Kuo Estimates OpenAI and Anthropic Demand at 26 Million TPUs in 2028–29. Company Unit Counts Are Undisclosed.
A supply-chain analyst’s latest survey assigns most of Google’s estimated 2028–29 TPU shipments to two outside customers.
AI moves fast. We only write what’s worth reading.
Section
Coverage of AI chips and the hardware behind them. Every story, newest first.
92 stories published · newest first
A supply-chain analyst’s latest survey assigns most of Google’s estimated 2028–29 TPU shipments to two outside customers.
A campaign to reinstate Nvidia's co-founder, whose resignation Nvidia announced on November 12, says more than 115,000 people have signed.
A Wall Street Journal account, relayed by Wccftech and TechPowerUp on November 13–14, says Nvidia offered financial guarantees to AI startups and small cloud providers that could not otherwise afford its GPUs, while…
Reuters read more than 150 Chinese procurement notices, investment plans, construction contracts and corporate disclosures, and found projects worth more than 23 billion yuan (about $3.2 billion) designed around Nvidia…
SpaceX said on November 12 that it plans to invest more than $74 billion in two AI supercomputing campuses: over $24 billion at Starbase in Texas and roughly $50 billion in Cochise County, Arizona.
AMD announced the 288GB Instinct MI450X for Helios and separately reported MI455X inference results against a published Nvidia MLPerf entry. The release gives no MI450X price or availability date.
Nebius beat its third-quarter estimates and posted its first GAAP operating profit. Then it raised next year's spending plan to $40–45 billion, said prepayments and contract-backed loans would cover most of it in…
Will DePue of Unemployment Labs published a checkpoint-forensics report on November 12, 2026, recovering the hardware layout embedded in the starting weights of six Chinese frontier models released since April 2026.
Amazon announced SGLang-Neuron is publicly available on AWS Trainium on November 12, 2026, through a Deep Learning Container.
Nvidia's board accepted Jensen Huang's resignation as president, chief executive and director on November 11, effective immediately, the company announced on November 12. CFO Colette Kress serves as interim CEO.
Reuters reports that Nvidia's board met late into Wednesday night to decide whether to remove Jensen Huang. As of Thursday's premarket reports on November 12, Nvidia had announced no outcome, named no successor, and not…
In its IPO roadshow video, Anthropic says its Mythos model got Claude running on AMD chips in about three days, a job that used to take dozens of engineers months.
Procurement records from Northwestern Polytechnical University, on the US Entity List since 2001, show purchases of Nvidia H200 chips through a domestic supplier and rented access to servers built on restricted chips…
Nvidia shares fell for a second straight session on Tuesday, closing at $168.89. A new report alleging at least $7.5 billion in chip shipments to China, a reported naming of outside counsel for the board's…
TSMC's board on November 10, 2026 approved up to $500 million for Anthropic shares at the IPO price and up to $500 million for OpenAI preferred stock. Both figures are authorization ceilings, not completed purchases.
Google expanded its eighth-generation TPU line on November 9, 2026, announcing TPU 8tx and TPU 8i Max. Google itself draws the line that matters here: these are new systems built around existing chips, not new silicon.
Rebellions plans to connect its developing R200 processor through Nvidia's NVLink Fusion and use MGX for rack-scale systems.
Nvidia's board announced on Sunday, 8 November, that it is opening an independent investigation into recent reporting that founder and CEO Jensen Huang "has been the target of a foreign intelligence effort".
The Information reported on 6 November that TSMC had started drafting plans to make 3D DRAM at Nvidia's request. TSMC's answer, published by Tom's Hardware a day later, is two sentences about what TSMC is.
NVIDIA's largest FLUX 3 Klein speedups are measured against uncompiled BF16 inference with vanilla attention. The absolute times and output differences give a more useful picture of the gains.
Supermicro, IBM and NVIDIA have packaged shared KV-cache infrastructure using Petascale storage, IBM Storage Scale and NVIDIA Dynamo.
The Financial Times reported on November 4 that Oracle has cancelled a five-year, roughly $7 billion lease that would have given Tencent access to as many as 100,000 advanced AI chips in Southeast Asia, after US…
AMD reported record third-quarter 2026 revenue of $13.4 billion, up 45% year over year, with Data Center revenue nearly doubling to $8.5 billion.
onsemi reported third-quarter 2026 revenue of $1.722 billion on November 2 and disclosed $180 million in AI data center revenue.
Marvell Technology agreed to acquire Lumentum in a $115 billion implied enterprise-value, all-stock transaction on November 2, 2026.
Ceva announced on November 2, 2026 that Mobilint's REGULUS edge AI system-on-chip, combining its SensPro AI DSP and NeuPro-Nano NPU, had reached production.
ServeTheHome's comparison of Nvidia's preliminary and shipping Rubin specifications shows lower HBM4 and NVLink bandwidth, reduced NVL8 memory capacity, and higher scale-out networking bandwidth.
South Korea reported $115.28 billion in October exports, up 93.7% from a year earlier, with semiconductor shipments rising 259.4% to $56.44 billion.
Nine Companies House allotment returns underpin reports that London inference-chip startup Fractile issued about £577 million in Series C shares between September and October 2026.
Google said on October 30 that its Project Suncatcher prototype had returned its first AI response from orbit: a haiku generated by an onboard Gemma 4 model running on a data-center-class TPU.
PowerLattice, a Vancouver, Washington startup founded in 2023, has sold $86 million of a $111 million equity offering, according to an October 30 SEC filing.
Intel reported third-quarter 2026 results after the close on October 29: revenue of $16.5 billion, up 21%, Data Center and AI revenue up 78% to $7.3 billion, and non-GAAP EPS of $0.52 against its own $0.38 guidance.
South Korean inference-chip maker FuriosaAI announced a $620 million Series D at a post-money valuation of about $2.9 billion. Korea's National Growth Fund supplied $280 million.
Samsung Electronics released the final version of its third-quarter results in Seoul on October 29, 2026. The totals matched the guidance it gave on October 8: revenue of 195.25 trillion won and a record operating…
Moore Threads, China's listed GPU maker, filed its third-quarter report on October 28, 2026. Revenue rose 1,774% to 1.552 billion yuan and the company reported its first quarterly profit, ahead of the 2027 target it had…
Malaysian Customs disclosed on October 28 that it had seized 30 servers containing advanced AI chips, worth nearly RM49.5 million, at Kuala Lumpur International Airport on October 15.
LG Electronics reported its third-quarter results on October 28. Beyond the headline earnings figures, the release disclosed that cumulative orders for AI data center cooling exceeded 3 trillion won through September…
ChangXin Memory Technologies (CXMT), China's largest DRAM maker, plans to begin limited production of HBM3E, the stacked memory paired with AI processors, within the next two months, Reuters reported on October 28…
SK hynix reported third-quarter 2026 results on October 28 in Seoul: revenue of 104.62 trillion won and operating profit of 83.33 trillion won, an operating margin of 79.7%.
The Information reported that OpenAI plans to buy Google TPUs, citing one person familiar with the matter. Reuters said the report disclosed no terms or delivery details.
Teradyne reported third-quarter revenue of $1.306 billion, up 70% year over year on record Memory and Integrated Systems Test sales it attributes to AI demand.
Oracle and Google announced on October 27 that Google's Tensor Processing Units will be offered through Oracle Cloud Infrastructure, the first time Google's custom AI chips will run on a hyperscale cloud outside Google.
Nvidia announced on October 27 that it will invest $3 billion in Meta's equity offering and expand the two companies' work on Muse, Meta's personal agent.
Bank of America, Blackstone, Citi and Morgan Stanley announced on October 26, 2026 that they will finance $63 billion of Broadcom AI accelerators and networking equipment through a special purpose vehicle, which will…
Dell Technologies announced a broad expansion of the Dell AI Factory with NVIDIA on October 21, 2026: a 48U rack holding 256 NVIDIA Vera CPUs, the first rack-scale inference system built on Groq 3 LPUs, a containerised…
Cadence announced Maestro on October 26, 2026, a portfolio of domain-specialised AI models for its chip and system design agents.
The price of renting Nvidia's four-year-old H100 rose through October, on the lists buyers actually budget against.
On October 25, in a reply on X rather than a corporate announcement, Elon Musk said SpaceX had closed $40 billion in debt financing for GPU purchases.
Eleven union officials began a 24-hour warning strike outside Micron's Taoyuan plant at 09:00 on Thursday 22 October 2026, after talks the day before ended without agreement.
At GTC Berlin on Wednesday, 21 October 2026, NVIDIA introduced DSX POD, a reference design for complete Vera Rubin AI factories built by partners inside modular containers.
On 20 October 2026, Reuters reported, citing anonymous sources, that ByteDance, Alibaba and Tencent have collectively ordered more than 3 million of Nvidia's RTX PRO 5500 workstation cards — a combined total Reuters put…
On 20 October 2026, at PyTorch Conference 2026, Google open-sourced TorchTPU, a native PyTorch backend for its Cloud TPUs, co-developed with the PyTorch team at Meta.
Google completed its open Coral NPU chip design on 20 October and published the RTL — for what it describes as the complete 32-bit architecture — on GitHub, alongside a v1.0 open compiler.
Amazon announced TorchNeuron on 19 October and released the code as open source on 20 October, at the PyTorch Conference in San Jose.
Tenstorrent, the 10-year-old inference chip company led by Jim Keller, has raised $1.5 billion at a reported $6.2 billion valuation, TechCrunch reported on 19 October.
Modal said on October 19 that NVIDIA's Vera CPU delivered up to double the task throughput of the x86 Granite Rapids machine it chose for comparison, in Modal's own sandbox benchmark under heavy load.
Microsoft has terminated its agreement to anchor Nscale's planned data centre in Loughton, Essex, according to reports published on 18 October.
Li Auto announced a first external financing for its chip subsidiary on October 16. The headline total is real. So is the part most coverage skipped: the largest single contributor to the round is Li Auto itself.
Production reviews of NVIDIA's RTX Spark laptops landed on October 16, the day the machines went on sale. The integrated GPU and the memory capacity hold up. Two other numbers need a slower read.
NVIDIA's updated DLSS 5 neural rendering model starts rolling out on October 16, in NBA 2K27 and on RTX Spark laptops. The announcement's biggest number describes the model, not the game running on your screen.
TSMC reported its sixth consecutive quarter of record net profit on October 15, beat analysts' earnings expectations, and raised its 2026 revenue growth forecast.
GlobalFoundries has been selected to operate up to two new fabs in Malta, New York, with Japan providing the investment capital. The first is targeted for 2029, while the second remains conditional.
ASML published third-quarter 2026 results on 14 October 2026 with record net sales of €11.7 billion, net income of €5.0 billion, Q4 guidance well above Q3, and a statement that planned 2027 and 2028 supply will not…
At its ai-PULSE conference in Paris on 13 October 2026, French cloud provider Scaleway announced an AI supercomputer to be built in the Stockholm area on AMD Instinct MI355X accelerators, in partnership with Tele2, and…
Intel gave its Crescent Island inference accelerator a product name, the GPU D270, and final headline specs at the Open Compute Project Global Summit on 12 October.
Two of Wall Street's largest lenders are arranging a $5.012 billion loan for a neocloud founded this year. Their reported underwriting commitment covers the $1.3 billion lease-related facility, not the $3.712 billion…
South Korea's benchmark fell more than 6% intraday on Tuesday as Samsung and SK Hynix led a fifth straight day of selling in AI chip stocks — amid concerns about a Chinese memory claim and higher financing costs…
Nvidia's share of China's data-centre AI chip market fell to 34% in the first half of 2026, according to IDC figures reported by Reuters on 13 October, with Huawei second at 28% as US export controls limited Nvidia's…
AMD announced Helion on October 12, 2026: an open rack-scale reference design that gives AI agents their own CPU-heavy rack, separate from the GPUs that run the models.
Applied Materials unveiled its EPIC Center in Sunnyvale on 12 October, a 600,000-square-foot semiconductor R&D facility with TSMC, Intel, Samsung, Micron and SK hynix as founding partners.
The AI bottleneck has moved again. First it was wafers, then power, now, by Broadcom's own account, it is money — and Broadcom is not a bystander. It designs the TPUs, leases the racks and has committed the financing.
Microsoft unveiled Maia 300, the third generation of its in-house AI chip, on Monday, and said it will scale the Maia family to multi-gigawatt deployments starting in 2027, making Maia the primary inference system for…
Amazon is reportedly offering $7 billion for the chip-portable inference layer Anthropic declined at $6 billion.
AI-linked debt issuance collapsed to $23 billion in September, down roughly 80% from June's $113 billion peak, according to Morgan Stanley data reported by the FT.
The headline is acquisition talks. The substance is a menu: buy the company, hire the team and license the technology, invest more, or promise more chips.
Air cooling addresses one specific input: water used for cooling. It does not close the case on construction demand, grid load, land, ecology or community impact.
Most late-stage raises are stories about burn. This one is about lead time: a profitable hardware vendor that scaled production 20x in a year and still cannot build fast enough.
Three chip-financing deals worth roughly $140 billion surfaced within a day. Broadcom, Oracle and SpaceX are all looking to private credit or debt markets to fund AI hardware, because the buildout has outgrown the…
TSMC beat both its own July guidance and the analyst consensus with record third-quarter revenue. September's 0.6% dip looks weak only until you notice the comparison: August was the company's best month ever.
The deal puts GlobalFoundries inside one of the most constrained parts of the AI chip supply chain. TSMC will pay for U.S.-made silicon interposers, the dense connective layer that lets processors and high-bandwidth…
Microsoft's Brett Ostrum announced the preorder opening in a Devices Blog post on October 7, positioning two machines built around Nvidia's RTX Spark superchip: the Dev Box and the Surface Laptop Ultra.
Microsoft and NVIDIA held their Windows AI and Surface event in San Francisco on October 7. RTX Spark laptops went up for pre-order, led by the Surface Laptop Ultra at US$2,599, while Microsoft Execution Containers…
Lambda, the Nvidia-backed GPU cloud provider, is raising up to $4 billion at a $14.5 billion pre-money valuation, led by Coatue Management and Blackstone — likely its last private round before a planned 2027 IPO.
Etched is reviewing incoming funding offers that value the AI chip startup between $40B and $50B, according to TechCrunch's Marina Temkin — roughly double the $21B valuation of its $700M round a couple of months ago.
A federal grand jury indicted Greg Lui, 38, owner of California-based Earthmade Computer Inc., for allegedly smuggling $300 million of export-controlled Nvidia servers to China between 2023 and 2024.
Prime Intellect launched Prime Inference on October 3, a serverless inference platform for frontier open-source models running on the company's own NVIDIA Blackwell hardware.
Nvidia announced a 64GB DGX Spark configuration on October 2, starting at $4,999 and shipping October 23 through six OEM partners (Acer, ASUS, Dell, Gigabyte, HP, MSI).
Australian AI data centre operator Firmus priced its IPO at A$11 a share, raising about A$7.1 billion at a US$30.6 billion equity valuation — roughly triple what investors paid in August.
Google's first AI chips are in orbit. Four Trillium TPUs rode a Planet-built prototype satellite to space aboard SpaceX's Transporter-18 rideshare on October 1, and both companies confirm contact with the spacecraft.
Sam Altman personally confirmed Cerebras as "a close partner," with "a deep engagement pushing on the frontiers of speed," putting speculation about the alliance to rest.
Anthropic's IPO filing discloses a Broadcom financing facility of up to $42 billion tied to its TPU lease commitments — about a third of a $125.2 billion, five-year compute lease.
Amazon is reportedly preparing to sell about $8B worth of its Nvidia Grace Blackwell chips to outside investors, then lease them back. Same data centers, same GPUs, different balance sheet.