Jensen Huang Was Never Your Chip Supplier. He Was Your Central Bank.
Nvidia posted a record $115.8B quarter five days after Jensen Huang left. The chips keep shipping; what's at risk is his discretionary financing and allocation, and your compute plan sits downstream of it.
OPC Community
OPC Community

Nvidia just posted a record quarter without him. What you were leaning on never showed up on the income statement anyway.
Jensen Huang had been gone from Nvidia for five days when the company reported $115.8 billion in quarterly revenue. That's up 103% from a year ago. Guidance for next quarter: $130 billion.
So much for the single point of failure. Or at least the one everybody was staring at.
A day later, Elon Musk told the DealBook Summit that pushing Huang out was like kicking Michael Jordan off your team. He said he'd hire Huang himself, and he signed a reinstatement petition that has now passed 200,000 names. Palmer Luckey needed fewer words: "China wants Jensen out, so I want him in." Satya Nadella called the whole thing "pretty tragic," and then he and Sam Altman both said they were confident in Nvidia without him.
Put those reactions beside the earnings number and they stop making sense. A company that doubles revenue under an interim CEO shouldn't need three of the most powerful people in tech lining up in the same week to vouch for it or grieve over it.
Unless what they're grieving was never the company.
The Jordan Analogy Proves the Opposite of What Musk Meant
Michael Jordan walked away from the Bulls in October 1993. Chicago had just won 57 games and a third straight title.
The next season, with no Jordan, they won 55.
Two fewer wins. The system was the same, the coach was the same, and Scottie Pippen was playing the best basketball of his life. What they lost was the second round of the playoffs, to the Knicks. The team's floor barely moved. Its ceiling was gone.
Nvidia's quarter told the same story. The fabs are booked. The Vera Rubin racks are ramping. Every order in that number was signed long before November 12. Colette Kress, the CFO now sitting in as interim CEO, ran a clean quarter and bumped next year's growth target from 70% to 75%. One man cleared out his office and nobody's shipment slipped.
If your fear was "Jensen leaves and the GPUs stop," relax. You were never depending on that.
You were depending on everything Huang did that no operating plan can do. Musk knows this better than anyone who shared that stage with him, because in 2016 Huang personally carried the first DGX-1 supercomputer into OpenAI's office and handed it to him.
No allocation process produced that box. One man decided who mattered next, and he did it eight years before the market came around.
What One Person Actually Decided
Forget the chip sales for a minute and look at what Huang's Nvidia did with its discretion over the last eighteen months.
It agreed to buy up to $6.3 billion of CoreWeave's unsold capacity through 2032, and that promise is what made CoreWeave's debt financeable. It signed a letter of intent to put up to $100 billion into OpenAI, money that comes right back as chip orders. Anthropic got a commitment of up to $10 billion. Lambda got Nvidia as a tenant, leasing back its own GPUs. Intel got a $5 billion stake.
Component vendors don't behave like this. Lenders of last resort do. Huang picked which neoclouds got scarce chips early, which labs got balance-sheet support, and which customers counted as strategic when supply ran short. Each of those calls was a bet on the future, placed by a founder with 33 years of standing and a board that mostly stayed out of his way.
Now look at who has the chair. Reuters ran its profile of Kress under the headline "the CFO who challenged Huang's bets." By every account she is excellent. She is also interim, she answers to a board with an open investigation, and she is guiding to gross margins of 71% to 72% while memory prices squeeze her.
Nobody in that seat signs a $6.3 billion backstop for a customer's unsold inventory. She can't. A deal that reads as vision when a founder signs it reads as recklessness when a caretaker does.
“The chips were never the scarce thing Jensen controlled. The exceptions were.”
The First Thing to Change Was Policy, and It Took Five Days
Want proof that the risk lives in the discretionary layer? You already have it.
For two years Huang was the loudest voice in Washington for selling into China. He made the case in public and lobbied for it in private, and every time the rules tightened he went back and designed another compliant chip. You can agree with him or not. The position was his, and it cost him politically to hold it.
Then came Kress's first earnings call, where Nvidia said it will stop selling H200 chips in China. The $130 billion guide assumes zero data center compute revenue from the country. Yes, new U.S. rules were pushing the same way, and maybe Huang loses that fight regardless. But the company quit fighting inside a week of his exit.
I'm not here to tell you the China call is wrong. Luckey thinks Beijing is delighted. Others think the exit closes a hole. What should bother you is how fast it happened. A stance that shaped global chip flows for two years turned out to belong to a person, and when he walked out, it walked out with him.
So ask yourself which other Nvidia positions were personal. The willingness to finance customers. The habit of seeding neoclouds as a counterweight to Microsoft, Amazon and Google. The patience to fund CUDA for a decade while Wall Street griped about the margin drag. I can't tell you which of those make it through a CFO-led interregnum and a search for a permanent CEO. Nvidia can't either, yet.
And that gets us to the strangest detail of the week.
Nvidia Hasn't Finished Writing Down Its Own Risk
Nvidia normally files its 10-Q the same day it reports earnings. This time it didn't. On the call, Kress said the filing wouldn't come immediately. She gave no reason. UBS analyst Timothy Arcuri wrote afterward that management tied the delay to the ongoing investigation and was weighing whether to add risk factors.
The 10-Q is the long document. You get the revenue from the press release. The legal proceedings and the risk factors go in the 10-Q, in the section where a company has to put in writing what could go wrong.
Nvidia's stock fell four straight days after the resignation, and by Wednesday's close it had given back its post-earnings pop. Nobody in the market doubts the $115.8 billion. Investors are waiting for the part of the story that won't fit in a headline number.
To me, the delay is the most honest signal of the week. Here is a $130-billion-a-quarter company with the best finance team in semiconductors, and it needs extra time to work out how to describe its exposure to one man leaving.
Have you described yours?
Your Compute Plan Has an Undisclosed Risk Factor
If you build on rented GPUs, this is the part you won't enjoy.
The big labs hedged already. Anthropic signed for up to a gigawatt of Google TPUs. OpenAI struck a six-gigawatt deal with AMD and started a custom-silicon program with Broadcom. They did all of that while cashing Nvidia's checks, because they understood who they were dealing with: a counterparty whose generosity was a policy. Policies change.
Startups mostly skipped that step. Say you're a twenty-person company. Your compute plan probably goes something like this: hyperscaler credits now, a reserved contract with a neocloud at a price that assumes current terms, a move to the next chip generation when it lands. Every line of it sits downstream of Nvidia's discretionary behavior. Your neocloud prices off its cost of capital, and its cost of capital depends on whether Nvidia stands behind its capacity. Your spot in the queue depends on your provider's spot in Nvidia's.
You never wrote any of that down. For three years it was as reliable as the weather, so why would you?
I've read a lot of seed and Series A decks since 2023. "GPU availability" shows up as a risk in dozens of them. Not one has ever said "our provider's financing depends on a vendor relationship that depends on one executive's strategy." It would have sounded paranoid. This week it's a plain description of the cap table two layers above you.
So write the risk factor. It costs you an afternoon, and it forces three questions:
1. Who is my provider's Jensen? Find out how they're financed and what their Nvidia relationship rests on. A neocloud with a backstop, an equity tie and early allocation is a different counterparty than one buying at list. 2. What in my contract survives a tighter Nvidia? Check the renewal pricing, the delivery dates for next-generation capacity, and what happens if your provider gets acquired or restructured. 3. How many weeks to run on something else? Skip the question of whether it's possible. Count the weeks. If the honest answer for inference is "we've never tried," try it this month on one workload, on TPUs or Trainium or AMD, while it's still a choice.
You don't have to believe Nvidia is in trouble to do any of this. I don't believe it. The floor is 55 wins.
The Bulls kept selling out the United Center in 1994, too. They just weren't the team where anything could happen anymore, and every player whose career was riding on that found out in the second round.
Nvidia is taking a few extra days to write down its risks. You've had three years.
Frequently Asked Questions
Will Jensen Huang leaving Nvidia affect GPU supply for startups?
Not in the near term. Nvidia reported $115.8 billion in quarterly revenue five days after he left and guided to $130 billion, because the orders were signed and the fabs were booked long before his exit. The risk is in the discretionary layer he controlled: which neoclouds got scarce chips early, which customers got balance-sheet support, and who counted as strategic when supply ran short.
Why does the article call Jensen Huang a "central bank" for AI?
Because his Nvidia acted like a lender of last resort, not a component vendor. It agreed to buy up to $6.3 billion of CoreWeave's unsold capacity, signed a letter of intent for up to $100 billion into OpenAI, committed up to $10 billion to Anthropic, leased back GPUs from Lambda, and took a $5 billion stake in Intel. Those were bets a founder with 33 years of standing could sign, and an interim CEO answering to a board with an open investigation is unlikely to repeat them.
How should an AI startup assess its exposure to Nvidia after Huang's exit?
Write the risk down and answer three questions. Find out how your compute provider is financed and what its Nvidia relationship rests on, then check what in your contract survives a tighter Nvidia: renewal pricing, next-generation delivery dates, and what happens if the provider is acquired or restructured. Finally, count how many weeks it would take to run on something else, and if you have never tried, move one inference workload to TPUs, Trainium or AMD this month.
Join the OPC Community
Connect with solo founders worldwide. Get daily insights, curated opportunities, and peer support.
Join free