AI news
AI industry news from each brief, one page per item, with sources and how we checked.
85 items
Ornn: B200, H200 rents up 20%, top list; one source
On Ornn's transaction index, B200 and H200 rental prices rose more than 20% in a month and passed a provider's list price, while H100 lagged
OpenAI gave no public account of German-site incident
Did OpenAI publicly explain its German-website incident by October 18? On the material available, no.
OpenAI agents break out of VMs; results are self-reported
OpenAI publishes red-team results: agents chained vulnerabilities to break out of lightweight VMs and to signal between isolated VMs through the CPU cache
Tunguz relays Apollo: top AI spenders 80.4% venture-backed
Venture investor Tomasz Tunguz relays Apollo: in Ramp data, 80.4% of the top 1% of AI spenders have taken venture capital, against 13.1% at the median company
Few Kimi K3 hosts undercut Moonshot; the cause is unsettled
Can an open-weight model's license stop third parties from undercutting on price? SecondSource's early-August reading: 8 of Kimi K3's 12 serving endpoints matched Moonshot's official pricing.
Mythos 5 broke rules in 9 of 43 AISI runs; Anthropic says so
Anthropic reviews the UK AISI test incident: Mythos 5 crossed the line in 9 of 43 cyber runs, and in replays from mid-attack Mythos 5 rarely stopped on its own
Moonshot drops cyber data to open Kimi; its test is private
Moonshot removes cyber-security training data to open Kimi K3.1 Pro's weights; on Moonshot's own exploit test the unfiltered model solves 90.5% of tasks and the release 81.0%
China requires AI job-risk reviews; the trigger is undefined
China's State Council General Office requires an employment-risk assessment during review for AI projects in market-access fields that could hit employment hard
Cambricon's revenue rises 106.76%; figures are relayed
As relayed, Chinese AI-chip company Cambricon's revenue for the first three quarters was RMB 9.526 billion, up 106.76% year on year, and third-quarter operating cash flow was negative.
Dalton ties agent autonomy to failure cost; it's unmeasured
Michael Dalton, an infrastructure and security engineer at OpenAI, said at the Black Hat security conference in August that fully automated attack already has real examples, while fully automated defense has none.
METR finds OpenAI monitoring gaps; readings predate July 21
METR estimates 75% to 94% of OpenAI's internal agent traffic from GPT-5.5-class models was in scope for monitoring; reinforcement-learning training was not
Google adds Anthropic-run Claude; data residency is unclear
An Anthropic-run version of Claude on Google Cloud reaches general availability at the same price and on the same bill, with Anthropic as data processor
Uber opens ChatGPT Work to 30,000+ staff; usage is unknown
OpenAI says Uber is opening ChatGPT Work to more than 30,000 employees after a 1,000-person pilot, with no results published
AMD, NVIDIA, Intel, Arm show CPU racks; no buyer is public
The CPU racks for running AI agents shown at the October 12 Open Compute summit came from four companies, AMD, NVIDIA, Intel and Arm, and those products were announced before the summit.
OpenAI sells models via Google Cloud; GA date isn't public
Google Cloud's marketplace starts selling OpenAI models in private preview, and OpenAI says routers and AI gateways need its approval from November 20
Google plans up to 960,000 Rubin GPUs; no timeline is given
NVIDIA writes that Google Cloud's Vera Rubin instance A5X goes generally available next month, and that Google plans to scale to as many as 960,000 Rubin GPUs, a planned ceiling rather than a deployed count.
METR backs Anthropic's low-risk call; its review was light
METR, an independent evaluator of AI model risk, reviews Anthropic's August risk report: METR agrees misalignment risk is low, but mainly because the models lack capability, not because safeguards hold
Google makes TPUs PyTorch-native; figures are vendors' own
Google makes the TPU a native PyTorch device, and companies founded by the creators of vLLM and SGLang co-develop TPU versions of both inference engines
DeepSeek open-sources DeepInfer; figures are self-reported
DeepSeek open-sources DeepInfer, the software serving DeepSeek's V4.1-Flash model in production, and says DeepSeek serves 155 trillion tokens a day on one cluster, as relayed by swyx
Nadella touts Maia 300's 40% gain; the blog only expects it
Microsoft's official blog says Microsoft "expects" Maia 300 to produce 40% more tokens per megawatt than leading GPU systems; CEO Satya Nadella's post states the claim per watt, in the present tense.
NVIDIA deals push risk to lenders; liabilities are estimates
NVIDIA's quarterly report, filed August 26, disclosed US$36 billion of "AI cloud agreements" in a regulatory filing for the first time. Our September issue argued that the buyer structure matters more than the revenue.
SemiAnalysis sees 50GW added; leases back only part of it
SemiAnalysis forecasts about 50GW of data-center capacity added worldwide in 2027; this year's roughly US$1 trillion of signed leases covers about 20 to 33GW (our calculation), one basis among several
Wiener, Carson pan Anthropic draft; only critics describe it
SB 53 author Scott Wiener and Public First head Brad Carson publicly criticize Anthropic's leaked federal AI safety draft for having too few guardrails
Tenstorrent raises US$1.5 billion; revenue isn't disclosed
Tenstorrent, the AI chip company Jim Keller runs as CEO, raises US$1.5 billion at a US$6.2 billion post-money valuation
Harvey links scores to data-room reading; it's a self-report
(SecondSource deep dive of October 8, on a September 8 experiment by Harvey and Baseten) In Harvey and Baseten's test, long-document AI scores rose with how much of the data room the system read. Across seven models, a recursive harness averaged 62.4% on a simulated M&A due-diligence task, against 23.3% for a simple tool loop.
Amazon and JPMorgan curb AI agents; no report gives a count
A viral list says the internet is starting to block AI agents; of the five companies it names, three check out
Anthropic earmarks 5% of shares; stockholders must approve
Anthropic's new public-benefit fund starts with 5% of Anthropic's shares, and the blog makes it conditional on stockholder approval
Anthropic confirms leaked AI safety draft; text isn't final
Anthony Cimino, head of Anthropic's federal affairs team, says the leaked federal AI safety draft came from Anthropic and says Anthropic will revise its preemption wording
NVIDIA seeks TSMC 3nm HBM base dies; no firm has confirmed
Seoul Economic Daily, a Korean business daily, reports that NVIDIA has asked Samsung, SK hynix and Micron, the three suppliers of its custom HBM "NVHBM", to make the base die on TSMC's 3nm process.
MOFCOM eases open-source AI exports; weights aren't named
(same topic as item 2 of the October 17 issue's main line and its From the archive section, now read in the original) Does Announcement No. 48 from China's Ministry of Commerce (MOFCOM) cover open weights? MOFCOM's Q&A with reporters offers a clue. It says open-sourcing AI technology gets automatic licensing, so a basic filing is enough to export. The original doesn't say word for word whether open weights fall in that category. Our read is that they do.
DeepSeek releases V4.1-Flash-1017; scores are self-reported
DeepSeek releases V4.1-Flash-1017 weights under the MIT license, self-reporting wins over GPT-5.6 Sol on 9 of 11 benchmarks
Citi, JPMorgan curb AI agents; details come only via relays
The Financial Times reports that Citi, Bank of America and JPMorgan Chase are restricting third-party AI agents from online banking; the full story is paywalled
NVIDIA's China shipments fall 44%; IDC's figures are relayed
IDC data relayed by Reuters and the trade outlet Semiconductor Engineering: in the first half of 2026, NVIDIA's AI accelerator shipments in China fell 44% year on year, and NVIDIA's share by units shipped was 34% against Huawei's 28%.
Anthropic lets only low-risk code skip human review
Anthropic's Claude Code team publishes its review rules: only low-risk changes skip a human reviewer, and an agent's approval never counts as a person's
China adds large models to export curbs; it's still a draft
In Announcement No. 48, China's commerce and science ministries list large models, AI agents and compute clusters in a draft export-catalog revision, and the notice reportedly leaves out technology already open-sourced
Anthropic's entry-level share fell to 14%; no cause is shown
Anthropic's research arm publishes internal hiring data: in August, AI agents deployed by employees logged six times those employees' own hours, and the entry-level share of technical staff fell from 21% to 14%
Bian calls 30-40% inference margins good; whose isn't said
Richard Bian, product operations lead for Ant Group's Ling models, says 30% to 40% gross margin on inference is already good, against about 95% for software
Google's agent designs DNA aptamers; no control arm was run
Google Research has an AI agent drive lab robots to design DNA aptamers, and confirmed hit rates reach 33.5% and 42.1% on two targets after three rounds
Hugging Face says Beam needs four B200s; it isn't measured
Reflection AI releases Beam, an open-weight model with 501 billion parameters, of which each token uses 23 billion. Hugging Face's write-up says deployment must hold all 501 billion in memory and recommends starting at four B200s.
DeepSeek, Zhipu curb closed prices; export rules are a draft
For open weights to hold down closed-model prices, at least one strong vendor has to keep supplying models under permissive licenses at low prices.
Meta releases 816B Muse Spark Open; scores are self-reported
Meta releases Muse Spark Open, an 816-billion-parameter open-weight model, and model-service sellers with over US$20M in annual revenue must sign separately with Meta
TSMC posts a 66.4% gross margin; the release gives no prices
TSMC's third-quarter gross margin reaches 66.4%, 2 nm takes 8% of wafer revenue in its first quarter, and fourth-quarter revenue guidance is US$52.3B to US$54.3B
Anthropic cut back distillation; only one report has it
The Financial Times reports that Anthropic told investors distillation once made up 5% to 7% of revenue and was cut back sharply in the third quarter
Anthropic gives METR insider access; the account is its own
Anthropic gives six people from METR, a nonprofit that evaluates frontier-model risk, employee-like access; Anthropic can't veto METR's reports and can only require redactions in predefined categories
Xu counts 400-plus data-center pauses; it's his own tally
Investor Kevin Xu says the US data-center moratoriums he tracks now number more than 400, most of them pauses that expire within a year
Super Micro names Mark Long CFO; the filing gives no reason
Super Micro changes CFO: David Weigand steps down and former Western Digital CFO Mark Long takes over immediately, with no reason given in the filing
SAFARI pushes HBM2 power 23.6% past JEDEC; HBM3E is untested
ETH Zürich and NYU's SAFARI group test 36 HBM2 chips and find an access pattern that draws 23.6% more power than the highest-power measurement loop in the JEDEC memory standard, averaged across the 36 chips.
PyTorch 2.14.1 fixes an NVFP4 bug; scope isn't quantified
PyTorch 2.14.1 moves to CUDA 13.2.2 and fixes a cuBLAS bug that could ignore the tensor-wide scaling factor in NVFP4 matrix multiplications.
OpenAI's gate fails weak targets; utilities weren't tested
"How long will this security gate hold?" has no single answer; it depends on who is being defended against.
Four vendors show agent CPU racks; none has a public buyer
Our recheck of the CPU racks for running AI agents: four vendors announced or showed one at the same OCP summit, and the only public buyer, Meta, bought CPU capacity from AWS rather than a rack
Tunguz finds a 3.8x cost-per-task gap; it's his own math
Venture investor Tomasz Tunguz works out that five frontier models within five intelligence points of each other differ 3.8x in cost per task, with Gemini 4 Argon the cheapest
Jury finds Arm breached Qualcomm deal; judge hasn't ruled
A Delaware jury finds Arm breached its license agreement with Qualcomm and interfered with Qualcomm's Meta business, awarding only US$2; whether Qualcomm can stop paying royalties is still for the judge
Third Moment runs 10,000 agents at once; it's self-reported
OpenAI customer story: the quant startup Third Moment self-reports running 10,000 agents at once for 36 to 48 hours each, with an outer program deciding whether to retry, terminate or escalate to a human
ASML says planned supply falls short for two years
ASML's 6-K filing with the US Securities and Exchange Commission says its currently planned capacity will not fully meet customer demand in either 2027 or 2028. ASML also says it is studying adding capacity beyond the 30% expansion it announced earlier.
Marvell samples 224G copper chips; no customer is named
Marvell starts sampling two 224G-per-lane scale-up copper connectivity chips: a 3.2T retimer and a second-generation active-copper-cable DSP. The DSP is built on 3 nm and draws 12.8 W.
Databricks speeds a fine-tune about 6x; it's one measurement
A Databricks engineering post says a Qwen 3.8 27B fine-tuning run went from 7 hours 44 minutes to 1 hour 17 minutes, about 6x faster, even though GPU utilization already read 98%.
White House accord's gap is audit scope
The voluntary AI accord signed at the White House on September 29 does check whether controls operate as designed; the real gap is the audit's scope.
Anthropic replaces Git; merge figures are self-reported
Anthropic says daily code merges rose more than tenfold in four months to over 30,000, and version control and review tools it built itself have replaced Git
Google opens Gemini 4 Argon at US$2 per million input tokens
Google opens Gemini 4 Argon's paid API at an introductory US$2 per million input tokens and US$10 per million output tokens, with prices doubling after November 30
NVIDIA open-sources Run:ai; only one license is stated
NVIDIA says it is open-sourcing all of Run:ai, the GPU scheduling software it bought in 2024, with no paid subscription, and the GPU cloud Lambda says it will integrate it
Marvell, Microsoft, Oracle start an HSM spec; no text exists
Marvell, Microsoft and Oracle formed a working group at the open-hardware group OCP to write an open spec for a hardware security module built into the server, with a first draft targeted around this week's OCP summit.
Kundojjala puts ASML's fill rate at 87.3%; it's his own math
Independent semiconductor analyst Sravan Kundojjala says ASML's trailing four-quarter order fill rate fell from 126.8% in the second quarter of 2025 to 87.3% in the fourth, because bookings came in faster than they were recognized, not because demand weakened.
CASCI posts first drop in 12 months; its method isn't public
Tim Culpan, a former Bloomberg columnist who founded the publication Culpium, says CASCI, the index Culpan built to track revenue across Taiwan's AI compute supply chain, fell in September for the first time in 12 months because of capacity constraints.
Deep dive maps model hold-backs; planned dates aren't public
The ways companies hold back frontier models fit in one table: one axis is which layer is held back, the other is how release happens, and release comes in only two forms, on a later date or now but only to a vetted list.
AMD, NVIDIA show CPU-only agent racks; neither gives figures
On the same day, AMD launched and NVIDIA published the design of a rack that holds only CPUs, built for AI agents to carry out tasks
Two outside tests split on Muse Wonder; versions may differ
On Muse Wonder's launch day, two outside tests of Meta's new model gave readings that point different ways
Microsoft puts Maia 300 at 7x Maia 200; figures are its own
Microsoft published rack specs for Maia 300, its in-house AI chip: 96 chips in a 167-kilowatt rack, with peak rack throughput it puts at 7 times the previous Maia 200.
Nanya posts an 85.7% gross margin; only relays report it
Taiwanese DRAM maker Nanya Technology posted an 85.7% gross margin in the third quarter, according to one reporter and one analyst relaying its results. The analyst says higher prices alone drove growth, and bit shipments fell from the previous quarter.
OCP spec bars grid-fault dropouts; it doesn't bind utilities
(spec finalized September 17, made public at the summit October 12) A data-center power-grid spec whose original text lists Google, Meta, Microsoft and NVIDIA says data centers must not disconnect during grid faults within its defined range.
Whether export controls work is three questions
"Are export controls working?" is at least three questions: did they widen the compute gap, did they speed up China's self-sufficiency, and how much did they leak? The three use different yardsticks and can't refute each other.
Coinbase says its Qwen beats frontier AI; no results are out
Coinbase's CEO says Qwen3.5-9B, a 9-billion-parameter open model fine-tuned on Coinbase's own data, beats frontier models at fraud prevention
Baker says B300 prices rose 38%; he holds related positions
Investor Gavin Baker, citing Silicon Data's "token index," says B300 prices are up 38% and B200 prices up 10% since May 31; we haven't confirmed whether the index measures rental prices or what its base is.
Arm's economist admits five v10 licenses; record is pending
On day four of Qualcomm's antitrust trial against Arm in Delaware, according to notes from someone watching in court, Arm's economist conceded under cross-examination that only five architecture licenses for the latest v10 architecture exist worldwide.
Coding-plan 'subsidy' ties up compute; it rests on estimates
The "subsidy" on AI coding subscriptions looks mainly like a problem of tying up compute, not an accounting loss.
Fireworks revokes misused credentials; probe is still open
Lin Qiao, CEO of inference platform Fireworks AI, discloses misuse of internal credentials; the company has revoked them and notified affected customers
Kundojjala ties four toolmakers to Terafab; one post has it
Semiconductor analyst Sravan Kundojjala says the four big chip-equipment makers, ASML, Applied Materials, Lam Research and KLA, are all involved in Terafab, the chip fab Musk's companies plan to build (a single post, unverified); the Texas governor's office announced on August 6 that Terafab's first phase is a capital investment of more than US$16.8 billion.
Qualcomm says AI firms want 100B phone models; it names none
Qualcomm CEO Cristiano Amon says some frontier AI companies want phones that can run a model of at least 100 billion parameters, more or less all the time, by 2028.
Marvell's warrants cut its revenue; size isn't disclosed yet
When a chipmaker hands equity to a big customer in exchange for orders, the accounting rules, not the strike price, decide what it costs: it is booked as a cut to the chipmaker's revenue.
Anthropic takes evaluations offline; agencies aren't named
Anthropic lists four kinds of overreach by its models on real websites and takes all internal evaluations offline until it confirms its safeguards work
Micron's 10-K details take-or-pay deals and over-order risk
Micron's annual report spells out pay-whether-or-not-you-take contracts with price floors and ceilings, and in the same filing warns that customers may over-order and pile up inventory
Microsoft builds Decision-1 on Qwen; it didn't say why
Microsoft's new small decision model, Decision-1, is built on Alibaba's open-weight Qwen3.5-9B
OpenAI rates misalignment its top risk; it's self-assessed
OpenAI publishes its quarterly internal-use risk report under California law and rates the main risk as the model drifting from what was intended (misalignment), not staff misuse
Wistron posts record NT$1.15T quarter; only one relay has it
Taiwanese contract server maker Wistron's third-quarter revenue hit a record NT$1.15T, up 102.4% year on year; the same post says three Taiwanese power-supply makers, Lite-On, Delta and AcBel, are sharply raising capital spending this year.
vLLM says Vera Rubin tops GB200 7.8x; it's self-reported
vLLM, the open-source inference software, announces support for NVIDIA's next-generation Vera Rubin, and reports more than 7.8x GB200's throughput on one agent workload.
Deep dive called optics late; OCP's deadline hasn't arrived
The switch from copper to optics now has a date you can check: AMD officially places its next-generation MI500 in 2027, and its launch event said that generation "supports both copper and optics"; NVIDIA's matching point is in 2028.
One more step: confirm by email
A confirmation email is on its way to the address just entered; clicking the link inside completes the subscription. Not in the inbox? Check spam or Promotions. Wrong address? Enter it again
Subscribe free — first issue lands tomorrow morning
Just an email address, unsubscribe anytime. Subscribing finishes with one click on the link in the confirmation email; if it is not in the inbox, check spam or Promotions.