NVIDIA’s 22x Revenue Surge Shows Why Quantum’s Quiet Phase Deserves Attention

The numbers are extraordinary.

NVIDIA generated $9.71 billion in revenue in fiscal 2018, the year ended January 28, 2018. By fiscal 2026, revenue had reached $215.9 billion, according to the company’s financial results.

That is an increase of more than 22 times in eight years, equivalent to a compound annual growth rate of roughly 47%.

The comparison becomes even more striking at the quarterly level. NVIDIA generated $68.1 billion in revenue during the fourth quarter of fiscal 2026 alone — roughly seven times what the company produced during the entire fiscal year eight years earlier.

Much of that expansion eventually became associated with generative AI, data centers and the extraordinary demand for GPUs required to train and run large language models. Today, investors routinely debate AI infrastructure spending, data-center capacity and whether hyperscalers can earn adequate returns from hundreds of billions of dollars in capital expenditure.

But one of the technologies that ultimately helped create that demand arrived long before the market began treating generative AI as a dominant investment theme.

On June 12, 2017, eight Google researchers submitted a paper titled “Attention Is All You Need.” The research introduced the Transformer architecture, replacing the recurrent and convolutional approaches then common in sequence processing with an architecture based around attention mechanisms.

The paper was initially demonstrated on machine-translation tasks. There was no ChatGPT, no consumer generative-AI boom and no obvious reason for public-market investors to expect it to transform NVIDIA’s income statement.

ChatGPT did not launch publicly until November 30, 2022 — more than five years later.

During that period, researchers continued developing increasingly capable Transformer-based models while improvements in computing hardware, datasets and model scaling gradually turned an academic architecture into commercially useful products.

The eventual explosion created businesses across almost every layer of the technology stack. Model developers such as OpenAI and Anthropic became some of the world’s most closely watched private technology companies. Meanwhile, companies surrounding the models — including cloud providers, chipmakers and businesses providing AI infrastructure — attracted enormous amounts of capital.

The point is not that investors could easily have identified the Transformer paper in 2017 and predicted NVIDIA’s eventual revenue trajectory.

They could not.

Many promising research papers never become commercially important. Even technologies that work scientifically can fail because they are too expensive, difficult to scale or unable to find a compelling application.

But the timeline demonstrates something important: major technological shifts can spend years developing inside research communities before the commercial evidence becomes obvious.

Quantum Error Correction Is Producing More Concrete Milestones

Quantum computing may be entering a similarly important research period, although the comparison remains highly uncertain.

The most meaningful advances are increasingly centered on error correction and logical qubits rather than simply increasing the raw number of physical qubits.

Quantum information is extremely sensitive to noise. A useful large-scale quantum computer therefore needs to combine physical qubits into logical qubits that can detect and correct errors while calculations are running.

Google demonstrated an important milestone with its Willow processor when researchers showed that increasing the size of a surface-code logical qubit reduced errors rather than making them worse. The result, published in Nature, demonstrated operation below the error-correction threshold — a requirement for building increasingly reliable logical qubits as systems scale.

Google has continued pushing that work forward. In 2026, its researchers published results using dynamic surface codes and demonstrated techniques designed to help quantum computers continuously adapt to hardware drift during long computations.

IBM and the University of Chicago reported another notable result in July. Researchers encoded 70 logical qubits and performed more than 2,400 logical two-qubit operations in an experiment designed to solve a computationally difficult sampling problem. IBM said the effective logical error rate was about ten times lower than the underlying physical error rate.

Quantinuum has also reported rapid progress in logical-qubit efficiency. The company said in March that its hardware produced as many as 48 error-corrected logical qubits from 98 physical qubits, while larger error-detected configurations reached 94 logical qubits.

These experiments are not equivalent to commercially useful fault-tolerant quantum computing.

They do, however, make the industry’s claims increasingly measurable.

Instead of simply promising that quantum computers will eventually transform chemistry, materials science, cryptography or optimization, researchers can increasingly specify how many logical qubits were created, how much error reduction was achieved, how many logical operations were completed and how the result compared with classical computation.

Quantum Still Has a Much Harder Commercialization Problem Than AI

The NVIDIA analogy becomes dangerous if taken too literally.

Transformers were software architectures that could scale rapidly once sufficient compute became available. Quantum computers require entirely new physical computing systems, highly controlled environments and error rates low enough to maintain fragile quantum states through large numbers of operations.

The engineering problem is much harder.

IBM is targeting a fault-tolerant system called Starling for 2029. Google has said it is increasingly confident that commercially relevant superconducting quantum computers could emerge before the end of the decade. Microsoft has also targeted 2029 for a scalable system using its topological-qubit approach.

Those are company roadmaps, not guarantees.

Microsoft illustrates why skepticism remains necessary. Its Majorana quantum program has produced ambitious claims about topological qubits, but outside physicists have repeatedly questioned whether the published evidence fully supports those claims. Even after Microsoft unveiled an improved Majorana 2 chip in 2026, Nature reported that researchers remained skeptical.

Commercial demand is another unresolved issue.

Generative AI found an obvious mass-market interface in chatbots and is now spreading into coding, search, enterprise software and AI agents. Quantum computing does not yet have an equivalent application capable of pulling millions of ordinary users into the technology.

Most proposed quantum advantages remain concentrated in specialized fields such as molecular simulation, materials discovery, cryptography and certain optimization problems.

The Investment Opportunity May Be in Identifying the Bottleneck Before the Product

This is where the NVIDIA comparison becomes more useful.

The lesson from AI is not that investors should find one promising research paper and immediately buy whatever companies are associated with it.

The better lesson is to identify what must become scarce if the technology works.

For generative AI, the bottleneck became compute.

Transformers scaled extremely well with more data and processing power. Once the models became commercially useful, demand cascaded backward through the technology stack: GPUs, networking, high-bandwidth memory, cloud infrastructure, cooling systems, electricity and data centers.

NVIDIA happened to control one of the hardest bottlenecks.

Quantum computing may develop very differently.

The eventual winner may be a quantum-hardware company. But it could just as easily be a supplier of cryogenic equipment, control electronics, photonics, error-correction software, semiconductor manufacturing technology or hybrid classical-quantum infrastructure.

That is why focusing only on finding “the NVIDIA of quantum” may be the wrong question.

Investors did not need to correctly forecast ChatGPT in 2018 to notice that machine learning was becoming more computationally intensive. Likewise, investors today may not need to know which quantum architecture ultimately wins to monitor which technical bottlenecks are becoming increasingly important.

Large technology companies have another advantage: they can fund research for years before commercialization becomes obvious. Alphabet can support long-duration quantum research partly because Google’s core advertising business continues producing enormous revenue and cash flow. Microsoft and IBM have similarly deep financial resources.

That makes the current phase particularly difficult for investors. Much of the important progress is happening inside companies where quantum contributes almost nothing to current valuation, while publicly traded pure-play quantum companies can be priced heavily on expectations rather than revenue.

The distinction between technical progress and investable economics therefore matters enormously.

Logical error rates should improve. Useful circuit depth should increase. Hardware must scale. Customers must eventually demonstrate that quantum calculations provide enough economic value to justify their cost.

Those are measurable milestones.

If they continue improving, the market may eventually experience its equivalent of the November 2022 ChatGPT moment — the point when years of obscure technical progress suddenly become visible to everyone.

There is no guarantee that moment arrives.

But NVIDIA’s rise from $9.71 billion to more than $215 billion in annual revenue is a reminder that by the time a technology’s commercial importance becomes obvious on an income statement, the research phase that created it may already be many years old.

For quantum computing, the more useful question may therefore be shifting from “does this technology work?” to “which technical claims are becoming reproducible, which bottlenecks are disappearing and who controls what remains scarce?”

That is where the next major technology cycle usually starts becoming investable — long before the revenue chart makes it obvious.