The one-screen version: On November 12, Nvidia's board announced that Jensen Huang resigned as president, CEO and director — effective immediately — and installed CFO Colette Kress as interim chief while it searches for a permanent one. The man who personally brokered the supply deals your inference costs ride on is gone overnight. And here's the part that actually tells you something: in the same seven days, Nvidia is wiring roughly $10B into Anthropic's IPO, Broadcom agreed to lend Anthropic up to $42B, and the chip world kept de-risking its dependence on Nvidia silicon on schedule. The founder left; the machine didn't stop. Here's what each half asks of you.
1. Read this as key-person risk in your supply chain, not as a market omen#
The temptation is to trade the headline — "the AI trade is cracking." Don't. The 8-K is terse: Huang resigned effective immediately, Kress (CFO since 2013) is interim CEO, board member Harvey Jones is now the company's first-ever chair, and the search will weigh internal and external candidates. The company gives no reason. A lot of reporting links the exit to scrutiny of Huang's personal conduct and an internal investigation — we flagged that the board had opened one — and some coverage frames it closer to an ouster than a resignation. None of that is confirmed, so don't launder the allegations into fact.
What is fact is the operational exposure. Huang didn't just run Nvidia; he was the relationship. The allocation priority, the pricing, the roadmap promises that your GPU budget quietly depends on were brokered by a founder who is no longer in the building, and that relationship capital doesn't hand off cleanly to an interim chief and a board in transition.
What it changes for you: nothing today, everything in your planning assumptions. Stop treating your current GPU allocation, discount, or roadmap expectation as a fixed input. The person who could personally guarantee it is gone. Re-forecast on the assumption that any of those can move — not because they will, but because the one anchor that made them feel permanent just walked out.
2. The money didn't flinch — watch where it kept flowing#
Now put the resignation next to the other thing that happened this week, because the contrast is the whole briefing. While Nvidia lost its founder, the capital stacked around the model layer moved exactly on schedule. Seven firms — Nvidia and Google in for roughly $10B each, AMD around $2.5B — are putting a combined ~$26B into Anthropic's IPO at a reported $70–$80 range, a deal that with cornerstone money could raise up to ~$90B. Separately, Broadcom agreed to lend Anthropic up to $42B to finance leased chips.
Sit with the shape of that. The single most important person in AI hardware resigned, and the checks flowing into an AI model company didn't pause. That's not a coincidence of timing; it's the market telling you where it thinks the durable value sits. The chip is critical and the chipmaker just proved it's mortal at the top. The model — and the IPO wrapping it — kept its financing. We've argued for months that the model was never the moat; this week the capital said something adjacent and sharper: the chip vendor's org chart isn't the moat either.
What it changes for you: when you're deciding where to spend your own scarce capital — engineering, not just dollars — follow the money's revealed preference. It's financing the model-and-governance layer through open turmoil at the chip vendor. Build where the value is accruing, not on the single supplier whose continuity just got a lot less certain.
3. Chip diversification stopped being a hedge and became the base case#
Here's the tell that makes both halves cohere. The groundwork for not depending on Nvidia was already being poured before Huang left. Oracle is preparing to sell Google's TPUs in its cloud starting in 2027 — the first hyperscaler outside Google to offer them — and OpenAI is reportedly eyeing Google silicon directly. AMD is a $2.5B cornerstone in the Anthropic deal and sells the Instinct accelerators that increasingly show up in real inference stacks. The entire industry has spent 2026 quietly building a second and third path off Nvidia, and this week handed them the clearest possible reason to finish the job.
What it changes for you: treat multi-silicon as the default, not the contingency. You don't need to abandon Nvidia — its chips are still the best in most workloads. You need your inference layer abstracted so that moving a workload to a second path (AMD Instinct, a TPU via cloud, or a managed inference provider) is a configuration change, not a re-architecture. The same discipline we keep preaching for models — keep the boundary swappable, never marry one vendor — now applies with fresh urgency to the silicon underneath them.
The throughline: this week the AI economy ran a live test of its own fragility — pull out the single most important person in the hardware layer and see what breaks. The answer, so far, is: less than the headline implies. The capital kept flowing into the model layer, the chip-diversification plans kept advancing, and the machine kept running past an empty chair. The lesson for a founder isn't to panic about Nvidia; it's to build like the industry already is — assuming no single person, chip, or lab is load-bearing, because this week proved the market is pricing exactly that.
