AI Industry Daily Briefing — November 7, 2026

The Executive Read

Thursday’s stock market treated AI infrastructure as a bet that could be repriced, and Friday’s news kept that theme going. CNBC reported that the Nasdaq fell 5.5% on Thursday, its steepest drop since April 2025, with Nvidia down more than 9% and AMD and Micron down by double digits. CNBC attributed the fall to a mix of causes, including worries about data center construction after the midterms. On Friday, Bloomberg reported that OpenAI has raised $11 billion of a larger round at a $1.25 trillion valuation, below the $1.4 trillion it was reportedly seeking in September. TechCrunch, relaying Bloomberg, reported that buyers of Anthropic’s planned IPO shares are discussing valuations as low as $1.1 trillion. Ramp, a corporate card company, published data showing U.S. business spending on AI model usage was roughly flat across August and September. Meanwhile the physical build continues: Google took over a Wyoming site and put its commitment there at $17 billion. And capability claims are rising, not falling. METR, an outside evaluator, said its staff believe Anthropic’s models are improving about twice as fast as before, though Anthropic withheld most of the evidence METR found most useful. So the story is a gap. Money is getting more careful while the technology, by some accounts, is speeding up. We did not find a single cause for Thursday’s fall, and CNBC lists several.

One correction of omission: Friday’s edition did not cover Thursday’s selloff. This one does.

Top AI Headlines

Tech Stocks Drop Sharply as Data Center Politics Meet a Weak Chip Forecast

What happened. CNBC reported that the Nasdaq fell 5.5% on Thursday, November 5, its worst day since April 2025. Nvidia and Qualcomm each fell more than 9%, and Nvidia lost nearly $450 billion in market value, according to CNBC. AMD, Micron and Applied Materials each fell by double digits. In a separate CNBC piece, the Thursday declines are listed as Nvidia 9.3%, AMD 17.3%, Micron 17.5%, Broadcom 10.4% and Microsoft 3.6%. CNBC said Qualcomm reported Wednesday and gave a weak forecast for the current quarter. Its chief executive, Cristiano Amon, said the company has “a supply issue,” not a demand issue. CNBC listed several causes: fears that the AI building boom slows after the midterms, rising interest rates, heavy capital spending, a global memory shortage that raises component prices, and voters rejecting local data center measures in 40 of 41 votes.

Why it matters. The selloff mixes real events with guesswork. Qualcomm’s forecast and the election results are facts. How much each contributed to the drop is not known. Barclays’ Dominique Toublan told CNBC the pushback is “bipartisan but local,” and that “scarce power favors the biggest balance sheets.” Dan Ives of Yorkville Ives & Co. now sees Big Tech capital spending growth peaking in 2027, earlier than he had thought. That is one analyst’s view. CNBC said tech stocks showed signs of recovery by Friday morning, and gave no Friday closing figures.

Business implication. Companies that fund data centers with debt or small balance sheets are the most exposed to a shift in sentiment. Customers should expect suppliers to push for longer contracts. This is informational, not investment advice.

Sources: CNBC, “Qualcomm earnings, chip stocks, election” (November 5); CNBC, AI data centers, midterms and the selloff (November 6).

OpenAI Reportedly Raises $11 Billion at $1.25 Trillion; Anthropic IPO Pricing Reportedly Softens

What happened. Bloomberg reported on Friday, citing unnamed people familiar with the matter, that OpenAI has completed an $11 billion portion of a funding round at a $1.25 trillion valuation. According to TechCrunch’s account of the Bloomberg story, the round has not closed, and one source put the eventual total at about $20 billion within weeks. In late September, Bloomberg reported OpenAI was seeking $30 billion at $1.4 trillion. TechCrunch says no investors were named, and that OpenAI had not responded. Separately, TechCrunch cites a Bloomberg report from Friday that potential buyers of Anthropic’s IPO shares are considering a valuation as low as $1.1 trillion, against roughly $2 trillion that investors were reportedly expecting about a month earlier, per CNBC. Anthropic’s registration statement, filed October 26, left the share count and price blank.

Why it matters. Both numbers are reporting, not announcements. We could not open Bloomberg’s pages (the site refuses automated access) and the CNBC page TechCrunch cites for the $2 trillion figure returned an error for us. Neither company has confirmed a valuation. Yesterday’s edition noted Greg Brockman saying OpenAI could be ready for an IPO in 2027. The new reporting suggests investors are paying less than the sellers hoped.

Business implication. A lower valuation means each dollar of new funding gives up more of the company. Both labs also have pressure on prices: TechCrunch says OpenAI cut API prices in half for two GPT-6 models in September, and Anthropic cut prices for two of its models. TechCrunch says a company filing put the cost of Anthropic’s cuts at about $600 million of revenue growth. We did not open that filing.

Sources: Bloomberg, OpenAI completes $11 billion portion (November 6), reporting is Bloomberg’s, paywalled, not opened by us; Bloomberg, Anthropic IPO valuation expectations (November 6), reporting is Bloomberg’s, not opened by us; TechCrunch, OpenAI raise (November 6); Anthropic, S-1 announcement (October 26).

Ramp: Business AI Spending Has Stalled, Though Usage Keeps Rising

What happened. Ramp’s AI Index, a measure built from corporate card and bill-payment data from its customers, says token spend (what businesses pay for model usage) was roughly flat for two straight months, August and September, while token volumes kept rising. Ramp’s economist Ara Kharazian wrote the October 14 update. It found the top 1% of firms spent $9,568 per employee per month in September, down 3.1% from the July high of $9,873. About 80% of OpenAI and Anthropic spending comes from that top 1%. Ramp says 43.8% of U.S. businesses on its platform paid Anthropic in September and 40.6% paid OpenAI, with overall AI adoption at 55.9%. TechCrunch reports a later reading: spend rose 8.9% in the month to October 23.

Why it matters. Ramp attributes the flat line to falling frontier model prices and efficiency gains, which lower the blended price per million tokens, rather than to lower use. It also says new-customer growth has slowed, and compares the market to one nearing the top of an S-curve. It adds that customers hold the stronger bargaining position and that contracts are “not particularly restrictive.” Ramp’s sample is its own customers, and the top-1% estimates are volatile because the group is small, as Ramp notes. The October 23 figure comes from TechCrunch; we did not see it in Ramp’s October post.

Business implication. Buyers who negotiate now have leverage. Labs may sell more usage without gaining more revenue.

Sources: Ramp, AI Index, “Cracks in the AI Thesis” Part 3 (October 14); TechCrunch, “Ramp data shows AI spending stalling just as Anthropic nears its IPO” (November 6).

Google Takes Over Crusoe’s Wyoming Site, With a 702-Megawatt Fuel Cell Plan

What happened. Google announced on November 5 a “$17 billion investment” in its first Wyoming data center, in Laramie County near Cheyenne. Power comes from a 702-megawatt natural gas fuel cell project with carbon capture and storage, contracted with the utility AEP. Google says the fuel cells are made by Bloom Energy, and that other power comes from gas generators run by Tallgrass Energy and Black Hills Energy. Captured carbon dioxide would go to Tallgrass’s nearby storage hub. Google says it will pay for 100% of the power it uses, and that the building will be air-cooled to save water. A day later, Crusoe, the Denver developer that had planned the site, announced that Google has acquired the site, completing Crusoe’s pre-development work. Crusoe did not give a price. TechCrunch adds that the project was planned to scale to 10 gigawatts, that Bloomberg reported in June that Google had raised concerns about cost and timeline, and that Crusoe was valued at $30.9 billion in September.

Why it matters. 702 megawatts is on the scale of a large power plant. The sale shows how hard it is for a developer to deliver at this scale. The Google release leans heavily on the same points Meta’s president listed this week as keys to local approval: paying its own power bill, local property taxes (Google expects “hundreds of millions of dollars”), and a community council. Crusoe’s CEO said the company has “more than 6 GW contracted and over 40 GW in our pipeline.”

Business implication. The largest buyers can take over projects that smaller developers cannot finish. The cost of power and local consent is becoming a larger share of what a data center costs.

Sources: Google, Wyoming data center announcement (November 5); Crusoe, Wyoming campus update (November 6); TechCrunch, “Crusoe sells troubled Wyoming data center project to Google” (November 6); Bloomberg, June 11 report on Crusoe and Google, linked by TechCrunch, not opened by us.

Hundreds of Exposed Copies of ARTEX, an AI Hacking Framework, Found Online

What happened. Censys, an internet scanning firm, published research on ARTEX, an open-source framework in which an AI model drives penetration testing, the authorized practice of probing a network for weaknesses. Censys says it was published on GitHub on July 26 by the developer of ScopeSentry. The author later said it would go closed-source and the repository has been taken down, but archived copies and rebrands circulate. As of early November, Censys counted 1,465 hosts running an exposed ARTEX console. About 650 sit in two identical hosting blocks, leaving roughly 815 distinct-looking deployments. By country: 726 in the United States, 387 in Hong Kong and 177 in China. Censys says 155 of the hosts also run other known attack tools, and 32 expose open directories holding implants, the small programs attackers leave behind to keep access. Censys notes some hosts are likely run by authorized testers, and that ARTEX’s developer denies involvement in the attacks.

Why it matters. Microsoft’s October 21 report says Storm-0940, a China-aligned group active since 2021, used ARTEX in the ConnectWise cloud compromise, attempting sign-ins to more than 2,250 customer tenants. In some environments the group reached privileged cloud access in under three hours. Microsoft says most targeting failed and its own systems were not compromised. CrowdStrike separately linked ARTEX to a campaign against South Korean finance firms, according to Censys.

Business implication. Attack tools that run themselves lower the skill needed to attack. Microsoft and Censys both stress ordinary weaknesses: weak passwords and missing multi-factor authentication. Those still decide most outcomes.

Sources: Censys, ARTEX research; Microsoft Security Blog, Storm-0940 and ARTEX (October 21); TechCrunch, “Hundreds of Artex AI hacking tools found publicly exposed” (November 6).

Model and Product Updates

Thinking Machines releases Inkling 2 Small, with open weights promised. The company describes a mixture-of-experts model (one that activates only part of its parameters per query) with 594 billion total parameters and 17 billion active, accepting images. It is available now on the company’s Tinker platform with closed weights. The company says it intends to release the full weights under an Apache 2.0 license “next week,” after external safety testing, and calls that timing conditional. All benchmark figures are the company’s own: 86.3% on SWEBench Pro v2 (Hard), 72.1% on DeepSWE v1.1, and 30.6 to 31 on the Artificial Analysis Intelligence Index. The company’s tables show some rivals scoring higher on specific tests, and it says the model still trails closed models such as GPT-6.1 Sol on its alignment evaluations. No per-token price is given on the page. Thinking Machines Lab, Inkling 2 Small

Anthropic describes “dreaming” for agent memory. An Anthropic engineering post by Lance Martin, published November 6, argues that agents should log work in transcripts and review them offline to fix errors and build organized notes. In the company’s tests on a roughly 350,000-file corpus, accuracy from a generated wiki alone rose from 13% to 48% over 128 such sessions, and wiki plus search reached 61% against 53% for search alone. These are Anthropic’s own experiments. Anthropic Engineering, “The components of agent memory”

METR: Anthropic’s progress is roughly twice as fast, with caveats. METR, a nonprofit that evaluates AI systems, ran a high-access assessment from August 12 to October 29. Its staff believe that as of October, Anthropic’s models improve on well-specified tasks at about twice the rate of March 2024 to October 2025, and they “would not be surprised” by 1.3x or 3x. Benchmark scores grew at about 1.9x the prior rate, with a 95% interval of 1.5x to 2.8x. Most staff guess about 80% of the acceleration comes from AI, a figure METR calls highly uncertain. Including all frontier developers, the benchmark estimate falls to about 1.4x. METR says Anthropic redacted most of the evidence METR found most informative, so readers must rely on METR’s judgment, and that Anthropic cancelled a compute verification component midway. It found no convincing bottleneck over the next 12 months. METR, assessment of AI R&D acceleration at Anthropic (November 5)

Meta adds Spaces to its Muse agent. TechCrunch reports Spaces are goal-based workspaces in the Muse sidebar that Muse builds from a chat description, such as a planner, tracker or project board, with outside apps connected. Meta’s newsroom post was not linked by TechCrunch; its closest source is Mark Zuckerberg’s X post. TechCrunch notes Zuckerberg said on an October earnings call that the agents had 10 million users in the U.S. and Canada. TechCrunch, Meta Spaces (November 6)

Regulation and Policy Watch

Justice Department forms a “Super Intelligence” task force. In a release dated November 5, the department announced the SI Task Force, to protect American leadership from misuse by foreign state actors, criminal organizations, and people or companies engaged in fraud. Associate Attorney General Stanley Woodward and Associate Deputy Attorney General Stephen Weiss lead it, working with SI Force chair Jay Clayton. Attorney General Todd Blanche said development “will continue to be powered by American innovators.” The release as we read it gives no budget, staffing or specific cases. U.S. Department of Justice, press release (November 5)

Meta pauses its scraping terms. Nieman Lab, which first reported the proposed terms on October 15, says Meta on November 5 removed the preview link and indefinitely paused changes that would have let it scrape websites of Meta Business account users. A Meta spokesperson said: “We’re pausing the changes as we continue to engage with stakeholders.” Nieman Lab says courts in Norway and Brazil issued preliminary injunctions, the UK Competition and Markets Authority opened an investigation, and Senators Adam Schiff and Marsha Blackburn wrote to Zuckerberg asking him to withdraw them. We have not seen those rulings or the CMA notice ourselves. The existing terms remain in effect. Nieman Lab (November 2026)

House Republicans ask Jensen Huang for a closed-door briefing. CNBC reports that Representatives Brian Mast, John Moolenaar and Rick Crawford, who lead the Foreign Affairs, China Select and Intelligence committees, asked Nvidia’s chief executive for a joint briefing on his interactions with Alice Huang, a Megaspeed executive, and on Nvidia’s protections against foreign espionage. CNBC says the New York Times reported U.S. intelligence found she was passing information to Beijing. We have not seen that reporting. Megaspeed did not respond to CNBC. Nvidia said its regional cloud customers receive ordinary business information and that it would be notified if the government considered anyone a threat. The proposed date is November 18, the day after Nvidia’s earnings. CNBC (November 5)

Bank of England: modest productivity gains from AI. According to The Register’s account of the bank’s November Monetary Policy Report, UK firms report AI has raised productivity by about 0.5% over three years, and AI-related activity accounts for just under a fifth of UK GDP growth this year. We did not find those passages in the portion of the report we could read, so the figures rest on The Register. Bank of England, Monetary Policy Report (November 2026); The Register (November 6)

Emerging Startup Radar

Future Secure AI raises US$80 million at a US$1.2 billion valuation. Startup Daily reports the Sydney-founded, now Austin-based company raised convertible notes from Square Peg, Yarra Capital Management, Macquarie Asset Management and Firetrail. It sells AI “co-workers,” named agents working under a human supervisor, to firms in banking, consulting, healthcare and law. Startup Daily cites the Australian Financial Review for annual recurring revenue (yearly subscription value) approaching US$100 million by the end of 2026 and a target ASX listing in late 2027. That is reporting, not a company filing. Startup Daily (November 6)

Axon invests $1.2 billion in drone maker Skydio. Axon’s November 5 release says it will invest about $1.2 billion in cash and stock at a $5.2 billion pre-money valuation, add roughly 20% to its fully diluted ownership, and become Skydio’s exclusive U.S. and Canadian public safety distributor. Two transactions, a primary investment and a purchase from existing shareholders, are expected to close in the fourth quarter and first quarter of 2027. DroneXL, relaying figures we did not check against the filing, reports Axon’s stock fell about 18% Friday after weak third-quarter results. Axon release on PR Newswire (November 5)

AI Infrastructure and Market Signals

Blocked projects. CNBC cites the International Data Center Authority, a think tank, as counting at least 46 AI data center projects across 20 states, with more than $170 billion in announced investment, blocked, withdrawn or stalled after community opposition between January 2024 and May 2026. Goldman Sachs estimates AI-related capital spending could reach about $7.6 trillion from 2026 through 2031, per CNBC. Both are estimates.

Supply, not demand. Qualcomm’s Amon told CNBC “We don’t have a demand issue; it’s a supply issue.” CNBC says its data center business grew strongly but brought in about $650 million in the quarter. A memory shortage is raising component and device prices, per CNBC.

Crusoe’s pipeline. Crusoe’s own release claims more than 6 gigawatts contracted and over 40 in its pipeline. A gigawatt is roughly one large nuclear reactor. These are the company’s figures.

Sources: CNBC, data centers, midterms and the selloff (November 6); CNBC (November 5); Crusoe (November 6).

Public Investment Watchlist

Informational only; this is not investment advice. Thursday declines are from CNBC.

• Nvidia (NVDA). Down 9.3% Thursday, nearly $450 billion in market value. Reports third-quarter earnings November 17. Congressional briefing request proposed for November 18.
• Qualcomm (QCOM). Down more than 9% Thursday after a weak current-quarter forecast, per CNBC. We did not read the earnings release itself.
• AMD, Micron, Broadcom. Down 17.3%, 17.5% and 10.4% Thursday.
• Alphabet (GOOGL). $17 billion Wyoming commitment announced; price for the site undisclosed. Alphabet fell 2% to 4% Thursday, CNBC reports for the group of Alphabet, Amazon, Meta and Microsoft.
• Axon (AXON). $1.2 billion Skydio investment, per its release; stock reaction reported by DroneXL, not verified.
• Anthropic and OpenAI (private). Valuations are reported, not confirmed. Anthropic’s S-1 has been filed but no price range is set.

Watchlist

• November 17: Nvidia’s third-quarter earnings.
• November 18: Proposed closed-door briefing for Jensen Huang with three House committees.
• Next week: Thinking Machines’ promised Apache 2.0 release of Inkling 2 Small weights, if safety testing finishes.
• Coming weeks: Whether OpenAI’s round closes near $20 billion, and whether Anthropic sets an IPO price range.
• November 20: Date of Meta’s paused scraping terms. Watch for a formal withdrawal or new date.
• Anthropic and METR: Whether Anthropic publishes any of the redacted evidence.

---

A note on today’s reporting: Bloomberg’s pages and OpenAI’s newsroom blocked our tools, so those items rest on TechCrunch’s or CNBC’s account of them, flagged where they appear. We left out aggregator-only items, including a claim that Google DeepMind found cryptographic advances against 1024-bit RSA, because we found only a crypto-news rewrite and no primary source.