AI Markets: SpaceX borrows $40 billion, Anthropic eyes November

SpaceX is looking for $40 billion to buy Nvidia chips, and Anthropic is preparing its November listing. Google locks in nuclear power, Intel and TSMC jockey over Terafab, and AMD is short on capacity.

SpaceX wants to borrow $40 billion to buy Nvidia chips

According to the Financial Times, SpaceX is looking for $40 billion to pay for Nvidia chips. The deal would reportedly be led by Apollo. The news lands just as the stock posted its best week since June.

Key points:

  • The target amount: about $10 billion in bank loans and $30 billion in “investment grade” debt (FT, via Reuters and Investing.com).
  • Apollo would lead the deal and place the debt with investors; Pimco is part of the small group of lenders in talks.
  • The money would go solely toward buying Nvidia chips, according to the FT’s sources. Closing is expected in 2027.
  • SPCX shares slipped slightly after hours, following a 0.49% gain during the session, according to Yahoo Finance.

Why borrow instead of dipping into cash?

SpaceX chose to use Nvidia chips exclusively in its Colossus compute centers, which run Grok and rent out capacity to customers like Anthropic and Google. And those chips are very expensive, with the bill climbing with each new site.

The debt is rated BBB, the second-lowest rung of the “investment grade” category. At that level, insurers and pension funds are allowed to buy it. SpaceX had already sold $25 billion of this type of bond less than two weeks after its June IPO (Benzinga).

A reminder of context already covered here: SpaceX’s AI investments are hefty. The group reported $18.4 billion in capital expenditures in the second quarter. This new borrowing would add to that trajectory.

What it means for European readers

For a shareholder, there are two possible readings, and neither is settled. On one hand, easy access to credit shows that lenders trust the model. On the other, more debt means more interest to pay before AI revenue covers the costs.

For Nvidia, it’s one more order in an already packed backlog: a customer is financing its purchases through debt markets, not just through its profits. That’s the famous debate over AI’s “circular financing.”

SpaceX (SPCX) isn’t eligible for the PEA: in Europe, you can access it through a regular brokerage account or through funds that track the Nasdaq-100. This is factual information and not advice. The project remains a deal under discussion, unsigned as of today.

Sources:

Anthropic: IPO slips to November as two big customers pull back

Anthropic’s IPO is taking shape for mid-November. But at the same time, Meta and Microsoft are cutting back internal use of Claude. The timing is bad, since the prospectus already flags a heavy dependence on a handful of customers.

Key points:

  • Anthropic is reportedly targeting a listing as early as mid-November, with the investor roadshow kicking off the week of November 9 (financial press, including PYMNTS).
  • The hoped-for valuation would reach up to $2 trillion, versus $965 billion in the May round (not confirmed by the company).
  • According to The Information, Meta has halved its internal Claude Code users, and Microsoft has cut its budgets.
  • Michael Burry compares the $965 billion valuation to the price of 78 profitable S&P 500 companies.

What’s changed since our last edition

We had already covered the draft prospectus: strong growth, big losses, compute commitments topping $500 billion. What’s new is the timeline and the signal from customers.

According to The Information, Meta went from about 60,000 to about 30,000 Claude Code users. Yet it spent more than $105 million on it in a single 28-day period. At Microsoft, projected internal spending on Claude fell by more than a third from an annualized peak of over $1 billion. Both groups are pushing their teams toward their own tools.

Why it matters for a listing

The prospectus states that about a quarter of revenue comes from two customers. When customers that size reallocate, the market notices. These cuts are partial, and the amounts being spent remain high.

Then there’s the question of price. Burry, known for his bearish bets, sees the valuation as outsized. On the flip side, supporters point to revenue growth: $11.5 billion in the second quarter, according to the leaked prospectus. Both readings are opinions, not facts.

For a European saver, access will depend on the listing venue and on brokers. A U.S. listing wouldn’t be eligible for the PEA. The public S-1 hasn’t come out yet as far as we know: it must be published at least 15 days before the investor roadshow begins.

Sources:

Google signs 3.59 gigawatts with Constellation: nuclear power in the service of AI

Google is locking in a huge reserve of electricity for its data centers. The group is signing with Constellation Energy, the largest nuclear operator in the U.S. Constellation shares jumped more than 12% on the announcement.

Key points:

  • A 20-year contract for 890 megawatts of additional nuclear capacity on the PJM grid (Illinois, Pennsylvania, New Jersey).
  • A second, 15-year contract for 2,700 megawatts, for a total of 3.59 gigawatts.
  • Constellation will invest more than $4.3 billion; the price Google is paying isn’t public.
  • Constellation (CEG) gained 12.25% to close at $300.40, according to the Motley Fool.

How it works, in practice

These aren’t new reactors. Constellation will upgrade 11 existing nuclear units to get more output from them. The first power boost is expected in 2028, according to the two companies’ press release.

The 2,700 megawatts in the second contract come from Constellation’s existing fleet on PJM. Google doesn’t get a direct cable: the electricity flows through the grid, and the contract sets what Google buys.

In return, Constellation is picking Google Cloud and Gemini Enterprise in a five-year alliance, to better run its plants and its grid.

The bridge between AI and the stock market

Electricity has become AI’s bottleneck: having chips isn’t enough, you also need to be able to plug them in. Small modular reactors, for their part, have no target for coming online before 2028 at the earliest (Motley Fool). Upgrading existing plants is therefore faster.

This kind of contract turns a power producer into an “AI” stock in the eyes of the markets. The 12% jump proves it. It says nothing about what comes next: the share price already bakes in a lot of expectations.

On the European side, Constellation is a U.S. company, so it isn’t eligible for the PEA. The main takeaway is sector-wide: data center power consumption now weighs on electric utilities and grid operators. This article is for information only and is not advice.

Sources:

Terafab: Intel stays in Musk’s project as TSMC moves closer

Musk’s chip factory project is turning into a three-way game. Intel says it’s staying on, while TSMC is discussing a role. Markets have already reacted, especially on Intel’s side.

Key points:

  • Intel CEO Lip-Bu Tan says Intel “will continue to work with Elon Musk” on Terafab (Bloomberg, via Yahoo Finance).
  • Intel has been a development partner since April 2026; its 14A technology has been cited for the project.
  • Musk confirmed talks with TSMC about a role in this $16.8 billion project in Texas.
  • Nothing is signed, and the structure remains open.

What’s changed since our last edition

We had reported that TSMC was exploring a role in Terafab, according to a Bloomberg column. Musk has since confirmed the talks: “just discussions, but something could come of it.”

Today’s news comes from Intel. Its chief responded ahead of an event in Tokyo marking the company’s 50 years in Japan. The message: Intel doesn’t feel pushed out. Even so, Intel had fallen about 2% on Monday after the TSMC talks were confirmed (TIKR).

The scenarios on the table

According to press reports, TSMC could own and operate the factory, while SpaceX and Terafab would invest and guarantee chip purchases. Another option would give SpaceX the majority, with a smaller stake for TSMC. These are options, not decisions.

The project targets two fabs: one for the chips in Tesla cars and robots, the other for AI compute centers, including in space. The stated goal is one terawatt of compute capacity per year.

For the chip supply chain, the stakes are simple: Intel wants to prove it can manufacture for outside customers, and TSMC remains the benchmark. The two may coexist, or not.

On the European side, neither Intel nor TSMC can be held directly through the PEA. A UCITS-type semiconductor fund is the usual route. Information, not advice.

Sources:

AMD: demand outstrips supply, capacity to rise “a lot” in 2027

AMD’s CEO says demand for AI chips remains stronger than what the industry can produce. AMD therefore plans to significantly increase its supply in 2027. The company is also reserving capacity for the years after that.

Key points:

  • Lisa Su: “Demand exceeds supply. We plan to increase supply substantially in 2027, but clearly more will be needed.”
  • She sees “very, very high demand for the next several years” and is planning three to five years ahead.
  • AMD is working with Foxconn, TSMC and its memory suppliers to secure CPUs, GPUs, wafers and memory.
  • According to Bloomberg, Su is set to meet the head of Samsung’s chip division, as the memory shortage persists.

A bottleneck shifting to memory

Chips don’t sell on their own: they need memory. High-bandwidth memory (HBM) is in short supply, and manufacturers reserve it first for AI customers. Su has said so publicly: memory is one of the constraints.

According to trade media, the EPYC “Venice” server processors are reportedly already sold out through 2027. AMD has yet to confirm this.

Key takeaways for investors

A manufacturer saying “we’re short on capacity” signals strong demand, but also costs: reserving factory capacity three to five years ahead ties up money. AMD stock has climbed sharply in 2026, so the margin for error is thinner if demand slows.

It’s also a signal for the entire supply chain. TSMC makes the chips, Samsung and SK Hynix supply the memory, and Nvidia, for its part, holds on to its own allocations. We had already covered AMD’s contract with Microsoft and its targeted trajectory of $70 billion in 2027: Su’s statement is the logical next step.

For a European saver, AMD is a U.S. company and isn’t eligible for the PEA. This article is for information only.

Sources:

Edition prepared by our monitoring agent (data and sources as of Wednesday, October 7, 2026), reviewed in line with our editorial policy. General information, not investment advice: investing carries a risk of capital loss.