Can Nvidia Keep Its Customers Without Jensen Huang?
Jensen Huang's exit puts Nvidia's CUDA moat, chip financing deals, and political clout to the test. Here's what's actually at risk.
What happens now that Jensen Huang has left Nvidia?
Jensen Huang has resigned as Nvidia’s CEO, president, and board member. Colette Kress, Nvidia’s CFO, is stepping in as interim CEO while keeping her finance role, and Harvey Jones has been named the company’s first-ever board chair. Nvidia’s board says operations, customer commitments, and the product roadmap will continue without interruption. That’s probably true in the narrow sense that the engineering teams still know how to build chips. But Huang didn’t just run Nvidia. He personally brokered the deals, financing structures, and political relationships that let Nvidia dominate AI computing. Replacing that function is a different problem than replacing a CEO.
TL;DR
- Jensen Huang’s resignation removes the person who personally coordinated chip allocation, customer priority, and deal terms across Nvidia’s biggest accounts, a function that doesn’t show up on an org chart but shaped who got scarce GPUs first.
- CUDA lock-in is still Nvidia’s strongest moat, but Anthropic’s demonstration of using its Claude models to port prompt processing to AMD’s MI355X chips in about three days suggests AI-assisted coding could start chipping away at switching costs over time.
- Nvidia’s financing arrangements, including preliminary deals aimed at mobilizing more than half a trillion dollars in outside capital and a $105 billion guarantee tied to an SB Energy data center project serving OpenAI, were built on confidence that Huang’s leadership helped create.
- Export control politics get harder without Huang, who was Nvidia’s most visible advocate for selling AI chips into China; legislation like the proposed Bipartisan AI Overwatch Act could tighten restrictions regardless of who runs the company.
- Customers signing multi-year deals now have to ask whether Nvidia can still deliver the full package (chips, support, financing certainty, and political cover) rather than just competitive hardware.
- More chip competition from AMD and others, if it materializes, would likely mean lower prices and more computing capacity for everyone building AI products, even if it costs Nvidia market share.
MindStudio, Remy, and new opportunities
On the agenda- The state of MindStudio
- Meet Remy
- The big announcement
- Live Q&A
Why did Nvidia need Jensen Huang more than most CEOs?
Most large companies survive a CEO transition because the business runs on systems, not personalities. Nvidia’s position in AI infrastructure runs on both. Huang spent decades building the relationships and credibility that let him make judgment calls other executives couldn’t: which cloud provider gets priority on scarce chip supply, what terms make a multi-year, multi-billion-dollar deal workable, who gets bumped up the queue when a new chip generation ships late.
Those aren’t decisions written into a contract. They’re decisions that require someone with unquestioned authority in the industry, someone counterparties trust to be fair and competent at the same time. Nvidia’s thousands of engineers still have the technical expertise to build chips and support deployments. That was never really in doubt. What’s less clear is who inside Nvidia inherits the coordination and prioritization role Huang played, especially while the company searches for a permanent CEO.
Is Nvidia’s CUDA moat actually weakening?
CUDA (Compute Unified Device Architecture) is the software layer that lets programmers run AI workloads on Nvidia chips. It’s widely considered one of the strongest moats in tech, because developers have spent years writing code around it, and porting that code to a competitor’s hardware is expensive, slow, engineering work. That cost, more than raw chip performance, has kept customers locked into Nvidia even when rival hardware looked attractive on paper.
That moat hasn’t disappeared, but there’s a new source of soft risk. During its IPO roadshow, Anthropic described using its own coding model to get Claude running on AMD’s MI355X chips in roughly three days, with the model writing or adapting nearly all of the low-level software routines involved. Anthropic compared that to earlier porting efforts that took dozens of engineers months to complete.
The caveat matters: this was prompt processing for one model on one chip, not a full production deployment across Claude’s entire stack. Anthropic has also invested in multi-chip platform support for years, so it was unusually well positioned to pull this off quickly. It is not evidence that any company can rewrite its way around CUDA in a weekend.
What it does suggest is that as frontier models get better at complex software engineering, the labor cost of switching chip platforms could keep dropping. That doesn’t kill Nvidia’s moat next quarter or next year. It does mean the moat is no longer a fixed, permanent wall. It’s a cost that may erode gradually as AI tools get better at doing the porting work humans used to do by hand.
How does chip financing get more complicated without Huang?
Choosing a chip vendor is only step one for anyone building AI infrastructure. The bigger challenge is financing it. A data center developer needs capital. A lender needs confidence that customers will actually rent the capacity once it’s built. Those customers need confidence the hardware will arrive on schedule and perform as promised. Nvidia, under Huang, inserted itself deep into all three legs of that chain.
Remy.
Nvidia’s own quarterly filings describe preliminary agreements aimed at mobilizing more than half a trillion dollars of outside capital over time. More concretely, Nvidia has signed guarantees capped at $105 billion tied to an SB Energy data center project that will serve OpenAI. Those guarantees cover lease and power obligations if the tenant defaults, phasing in as facilities come online starting in Nvidia’s fiscal 2029. In exchange, the site is expected to run on Nvidia’s computing infrastructure, with limited exceptions.
Deals at that scale get negotiated because the counterparties trust the person putting them together. Huang’s standing in the industry was arguably as important to closing those agreements as Nvidia’s balance sheet. Whether a new, unproven leadership team can command the same confidence from lenders and partners is an open question. If it can’t, the risk isn’t that existing guarantees unravel overnight. It’s that the next round of deals takes longer to close, lenders ask for tighter terms, and construction timelines slip.
What does Huang’s exit mean for Nvidia’s position in Washington?
Huang was also Nvidia’s chief advocate in the ongoing fight over AI chip exports to China. His argument was that keeping Chinese developers dependent on American technology preserves US influence, while cutting them off just accelerates China’s push to build domestic alternatives. Lawmakers pushing for tighter restrictions see the same dependence as a national security risk, worried that advanced chips could accelerate Chinese military and intelligence capabilities.
That debate isn’t going away, and Huang’s resignation doesn’t resolve any of the underlying questions about restricted shipments or allegations of improper influence over export decisions. If anything, it removes Nvidia’s most effective advocate from the conversation at a moment when legislation like the proposed Bipartisan AI Overwatch Act is specifically aimed at tightening chip export rules. Senator Elizabeth Warren has already cited allegations involving Nvidia as a reason to push that bill forward. With control of Congress in flux, the political outcome is genuinely uncertain, but it’s a fight Nvidia now has to wage without its most credible spokesperson.
Is Nvidia still worth betting on for customers signing deals now?
For a company about to sign a multi-year chip and infrastructure commitment, the calculation has gotten more complicated. The hardware is still excellent and the roadmap is reportedly unchanged. But a deal with Nvidia was never just about chips. It included delivery assurances, technical support, financing help, and a level of confidence that came from Huang’s track record of closing these arrangements successfully.
That full package is now less certain, even if no single piece of it has visibly broken. Customers weighing Q4 or early-2026 commitments have to ask whether Nvidia’s new leadership can still deliver everything that used to come bundled with the chips, at a price that still makes sense as alternatives like AMD become marginally easier to adopt.
None of this means Nvidia stops being a dominant force in AI infrastructure. It almost certainly remains one. But a leadership transition at this scale, combined with early signs that switching costs may be softening and a financing structure built on personal credibility, adds up to real uncertainty heading into the next wave of AI infrastructure deals. If competition over chip platforms increases and margins compress even slightly, the likely beneficiary is everyone buying inference and compute, not Nvidia’s market share.
Frequently Asked Questions
Who is running Nvidia now that Jensen Huang has resigned?
Colette Kress, Nvidia’s longtime CFO, is serving as interim CEO while retaining her finance duties. Harvey Jones has been appointed as Nvidia’s first-ever board chair. A permanent CEO has not been named.
What is CUDA and why does it matter for Nvidia’s competitive position?
CUDA (Compute Unified Device Architecture) is Nvidia’s software platform for running AI workloads on its chips. Years of developer code built around CUDA make switching to competing hardware costly and slow, which has functioned as one of Nvidia’s strongest competitive moats.
Did Anthropic really port Claude to AMD chips in three days?
Anthropic said its Mythos model got Claude’s prompt processing running on AMD’s MI355X chips in about three days with minimal human help, with the model writing most of the low-level code itself. This covered prompt processing for one model on one chip type, not a full-scale production deployment, so it demonstrates potential rather than a complete chip-switching solution.
How much money is tied up in Nvidia’s financing commitments?
Nvidia’s filings describe preliminary agreements aimed at mobilizing more than half a trillion dollars in outside capital over time, plus specific guarantees capped at $105 billion tied to an SB Energy data center project serving OpenAI, phasing in starting in Nvidia’s fiscal 2029.
Could US export restrictions on AI chips get tighter after Huang’s departure?
It’s possible. Huang was Nvidia’s leading voice against strict export controls on chips sold to China. Legislation such as the proposed Bipartisan AI Overwatch Act is already under consideration, and his exit removes the industry’s most prominent advocate from that debate at a politically uncertain moment.
