By the time Jensen Huang walked off the SAP Center stage on Monday, the quantum question had already been answered. He didn't have to argue it. He just had to fold it into the survey — automotive, financial services, healthcare, industrial, media, quantum, retail, robotics, telecom — and keep moving. Quantum is part of the accelerated-computing ecosystem now, in NVIDIA's telling. It's just one of the verticals.
That is a quietly remarkable place to land, given where Huang was fourteen months ago.
At CES 2025, Huang told a roomful of analysts that "very useful" quantum computers were probably fifteen years away, and that twenty years was a more reasonable expectation. He delivered the line in his usual cadence — half exposition, half throwaway — and went on with the keynote. Inside an hour, the quantum-stock complex was on fire.
Rigetti dropped roughly 45% on the day. Quantum Computing Inc. fell about 43%. IonQ slid 39%. D-Wave fell 36%. These were not small moves on small floats; these were tens of billions of dollars in market cap erased on a single offhand answer from one CEO who, until that moment, was not even ostensibly in the quantum business. D-Wave's Alan Baratz went on television within a day and called Huang "dead wrong."
The story, as it was told that week, was that Huang had ambushed a frothy retail trade. The more interesting story, in retrospect, is that he gave the impression NVIDIA had no quantum thesis at all.
By GTC Paris in June 2025, the framing had shifted. Huang told the audience that quantum computing was "reaching an inflection point" and that the industry was "within reach" of applying quantum computers to "interesting problems in the coming years." He acknowledged, more or less directly, that his January remarks had come out wrong, and expressed surprise at how violently the market had read them.
This is a leader course-correcting in public. CEOs of his stature rarely do that on tape. The question worth asking is not whether Huang flipped — clearly he did — but what changed between January and June that made the flip cheap to perform. The likeliest answer is that NVIDIA had decided quantum was not going to be a side project, and the messaging had to stop pointing the other direction.
The proof arrived at GTC Washington in October 2025 with the launch of NVQLink — an open system architecture for tightly coupling GPU computing with quantum processors. The announcement read like a coalition statement: 17 QPU builders, five quantum-controller builders, and nine US national labs, including Brookhaven, Fermilab, Berkeley Lab, Los Alamos, MIT Lincoln Lab, Oak Ridge, Pacific Northwest, and Sandia. Huang called NVQLink "the Rosetta Stone connecting quantum and classical supercomputers."
That is not the language of a man who thinks quantum is twenty years away from useful work.
NVQLink is, mechanically, a low-latency interconnect for real-time control, calibration, and error correction. Strategically, it is something else: a bid to make NVIDIA the default classical half of every hybrid quantum-classical machine that gets built in the next decade. If a national lab wants to wire up an IQM superconducting processor or a Quantinuum trapped-ion stack to a Grace Blackwell rack, NVIDIA wants the bridge to be theirs, and it wants the runtime — CUDA-Q — to be theirs as well.
You don't ship that kind of platform unless you intend to sit underneath the entire category.
Which brings us back to this week. The GTC 2026 keynote, on Monday at the SAP Center, was overwhelmingly about the Vera Rubin platform — seven chips, five rack-scale systems, one supercomputer, NVLink 6, a multi-trillion-dollar revenue ambition that Huang now stretches out to 2027. The Groq integration, post-acquisition, took its own beat. Quantum did not get a Vera-Rubin-sized moment.
That is the point. Quantum did not need one. NVQLink already shipped. The CUDA-Q stack already exists. The labs are already wired in. By March 2026, NVIDIA's quantum story is a maintenance story — additions to the supported QPU list, latency improvements in the runtime, new error-correction tooling — not a launch story. Huang surveyed it inside the broader ecosystem because, from his perspective, that is now where it lives.
A reasonable observer looking at the arc — January's "20 years" remark, June's "inflection point" reversal, October's NVQLink launch, March's offhand inclusion in the ecosystem list — might fairly conclude that NVIDIA was building this entire stack while Huang was still publicly saying it didn't matter. The January comments did not reflect the company's plan. They reflected, at most, what Huang was willing to say while the plan was still in private.
For technology buyers, the practical read is narrower than the narrative.
First, classical-side AI compute budgets are not going to be redirected to quantum any time soon. Vera Rubin is the story for 2026 and 2027. Quantum is, for almost everyone outside a national lab or a large pharma research arm, still a watching-brief expense.
Second, "hybrid" is now the operative word. NVIDIA's bet is that the productive unit is a quantum processor wired to a GPU supercomputer with a fast bus between them — not a standalone QPU running on its own. Buyers evaluating any vendor's quantum roadmap should ask, very specifically, what the classical-side integration looks like, and whether it relies on NVQLink, a competitor's interconnect, or nothing at all.
Third, the credibility math has shifted. When Huang said quantum was twenty years away, a lot of buyers used that as cover to defer the conversation. He is not saying that anymore, and the infrastructure he has shipped since does not say that either. The deferral is harder to justify in 2026 than it was in early 2025.
None of this means quantum is suddenly close. It means Huang's read on how close has changed, the supporting infrastructure has been built, and the people writing capital-expenditure plans should at least know what NVQLink is, what CUDA-Q does, and which of their vendors plug into either. The pivot is real. The downstream decisions are going to take longer than the talking points did.
At CES 2025, Huang said "very useful" quantum computers were probably 15-20 years away. Within a single trading day, Rigetti dropped roughly 45%, Quantum Computing Inc. about 43%, IonQ about 39%, D-Wave about 36% — tens of billions of dollars in market cap erased on one offhand answer.
Launched October 28, 2025 at GTC DC, NVQLink is NVIDIA's interconnect fabric tying QPUs from 17 quantum hardware vendors and 5 controller builders to GPU supercomputers across 9 US national labs. Huang called it the "Rosetta Stone" between classical and quantum compute. It's the concrete infrastructure that made Huang's pivot expensive to walk back.
Most should not yet — Vera Rubin and accelerated GPU compute will absorb 2026 and 2027 capex. But buyers should know what NVQLink and CUDA-Q are, and which vendors plug into either, so the watching brief is informed when budgets shift.
Discover more content:
