AI Markets: Nvidia and the Anthropic IPO, Karp vs. the AI labs, Crusoe at $31 billion

Nvidia could put another $10 billion into Anthropic’s giant IPO, while Palantir’s CEO accuses AI labs of trying to dodge the courts. Elsewhere, Crusoe raises billions to build a Stargate showcase site, and humanoid robot announcements keep piling up, with varying degrees of nuance.

Anthropic IPO: Nvidia ready to add $10 billion, mid-October timeline holds

Two pieces of news made the rounds on September 17 and 18 about Anthropic’s giant IPO. One is genuinely new: Nvidia is reportedly discussing an additional check of up to $10 billion. The other mostly confirms what had been known since early September.

Key points:

Nvidia is reportedly in talks to become an “anchor” investor in Anthropic’s IPO, for up to $10 billion, according to a Reuters exclusive from September 11 picked up by Bloomberg and then by other outlets on September 17-18. Neither Nvidia nor Anthropic has confirmed the report.
That amount would come on top of the $10 billion Nvidia already committed to Anthropic in November 2025, alongside Microsoft’s $5 billion. Nvidia would then be playing three roles: indirect chip supplier, existing shareholder, and prospective IPO investor.
The “mid-October” timeline reported on September 17-18 isn’t new: it’s a rehash of a Reuters scoop from September 4.
Anthropic is still reportedly aiming to complete its listing before the US midterm elections in November 2026. The sources cited by Reuters themselves say that timeline could still shift.

The real news: Nvidia wearing three hats

Nvidia already put $10 billion into Anthropic in November 2025, alongside Microsoft’s $5 billion, at a valuation of about $350 billion — compared with the $2 trillion to $3 trillion being floated today for the IPO.

What the new report changes: Reuters said on September 11 that Nvidia is discussing an additional check of up to $10 billion, this time as an anchor investor in the IPO. An anchor investor commits in advance to buying a chunk of the shares at listing, which reassures other buyers before the first day of trading.

Several US commentators describe it as a closed financing loop: the same player funds a customer that, at the end of the chain, buys chips back from it — a structure worth watching before judging how solid the valuation really is. For now, the amounts remain information from anonymous sources.

The mid-October timeline: a confirmation, not a scoop

The timeline news is less new than it looks. Reuters had already written on September 4 that Anthropic’s investor roadshow would start in mid-October at the earliest, with a listing targeted just before the November midterms. The September 17-18 articles repeat that same date with no additional verifiable detail.

For readers who have followed the story since early September, it means one simple thing: the timeline hasn’t slipped a second time, it’s holding to the date already announced. Anthropic is still reportedly aiming for a public S-1 filing at the end of September, a $15 billion credit line under negotiation with a syndicate of 17 banks, and an investment grade credit rating before listing.

For a European retail investor: Anthropic isn’t listed yet, so no PEA or UCITS access for now. Once listed in the US as planned, the stock will stay outside the PEA anyway, accessible only through a regular brokerage account.

Sources:

Karp accuses Anthropic of trying to dodge the courts, Europe looks for its own sovereign Palantir

On September 17, 2026, Palantir’s CEO said on CNBC that the big AI labs want the sector “nationalized” to escape lawsuits. The same day, the fight over military AI sovereignty heated up in Europe, while an NHS contract remains stuck in the UK.

Key points:

Alex Karp (Palantir CEO) said on CNBC that AI labs want nationalization “because otherwise, all my clients are going to sue [them]” — Karp’s interpretation, not a quote from Anthropic.
Anthropic did not respond directly to Karp. Its CEO Dario Amodei has called for oversight by democratic governments and for gradual regulation, not for nationalizing companies.
In Europe, Thales is pushing alternatives billed as “sovereign” against Maven, Palantir’s military platform already deployed on NATO networks.
In the UK, Palantir just hired Tom Watson, former deputy leader of the Labour Party, while its £330 million contract with the NHS remains in limbo amid criticism over the sovereignty of patient data.

The controversy: who pays if AI causes harm?

Karp sees three camps in the AI safety debate: those who want to slow down (Amodei, Altman), the laissez-faire crowd (Nvidia, Meta), and himself, who champions civil and criminal liability as the “first line of defense.” His thesis: if the labs are pushing for government oversight, it’s to shield themselves from lawsuits by their customers, not out of concern for safety.

That’s an accusation, not an established fact. Amodei has indeed called for government oversight of AI — but describing a call for regulation as “nationalization” is Karp’s own shortcut.

For investors, this debate isn’t just rhetoric. If courts or lawmakers were to impose heavy legal liability on AI labs, it would change the risk calculation for the entire industry — an especially sensitive topic as Anthropic prepares its IPO (see elsewhere in this edition).

The European angle: Maven, Thales and the stalled NHS contract

Maven, Palantir’s military targeting platform, already equips NATO. Against it, Thales is betting on independence from US export controls (ITAR) as a selling point to European militaries — data sovereignty is becoming a purchasing criterion, not just a slogan.

In the UK, Palantir’s NHS contract illustrates the same tension: British lawmakers have recommended ending the £330 million contract, judging the risk of dependence on a US supplier too high. Hiring Tom Watson looks like a political lobbying move to save the contract.

PEA check: Palantir is listed on the Nasdaq, so it’s outside the PEA. Thales, on the other hand, is a French stock eligible for the PEA — but nothing here says whether the breakthrough of its “sovereign” military offering will translate into financial results. It’s neither a buy nor a sell signal, just a power struggle worth watching.

Sources:

Crusoe raises $3.9 billion and eyes the Nasdaq after building Stargate’s showcase site

This former bitcoin miner turned AI data center builder has seen its valuation explode in less than two years. It isn’t listed, but two of its investors are.

Key points:

On September 17, 2026, Crusoe announced a $3.9 billion funding round (Series F) valuing the company at $30.9 billion.
In October 2025, Crusoe was worth about $10 billion. At the end of 2024, $2.8 billion. Its valuation has grown more than tenfold in under two years.
Crusoe builds and runs the Abilene, Texas site: the showcase of OpenAI’s Stargate project, eventually 1.2 gigawatts of capacity, used by Oracle to run OpenAI’s models.
Nvidia and TPG, two publicly traded companies, are taking part in the round.

Who Crusoe is, and what this deal says

Founded in 2018 in Denver by Chase Lochmiller and Cully Cavness, Crusoe originally had nothing to do with AI. Its starting idea: capture the gas flared at oil wells, wasted for lack of a pipeline, to power bitcoin mining containers. A polluting waste product turned into useful electricity.

As demand for AI compute exploded, Crusoe changed lines of business. In March 2025, it sold its bitcoin business to NYDIG and went all in on AI data centers, with its own power generation. That pivot explains what followed: valued at $2.8 billion at the end of 2024, it’s now worth eleven times that.

The money raised funds its ongoing projects, including Abilene, and the development of “Spark,” modular data centers that can be moved by truck. According to Axios, Crusoe is also in talks about an IPO with Goldman Sachs, Morgan Stanley, JPMorgan and Bank of America. Nothing is settled: no date, no size, no target valuation.

What it says about the data center race, and for European investors

This round shows the outsized appetite for AI’s physical infrastructure: power plants, land, buildings, chips. Stargate, the $500 billion project backed by OpenAI, Oracle and SoftBank, is the most spectacular example. Crusoe is building a cornerstone of it.

For an individual with a PEA, Crusoe remains out of reach: it’s a private, unlisted company whose shares can’t be bought on the stock market. Two investors in this round are listed, however: Nvidia, whose chips power this kind of data center, and TPG, the asset manager that put money into the deal. Both are listed on the Nasdaq, so not eligible for the standard PEA, which is reserved for European stocks: for a French investor, they’re only accessible through a regular brokerage account, or indirectly through some UCITS trackers replicating US indexes.

One last caveat: a $30.9 billion valuation set by a handful of private investors has never been tested by a public market. It could move either way on the day Crusoe goes public, if it ever does.

Sources:

Starship slips a week, NASA hands SpaceX a completely different contract

Starship’s first orbital flight slides to September 28, with no public explanation. The same week, NASA awarded SpaceX a separate launch contract, for a science satellite that will lift off on a Falcon 9, no earlier than 2028.

Key points:

Starship Flight 14, announced for September 22, has been pushed back to September 28. SpaceX gave no official reason.
This flight is meant to be the first to put Starship into Earth orbit, after 13 suborbital tests, and to deploy 26 Starlink V3 satellites: SpaceX’s finance team calls it a “revenue-generating” flight, even though the satellites are the company’s own and not a paying third-party customer’s.
On September 17, NASA announced it had picked SpaceX to launch StarBurst, a small satellite for detecting gamma-ray bursts, but on a standard Falcon 9, no earlier than 2028: a contract unrelated to Starship.

Starship: a one-week slip, no surprise for the market

SpaceX announced the delay on its X account, saying only that liftoff remains “pending regulatory approval,” without detailing the technical cause of the slip.

The flight is set to take the ship around roughly six orbits at an altitude of 275 km, with a splashdown of the Super Heavy booster in the Gulf of Mexico and a splashdown of the upper stage in the Pacific, west of Chile. The 13 previous test flights never went beyond a suborbital trajectory.

For investors, this six-day delay is a non-event: SPCX stock kept climbing this week despite the announcement, just as it had already absorbed the earlier waves of post-IPO share unlocks without a jolt. An experimental rocket test slipping by a few days changes nothing about the thesis that has driven the stock since the summer: AI compute contracts and Starlink revenue, not a test flight schedule.

StarBurst: a NASA contract not to be confused with Starship

StarBurst is a small satellite designed to spot gamma-ray bursts emitted by neutron star mergers, complementing ground-based gravitational wave detectors. It’s part of NASA’s Astrophysics Pioneers program, dedicated to low-cost science missions.

NASA awarded the launch to SpaceX through a fixed-price contract under the VADR (Venture-Class Acquisition of Dedicated and Rideshare) framework contract, which lets the agency order launches over ten years, up to a combined ceiling of $1 billion across all providers. The exact value of this StarBurst contract was not disclosed. Liftoff is planned on a Falcon 9, in a Bandwagon rideshare configuration, from Cape Canaveral, no earlier than 2028.

For investors, this contract says one simple thing: NASA keeps entrusting its small science missions to Falcon 9, the proven rocket, while Starship stays in its test-flight slot. The two announcements landed the same week, but they tell the story of two different businesses at SpaceX: one routine and profitable, the other still experimental. As a reminder, SpaceX remains off-limits in a PEA (brokerage account only).

Sources:

Toyota promises 400,000 robots, UBTech opens a factory that turns one out every 10 minutes

Two announcements made the rounds on September 17-18, 2026: 400,000 robots promised at Toyota, and a Chinese UBTech factory claiming one robot every 10 minutes. After Unitree’s rollercoaster IPO and doubts about how productive these machines really are, here’s what holds up, and what to look at beyond the headline.

Key points:

Toyota (Nikkei, September 17, 2026) plans to invest about 1 trillion yen a year, starting in 2028, to equip 60 factories worldwide with 400,000 robots in total: 150,000 at Toyota and its group, 250,000 at its suppliers.
Toyota’s robot, nicknamed “Elly,” isn’t a bipedal humanoid: it rolls on a wheeled base, weighs about 50 kg, and learns its movements by imitating expert workers fitted with finger sensors.
UBTech (Hong Kong, ticker 9880) brought online on September 12 a factory in Liuzhou, China, targeting 10,000 industrial robots a year, with an adjoining automated warehouse able to store 112 robots in 65 m².
UBTech is still losing money despite its growth: a net loss of about 339 million yuan in the first half of 2026, on 1.27 billion yuan in revenue.

Toyota: the number is real, the horizon is 2028

The 400,000-robot figure does come from a September 17, 2026 article in the Japanese daily Nikkei, but the headlines picked up elsewhere cut out the context. The plan starts after 2028, with an annual investment of about 1 trillion yen, to overhaul 60 factories worldwide. Of the 400,000 robots targeted, 250,000 are expected at Toyota’s suppliers, not in the automaker’s own plants.

Today, the actual deployment amounts to a handful of units: seven Digit robots from Agility Robotics have been working since April 2026 at the Woodstock, Ontario plant, on lease. No Toyota press release confirms the 400,000 units: the information comes from Nikkei’s reporting, not from a figure stated by the automaker.

Toyota is listed in Tokyo and via an ADR in New York, outside the scope of the PEA: a European investor can access it through a regular brokerage account or a UCITS fund with exposure to Japan. Robots are only a tiny line item in a group that sells cars first and foremost.

UBTech: the factory is running, profitability remains to be proven

The Liuzhou factory really does exist: 14,000 m², 13.8 meters tall, started up on September 12 according to Global Times and eWeek. It builds the Walker S and Cruzr series, with help from UBTech’s own logistics robots.

No large customer contract has been announced to absorb this capacity of 10,000 robots a year. And UBTech’s financials are a reminder that building a robot every 10 minutes doesn’t mean selling it at a profit: the loss remains high relative to revenue, even though both are growing fast.

UBTech is listed in Hong Kong, also outside the PEA, accessible through a regular brokerage account or certain Chinese tech UCITS trackers — a young, volatile stock since its IPO in late 2023.

Sources:

Edition prepared by our monitoring agent (data and sources as of Friday, September 18, 2026), reviewed in line with our editorial policy. General information, not investment advice: investing carries a risk of capital loss.