Known for
Building the accelerator monopoly the entire buildout depends on.
Worth knowing
Featured · Model
Andy Jassy put Z.ai's GLM 5.3 on Bedrock's six-to-eight-model shelf, called the platform gigantic, then said the West is hungry for a Western replacement. Z.AI's Singapore shares fell 8.62% the next day.
· 4 min read
Platforms
Nvidia's October 30 NemoClaw post ships LangChain's tuned Deep Agents harness free, with its own sandbox and tracing bolted beside it. The enterprise paying to license that layer, DXC, printed the other half of the story on October 29: agent usage six times higher, bookings down 10 percent, a cut to the year.
· 3 min read
Chips
Ibiden expects demand for the boards under AI accelerators to outrun industry supply through 2030, and its own shares still fell hard on the day. AT&S, left off Nvidia's paid-capacity list, must answer with two cash numbers on November 5.
· 3 min read
Model
Nvidia released a free, open-weights 40-language transcription model eight days after Mistral priced batch speech-to-text at fractions of a cent. Per-minute APIs still sell where latency forces streaming, and independent benchmarks keep undercutting the vendors' own clocks there.
· 3 min read
Chips
American agencies concluded a Singapore executive with direct access to Jensen Huang was reporting to Beijing, and the president, briefed roughly a year ago, did nothing. For Nvidia holders, the China reopening is a favor granted on market grounds, not a rule.
· 3 min read
Chips
Google's TPUs enter another company's data center for the first time, beside Cerebras and SambaNova racks and an expanded Nvidia commitment, all announced in one morning. Oracle now carries the incumbent and its challengers on a $90 billion-plus annual spending plan.
· 3 min read
Data Center
CoreWeave switched on Copenhagen and committed to a four-continent AI backbone the same month its bonds fell below 80 cents and default odds passed 60%. November earnings must show backlog converting to cash before the expansion bill comes due.
· 3 min read
Data Center
Mira Murati's lab signed its models to Firmus's Asian AI factories mid-fundraise. The contract is real, but so is the US$3 billion Firmus must raise first, and the OpenAI termination clause shows what a delay would cost.
· 3 min read
Chips
A record plant purchase gave Unimicron the factory Kinsus wanted: now Nvidia is grooming its rival anyway.
· 2 min read
Data Center
CoreWeave's $104.2 billion revenue backlog rests on 4.2 gigawatts of contracted power, of which only 1.5 gigawatts is switched on. The November earnings report will show whether the backlog is a contract or a construction schedule.
· 2 min read
Data Center
The sector's best-financed buyer says buying a neocloud is "pointless," and its biggest name now owes $51.6 billion against $2.6 billion of quarterly revenue. CoreWeave, Nebius and IREN slid again overnight.
· 3 min read
Chips
Sam Kovacs of the Babylon Burns letter sold Nvidia short at about $211 into Friday's close, betting the last leaders of the AI trade fall next. His line is a weekly close below $200, and Alphabet, Microsoft and Amazon report this week.
· 3 min read
Data Center
Nvidia's own filings show $36 billion of neocloud backstops, a $105 billion guarantee on OpenAI's Ohio campus and up to $125 billion of residual value promises, while default insurance on its biggest neocloud prices a 63 percent chance of failure. The credit market is pricing the case where the guarantor pays.
· 4 min read
Chips
Nvidia is paying Ibiden, Samsung Electro-Mechanics and Unimicron to reserve production of the substrates that sit under its AI chips. AT&S, the Austrian substrate maker absent from Nvidia's list, is building the same capacity on AMD's and Marvell's commitments, and its stock sits 43% below its high with half-year results due on November 5.
· 3 min read
Chips
The Information counts more than $170 billion of Nvidia commitments behind its own customers, Huang says only working capital is missing, and 81% of production AI operators run their GPUs at half capacity. The market spent Thursday pricing who pays.
· 2 min read
Chips
On October 22, Nvidia named Taiwanese partners for a fifteenfold optical-module capacity build with no mainland names; on October 23, five Shenzhen optical companies answered with a noon announcement cluster and a CNY 1 billion buyback that flipped the ChiNext intraday. The order books are real, but the partner-list question is still open.
· 3 min read
Platforms
Mistral and Nvidia plan open benchmarks for engineering simulation. If toolmakers publish the time and compute a result takes, the per-seat software license loses its cover.
· 3 min read
Platforms
At GTC Berlin, Nvidia's partner conference, the company launched OpenShell, a free open-source runtime that contains, monitors and enforces AI agents, and named Modal's four closest rivals as integration partners. Six days earlier, Modal had priced its paid version of that isolation at $15.75 billion.
· 3 min read
Chips
Nvidia is prepaying for capacity and Besi says deliveries are slipping into 2027. SUSS MicroTec's record order book only converts to revenue if about 240 million euros ships in the second half, and the third-quarter report on November 5 decides it.
· 3 min read
Security
Nvidia named 11 security partners for its OpenShell agent architecture at GTC Berlin. CrowdStrike claimed the silicon enforcement layer with no dollar terms attached, and the security complex jumped up to 8.3% in a session on estimates that have barely moved.
· 3 min read
Data Center
Nvidia named Schneider a factory builder of its 14-day DSX POD data centers on October 21, the same day Schneider launched its RapidScale design for Vera Rubin, while Barclays warned the day before that peak pricing for power-gear makers is ending. Volume and price now point opposite ways on the same company.
· 4 min read
Chips
Nvidia named Amkor one of six suppliers in its October 21 capacity release, taking their multi-year agreements to $3 billion. Nvidia's own filing says commitments of this kind can be canceled before firm orders; the factories in Arizona and Gwangju cannot be.
· 2 min read
Chips
Samsung announced a second, bigger 2nm fab in Taylor, Texas on October 16, one day after TSMC confirmed it is only scouting the state. One foundry is building before it has the customers; the other has the customers and has built nothing in Texas yet.
· 3 min read
Data Center
The landlord hit a 52-week low of $12.79 as Nvidia's chief executive vouched for its main tenant and default insurance on CoreWeave hit a record 1,194.5 basis points. One customer supplies most of the rent, and that customer's own filing shows a $10.6 billion wall due by the end of 2027.
· 3 min read
Chips
China's accelerator market still grew, to 2.008 million units in the first half of 2026, and four Chinese designers now hold 95% of the non-GPU segment that overtook GPUs. Huawei names design software, not lithography, as the bottleneck in the chip-stacking strategy it built around export controls.
· 3 min read
Data Center
SEC filings posted on October 14, 2026, show scheduled sales by finance chief Nitin Agrawal and revenue chief Jonathan Jones cleared at $77 to $80.5 on October 12 and 13, the two trading days before the stock closed down 15.46% and the 2032 bonds fell below 80 cents.
· 3 min read
Robotics & Defense
The lab's FLUX 3 Action robot model leads RoboLab-120, a benchmark Nvidia runs. Germany's reported reply is a physical-AI agreement with Nvidia worth hundreds of millions of euros, with Black Forest Labs among the participants.
· 3 min read
Policy & Labor
Seven Senate Finance Democrats set the deadline in an October 13 letter asking whether AI data center costs were properly claimed as US research credits. Meta's own filings already price the risk of an IRS reversal.
· 3 min read
Chips
ByteDance is in advanced talks to buy bespoke MediaTek data-center chips, a deal that would hand the Taiwanese phone-chip designer its second major custom-silicon customer after Google. Nvidia cannot sell its best chips into China; it has $3.5 billion invested in the company that can.
· 3 min read
Chips
Nvidia is paying a Taiwanese test house about $1.4 billion up front for a plant that does not exist yet, its second supplier prepayment since July. The same day, Barclays told clients the AI hardware boom is mostly memory inflation.
· 3 min read
Deals
Two former Groq engineers sued the board over the $20 billion license-and-hire deal that moved Groq's technology and its people to Nvidia without a stockholder vote. Nvidia is now weighing the same structure for Reflection AI, with a deal possible within weeks.
· 3 min read
Chips
Korea's chip complex fell 7% on a Chinese memory announcement that has shipped nothing, while the HBM4 despec rumor behind SK Hynix's slide points the wrong way against the shipment record.
· 3 min read
Chips
An IDC shipment ledger seen by Reuters puts Nvidia at 34% of China's data-center AI chips in the first half of 2026, on shipments 44% below a year earlier. Non-GPU accelerators outsold GPUs for the first time, and Huawei took 28% of a market that kept growing.
· 3 min read
Economy
KKR's $996 million retail credit fund paid withdrawal requests past its own 5% cap, the gentlest breach in a quarter when Blue Owl savers asked for 39% of a fund and Blackstone's for 10%, just as six firms start raising $500 billion to lend against Nvidia compute.
· 3 min read
Deals
Nvidia's 10-K books $14.4 billion of the $16.9 billion Groq license as goodwill for the workforce and its future work. Two former engineers who kept their shares are contesting that price in Delaware, and a Justice Department request for information sits on the deal's structure.
· 3 min read
Deals
The September 3 8-K discloses the price of Nvidia's second-largest acquisition ever but none of its terms: no merger agreement, no break fee, no outside date. Holders and rival chip vendors carry that gap into a regulatory review that runs to 2027.
· 2 min read
Deals
Nvidia confirmed up to $2 billion of equity in xAI's $20 billion round on October 8, on top of a roughly $17 billion SpaceX stake and the Wall Street machinery financing the chip purchases. The collateral underneath is GPUs that lose about 30 percent of their value a year.
· 3 min read
Model
Beam cost, at listed cloud rates, under 3% of the $7.3 billion of GPU rent Reflection has committed, trails the Chinese leaders on its own benchmark table, and ships free this month against a $25 billion valuation.
· 3 min read
Chips
Eric Xu says Huawei's own data puts Ascend ahead of Nvidia in China; the claim ships with no share figure, no period and no basis. Nvidia assumes zero China data-center revenue in its guidance, and Beijing's industry ministry is telling buyers approval may come.
· 3 min read
Deals
Two former Groq engineers told a Delaware court that Nvidia's transaction took the technology and the talent while common holders were cashed out cheap. The licensed silicon is already in full production.
· 2 min read
Chips
A $35 billion loan for Anthropic's chips priced 275 basis points apart with and without Broadcom's credit support. The AI lab's prospectus locks Broadcom into $161.2 billion of leases, and a mid-November IPO sets its choice: convert the $42 billion loan into equity, or hold the debt.
· 3 min read
Data Center
Days after Nvidia was reported to be shopping chip residual-value risk to insurers, Corgi launched a data center program whose GPU "residual value protection" is built to make chips a financeable asset. The carrier behind that cover is unrated, two years old, and marked at $5 billion against $40 million of revenue.
· 3 min read
Data Center
Lenders are rebuilding the first deals in Nvidia's $500 billion compute-financing plan around guarantees and customer contracts rather than chip collateral alone. The risk the loan market refuses to take is drifting toward the balance sheet that just committed $235 billion to buybacks.
· 3 min read
Chips
Commerce's CHIPS research office made equity a condition of this year's chip R&D awards, and GlobalFoundries has issued Washington 9,907,399 shares at $37.85. At the 1 October close the stake is up about 29 percent, barred from board votes, with resale paperwork due by about March 2027.
· 3 min read
Security
CrowdStrike put SafeMind on NVIDIA's stage on 3 October, six days after Check Point wired its sub-100-millisecond agent verdict into NVIDIA's OpenShell runtime. Both are bidding for the same agent-security budget, and Check Point reports on 2 November with two actively exploited 9.8-severity holes in its own gear on the record.
· 3 min read
Policy & Labor
Credit filings show a Shenzhen- and Beijing-controlled lessor financed 700-plus servers for an AI developer, 32 of them Asus machines on restricted B300 Blackwell silicon, bound for a China Mobile hub. The same state that meters Nvidia's legal channel financed the restricted one, then scrubbed the paperwork.
· 3 min read
Chips
SMIC paid $5.58B for the 49% of SMIC North it did not own, at full utilization and at the end of the worst quarter Chinese AI hardware has had. The same week, Beijing started counting how many Nvidia chips its biggest buyers want, and that meter can reverse the demand the deal is priced on.
· 3 min read
Chips
Six open-source Ascend modules, including a CUDA rival called TileLang, move China's chip software bid from roadmap to repository. For Nvidia, the reopened-China bull case now rests on Beijing's permission alone.
· 3 min read
Policy & Labor
The Justice Department has sent Nvidia a formal demand for information about its Groq deal, a license-and-hire structure that never faced merger review. The answer could re-price every acqui-license in AI, and sharpen the review of Nvidia's pending $12.93 billion Hugging Face buyout.
· 3 min read
Deals
Third Point, Nvidia and a hedge-fund syndicate lend a loss-making neocloud $3.36 billion one week after its IPO filing, in notes reported to convert at the IPO price minus a double-digit discount that freezes above a $30 billion valuation. The structure, not the size, is the story.
· 3 min read
Deals
SK Hynix is fielding bank pitches to list Solidigm, the storage unit it built from Intel's $8.8 billion NAND exit, at up to $150 billion. Per the reports the offering covers the NAND and SSD unit only; the HBM business powering the AI boom stays with the Korean parent.
· 3 min read
Chips
Since May, Samsung Electro-Mechanics has locked in about 3.3 trillion won of long-term capacitor contracts, with roughly 700 billion won more in final negotiation. Its self-declared "MLCC's Law" says the capacitance each GPU needs doubles every year; if that holds, the bill lands with whoever has to build the megawatt rack.
· 3 min read
Chips
MIIT asked Alibaba and ByteDance to log their planned Nvidia RTX Pro 5500 purchases, and Cambricon fell 5.7 percent the next session. The half-year ledger shows what a closed market pays its champion: RMB 2.3bn of net profit on RMB 5.9bn of revenue.
· 3 min read
Chips
Samsung Electro-Mechanics' record 6.78 trillion won substrate bet is part-funded by the big-tech customers who need a second supplier. Every layer of it still runs on Ajinomoto film.
· 2 min read
Applications
Public sector is Salesforce's fastest-growing business at scale, and OpenAI and NVIDIA now ship into US government work through its cloud. The market spent the same month knocking 4.5% off CRM on an enterprise stack Meta has not even priced.
· 2 min read
Chips
The AI packaging shortage just handed Amkor its best-ever second quarter, and the money relieving the bottleneck comes from the people squeezed by it: a $1.5 billion Nvidia prepayment, hyperscaler orders shifting to Intel's EMIB, and $407 million of CHIPS cash for an Arizona campus that produces nothing until early 2028.
· 4 min read
Chips
Morgan Stanley estimates put Nvidia, Alphabet and AMD at 85% of 2027 high-bandwidth memory, Nvidia alone at roughly 37%, backed by $279 billion in disclosed supply commitments. The DDR5 market is paying the toll now, and the next reading lands at Nvidia's November earnings.
· 3 min read
Model
Anthropic's own paying subscribers now call its caution a cartel, and the suit sits unanswered over a listing investors expect near $2 trillion. The compute bills under that mark are only partly financed.
· 3 min read
Chips
China's reported nod to Nvidia chip purchases sent its own chip champions to a one-year low, in the same session Asia's chipmakers slid on OpenAI's training pause. Michael Burry is short the boom and says Trump "cannot afford to let it fall."
· 3 min read
Policy & Labor
China's industry ministry is collecting order forms for Nvidia's newest workstation card from two named giants, volumes and use cases required, before it approves a thing. The chip sits outside the US control perimeter drawn in 2022, days after export controls went unmentioned at the Xi-Trump summit.
· 3 min read
Chips
Ayar Labs has lifted its 2026 primary funding to $650 million to build optical engines that replace the copper linking AI accelerators. Nvidia, AMD, Intel and Wiwynn are all on the cap table, and Wade's own deadline is volume qualification by the end of 2027.
· 2 min read
Deals
A $6 billion license and $1 billion of equity keep Poolside independent on paper, hand its backers a payout by 2027, and move 109 engineers to Nvidia. The Justice Department is already testing that structure against the reported $20 billion Groq deal.
· 3 min read
Model
Usage fees stacked on base subscriptions pushed Harvey's gross margin from about 50 percent to minus 50 percent before an August rebuild on Moonshot's Kimi K3 fixed it. The cheapest compliant seats for that Chinese model are American inference hosts running chips export controls keep from China.
· 2 min read
Policy & Labor
The first Chinese state visit in 11 years, on 24 September, exists to extend a trade truce that expires in November. Its currency is the AI supply chain: Nvidia silicon licensed east, rare-earth licenses dangled west, and a US mining trade that sells off when peace breaks out.
· 4 min read
Policy & Labor
The labs that built the technology now want Washington to slam the brakes. The president, on speakerphone with Nvidia, calls it a hoax, and the heaviest AI users are his only happy constituency.
· 3 min read
Policy & Labor
The president put Nvidia's CEO on speakerphone to declare a "SICK conspiracy" against AI. Back at the family firm, a Hong Kong platform reselling blacklisted Chinese models is paid for in the Trumps' own stablecoin.
· 6 min read
Policy & Labor
Nvidia disclosed $17 billion of consideration for a deal that left Groq independent. Newly reported scrutiny asks whether the licensing structure avoided merger review.
· 2 min read
Policy & Labor
A Senate proposal could protect coordinated safety delays. It would also shield the government notices explaining them from public records requests.
· 3 min read
Model
Chips, power, tokens, wrappers. The margin lives in exactly one of these.
· 4 min read
#5 of 227 people with a Forbes net worth here.
