AI infrastructure news, week of 9 November 2026: Jensen Huang leaves NVIDIA, neoclouds ask customers to pay up front

Jensen Huang resigns from NVIDIA, CoreWeave and Nebius lean on prepayments, Anthropic adds Microsoft compute, Cerebras signs Microsoft. What it means in Europe.

Older NVIDIA office building with a dark stone facade, glass stairwell and green NVIDIA logo, seen through trees from a car park
Photo: Nvidia HQ by Kevin McCarthy, Wikimedia Commons, CC BY 2.0

Five stories from the week of 9 to 15 November 2026 that matter if you rent or buy GPU servers in Europe. Each one rests on a company filing, a results release or an earnings call. Sources and dates are at the end.

This week in short

  • Jensen Huang resigned as NVIDIA's president, CEO and board member on 12 November. CFO Colette Kress is interim CEO, and the board says customer commitments and the roadmap continue.
  • CoreWeave expects about $9 billion in customer prepayments and deposits over the next three months and raised its 2026 capex range to $37 to $39 billion.
  • Nebius set preliminary 2027 capex at $40 to $45 billion and said delivery at several colocation sites has moved into the first half of 2027.
  • Anthropic's amended IPO filing raised its Microsoft commitment to between $47.2 and $49.2 billion. Its largest Lambda agreement is now terminable.
  • Cerebras said on its earnings call that it signed a cloud deal with Microsoft, its third hyperscaler customer.

Jensen Huang resigns from NVIDIA

In a release filed with the US SEC on 12 November, NVIDIA's board said Jensen Huang "has resigned as President, Chief Executive Officer, and a member of the Board, effective immediately". Colette Kress, the CFO, is interim CEO and stays CFO until a permanent successor is found. Harvey C. Jones, a former Synopsys CEO, became NVIDIA's first Chair of the Board. The board is considering internal and external candidates. The release gives no reason. Press reports link the exit to an independent board investigation opened on 8 November into Huang's relationship with a former executive of the Singapore company Megaspeed. NVIDIA reports its quarterly results on 17 November, as scheduled.

The board's message to customers is that "operations, customer commitments and product roadmap" continue without interruption. South Korea's science ministry said NVIDIA reaffirmed its plan to supply 260,000 GPUs, TechCrunch reported. Our view: nothing changes for a server ordered this quarter, because allocation runs through OEMs and distributors under existing contracts. The open question is export policy. Reuters found more than $3 billion of Chinese AI projects built on restricted NVIDIA chips, and the change at the top comes while US scrutiny of chip diversion is rising. If you buy in Europe, expect more end-user questions, not fewer.

CoreWeave: prepayments now fund the build-out

CoreWeave reported third-quarter revenue of $3.66 billion on 10 November, up from $1.37 billion a year earlier, with a net loss of $709 million and a revenue backlog of about $132.5 billion. Active power passed 1.7 GW and contracted power 4.6 GW, of which more than 1.3 GW is outside the US. The company raised 2026 capex to $37 to $39 billion. CEO Michael Intrator said CoreWeave expects about $9 billion in customer prepayments and deposits over the next three months. It also set up a $2.6 billion facility for shorter contracts.

For a European tenant the signal is about terms. When the largest GPU clouds ask big customers to pay part of a multi-year contract in advance, smaller buyers should expect deposits and longer minimum terms to become normal. Our neocloud explainer covers how these providers finance their GPUs.

Nebius: $40 to $45 billion for 2027, and some capacity slips

Nebius published third-quarter results on 12 November: revenue of $980.8 million, up 571% from $146.1 million, and $7.3 billion spent on equipment in the quarter. In its shareholder letter it raised 2026 capex to about $27 billion, above the earlier $20 to $25 billion range, and gave a preliminary 2027 outlook of $12 to $13 billion in revenue and $40 to $45 billion in capex. It expects prepayments and financing against investment-grade contracts to cover 75% to 85% of 2027 capex.

The letter also says delivery dates at several colocation sites moved into the first half of 2027, and that part of the Microsoft capacity in New Jersey is now due in the second quarter of 2027. Nebius names Lappeenranta and a second site in Mäntsälä, both in Finland, as owned sites deploying capacity in 2027. Our reading: European capacity is coming, but power and buildings set the date, not the GPU order. If a provider quotes a start date, ask whether the hall is already energised.

Anthropic adds Microsoft compute, and a Lambda contract wobbles

In an amended IPO filing submitted on 10 November, Anthropic raised its minimum Microsoft commitment by $2.3 billion to $47.2 billion, with up to $2 billion more, for a total of $47.2 to $49.2 billion. Total compute commitments rose to about $900.9 billion. The same filing says either party may end Anthropic's August agreement with Lambda, worth about $34.9 billion, because Lambda missed a deadline to secure a facility lease. Neither side had sent notice, and Anthropic said it does not currently intend to terminate.

The amounts are far beyond any European buyer, but the Lambda clause is a useful template. Large tenants tie payment to capacity that is delivered and accepted, and keep an exit when a site is late. Smaller contracts can include the same protection.

Cerebras signs Microsoft as a cloud customer

Cerebras reported record third-quarter revenue of $210.8 million on 12 November and raised its 2026 core revenue guidance to $940 to $960 million. On the earnings call, CEO Andrew Feldman said Cerebras had signed a cloud deal with Microsoft that day. Microsoft joins AWS and Oracle as its third hyperscaler customer. No size was disclosed, and revenue is expected to start in the first half of 2027. Feldman also said early tranches of its OpenAI contract run behind schedule because of land, power and building work.

Microsoft adding a non-NVIDIA inference supplier shows that the largest buyers want more than one source. For most European teams running open models, NVIDIA hardware and its software stack remain the practical default. See our private LLM hosting page for what a single server can serve.

Questions

Does Huang's resignation affect RTX PRO 6000 supply? No source we found suggests so. NVIDIA's board says customer commitments continue, and the product line is unchanged.

Where does NEOXIS fit? We run one 8 × RTX PRO 6000 server in a data centre near Sofia, from $1.80 per GPU-hour or about $10,510 a month for the whole machine. It is rented into early 2027 and we accept reservations for the period after that. Details are on the pricing page.

Sources

SHARE