacity and cost-per-token at scale. Thompson treats Cerebras's revised IPO pricing of $150 to $160 per share (up from $115 to $125) and Anthropic's lease of 220,000 NVIDIA GPUs at SpaceX's Colossus 1 (the 300-megawatt Memphis data center) as concrete evidence that buyers are now sorting their compute spend by workload category rather than picking a single vendor.
Nvidia has now committed more than $40 billion to equity investments in AI companies in 2026, including a single $30 billion stake in OpenAI, up to $3.2 billion in glassmaker Corning, and up to $2.1 billion in data-center operator IREN. Wedbush analyst Matthew Bryson described the pattern as "squarely circular" — money cycling between chip vendor, model customer, and infrastructure provider. The chipmaker has also closed roughly two dozen private startup rounds plus 67 venture deals across 2025, intensifying scrutiny of how concentrated the AI capital stack has become around a single supplier.
NVIDIA Research published Nemotron Elastic, a post-training method that embeds 30, 23, and
12 billion-parameter nested reasoning models inside a single 30 billion-parameter parent —
extractable via zero-shot slicing without further fine-tuning. The recipe achieves a 360-times
token reduction over pretraining from scratch, and the 30 billion checkpoint compresses to 18.7
gigabytes under NVFP4 quantization. The 23-to-30 billion configuration advances the
accuracy-and-latency Pareto frontier with up to 16 percent higher accuracy and 1.9-times lower
latency than the default Nemotron Nano v3 budget control. All three precision variants (BF16,
FP8, NVFP4) are available on Hugging Face under nvidia/NVIDIA-Nemotron-Labs-3-Elastic-30B-A3B.
Cerebras Systems filed to sell 28 million shares priced between $115 and $125 per share, targeting $3.5 billion in proceeds at a $26.6 billion valuation — the largest US tech IPO of 2026 so far. OpenAI is one of the chipmaker's largest customers under a multi-year contract worth more than $10 billion signed in January, and holds a $1 billion secured loan plus warrants for over 33 million shares, potentially making OpenAI a major shareholder post-listing. The offering puts Cerebras' Wafer-Scale Engine 3 against Nvidia in a public-market test of GPU pricing power and frontier-lab compute lock-in.
The Department of Defense announced contracts with Nvidia, Microsoft, AWS, and Reflection AI to deploy AI on Impact Level 6 and Impact Level 7 classified networks — the most sensitive systems short of compartmented intelligence. The deals follow earlier agreements with Google, SpaceX, and OpenAI, and are framed as a vendor-diversification push following a public dispute with Anthropic over usage restrictions. Contract values were not disclosed.
Mistral released Mistral Medium 3.5, a 128-billion-parameter dense model with a 256,000-token context window, available open-weights under a modified MIT license at $1.50 / $7.50 per million input/output tokens. The release pairs with Vibe remote agents — async cloud coding agents launched from CLI or Le Chat that handle refactors and test generation in parallel. Reported benchmarks: 77.6% on SWE-Bench Verified, 91.4 on τ³-Telecom, and self-hosting on as few as four GPUs.
Swedish legal-AI startup Legora closed a $50 million Series D extension led by NVentures (NVIDIA) and Atlassian at a $5.6 billion post-money valuation, claiming over $100 million ARR and 1,000+ law firms across 50 markets. Harvey still leads at an $11 billion valuation with 100,000 lawyers and 1,300 organizations as customers. Both companies are launching celebrity ad campaigns — Harvey with Gabriel Macht, Legora with Jude Law — and expanding into each other's geographies.