Nvidia’s Market Value and the Future of AI

Looking back at Nvidia’s market value over roughly the past decade, it is easy to develop an almost automatic expectation of further growth. According to CompaniesMarketCap’s historical data, the company was worth about US$57.5 billion at the end of 2016. By the close of trading on 13 November 2026, its market capitalisation was approximately US$3.93 trillion, around 68 times as much. There were substantial setbacks along the way: its value at the end of 2022, for example, was roughly half what it had been a year earlier. Those fluctuations do not change the extraordinary growth it achieved over this period.[1]

That growth has a concrete basis. Demand for high-performance chips expanded, and Nvidia captured a large share of the opportunity through its GPUs, software ecosystem and systems engineering. As large-scale AI training and inference became central to the industry, the computing infrastructure it supplied acquired unprecedented importance.

Before extrapolating from that history, though, I think we need to clarify something: how closely do the forces driving Nvidia’s growth correspond to those driving progress in artificial intelligence?

Nvidia’s business is driven fundamentally by demand for its chips and related computing products. More companies purchasing computing capacity and more large data centres coming into operation generally mean greater commercial opportunities. AI has its own questions to answer: how to improve learning, reasoning, understanding and generalisation; how to solve practical problems more reliably; and what resources those capabilities require.

At present, these two sets of interests overlap considerably. There is no reason to assume that their relationship will always remain as it is today.

Increases in computing power have played a major role in advancing frontier AI over recent years. With existing models and training methods, putting more computation into a system has indeed been an effective approach. But effectiveness at one stage of development should not be mistaken for an inevitable law of progress. AI’s current dependence on computing hardware does not establish that future AI should become ever more dependent on it.

I do not think a healthy course of AI development should take continually rising hardware investment as its default premise. If better algorithms, model architectures or learning methods allow a system to achieve more with less computation, that is also significant progress. It may sometimes matter more than simply expanding the scale of computation. Ultimately, we have to assess progress through capability, reliability and actual cost, rather than chip counts alone.

This does not mean demand for computing power must fall. When the cost of a training run or an inference comes down, more people may begin using AI, and applications that were previously uneconomic may become viable. Total demand could still rise. But the outcome depends on how much efficiency improves, how quickly applications expand and which capabilities users are willing to pay for. We cannot assume in advance that every breakthrough in efficiency will eventually produce more GPU orders.

The connection between technical progress in AI and Nvidia’s commercial growth is therefore conditional. More efficient AI could continue to enlarge its market, or it could change the kinds of hardware the market needs. To assess Nvidia’s long-term value, we have to ask how AI’s growth will be achieved, as well as how much growth there might be.

This is also where my doubts about Jensen Huang’s account of the AI industry begin.

On 10 March 2026, Huang published “AI Is a 5-Layer Cake” on Nvidia’s official blog. He described the industry as five layers: energy, chips, infrastructure, models and applications. He used the term “AI factories” for large computing facilities, describing their function as producing intelligence, and argued that trillions of dollars of infrastructure still needed to be built.[2]

As a description of industrial organisation, this account has considerable explanatory value. Any AI system that actually runs needs physical resources, and energy, chips and infrastructure do constrain its scale. But the account primarily explains how an industry is organised and expands. Why intelligent capabilities emerge, and how they might improve further, remain separate questions.

My criticism is that Nvidia’s emphasis on infrastructure expansion does not get to the central logic of AI development. It can explain how to scale up existing technology, but it does not adequately explain how new breakthroughs in intelligent capability will occur. When “producing more intelligence” is expressed mainly in terms of additional computing investment, the growth requirements of a computing supplier can easily come to be accepted as requirements of AI itself.

There is a clear commercial interest here. The more AI depends on large computing facilities, the more favourable Nvidia’s position generally becomes. Its emphasis on computing power is entirely understandable. That does not make its position a neutral account of AI’s future. A company’s products being indispensable to the current technical approach does not establish that this approach is the inevitable path of intelligence.

It is in this sense that I find calling Huang the “godfather of AI” quite absurd.

If the label merely refers to his influence within the industry, it is a piece of media rhetoric. If it implies that he laid AI’s theoretical foundations, or has special authority on the nature and future of intelligence, it confuses different kinds of contribution. Geoffrey Hinton, who is commonly described as a godfather of AI, holds that standing because of his research on neural networks and machine learning. In 2024, he and John Hopfield shared the Nobel Prize in Physics for foundational discoveries and inventions enabling machine learning with artificial neural networks.[3]

Huang’s industrial and engineering contributions are substantial. The GPU and CUDA ecosystem developed under his leadership at Nvidia provided important conditions for modern AI’s development. Those achievements are significant enough in their own right; they do not need to be recast as foundational contributions to AI theory. What I question is the technical authority that is extended from commercial success, and whether it leads people to surrender their independent judgement of his industry narrative.

His earlier remarks on quantum computing made me more concerned about this issue.

During an Nvidia analyst question-and-answer session on 7 January 2025, Huang discussed when “very useful quantum computers” might arrive. He suggested that 15 years would probably be early, 30 years would be late, and 20 years was a timeframe many people would find plausible. Precision matters here: he offered a notably conservative assessment of the timetable for practical use. He did not reject the principles of quantum computing or dismiss all its value. In the same answer, he also described Nvidia’s work on combining quantum and classical computing through CUDA-Q.[4]

Even so, publicly placing the crucial developments so far into the future struck me as strongly defensive towards a potentially competing approach. I cannot rule out the possibility that protecting an established industrial advantage influenced that judgement. This is my inference about his commercial position; his public statements alone do not establish his subjective motives.

The subsequent developments also belong in the picture. At GTC’s Quantum Day on 20 March 2025, Huang acknowledged problems with how he had expressed his earlier view and invited quantum computing companies to discuss the field’s development. On 14 September 2026, Nvidia announced a further expansion of CUDA-Q, adding CUDA-Q Logical for designing and testing applications for fault-tolerant quantum computing.[5][6]

These facts do not support portraying him as consistently rejecting quantum computing. Nvidia is clearly seeking a role in the field and connecting its classical computing capabilities to the new technology. But my original question remains. When a company with an advantage in the existing computing system assesses a path that could change that system, how much of its judgement comes from technical analysis, and how much from protecting its position? It can participate in a new approach while still underestimating developments that do not follow directly from its existing strengths.

Compared with continually expanding today’s GPU systems, I am more optimistic about quantum computing. I think it could deliver the next genuine breakthrough in the computing paradigm.

GPUs can continue to achieve substantial gains through greater parallelism, better memory and interconnects, and more efficient system design. These advances matter greatly, but they largely remain within the classical computing paradigm. Quantum computing uses superposition, entanglement and interference to organise computation. For suitable problems, it could change how the resources required for a calculation grow with the size of the problem. The possibilities extend beyond making existing computations somewhat faster.[7]

This is the main reason I take quantum computing seriously. If a different computational mechanism can solve some of the problems that are difficult to handle today, the change could reach the foundations of the existing computing industry. Increasing the number of devices of the same kind has a different significance from finding another way to approach a problem computationally.

There has already been concrete progress. On 11 November 2026, IBM reported that its Nighthawk r2 processor, introduced that September, could support computations involving more than 7,500 quantum operations across 120 programmable qubits. The next-generation Nighthawk r3 and larger systems announced on the same day remained plans for subsequent development. These advances deserve attention, but improved processor performance, quantum advantage in particular experiments and widely usable fault-tolerant quantum computing are distinct achievements.[8]

One boundary is especially important to preserve: quantum computing’s potential for major computational breakthroughs does not automatically make it the principal computing foundation for the next generation of AI. The suitability of certain problems for quantum algorithms does not establish that large-model training and everyday inference will enjoy the same benefits.

An OECD report published on 5 November 2026 identified continuing technical bottlenecks in quantum AI, including the input of classical data and the readout of quantum results. It included hybrid quantum–classical computing among the approaches more likely to contribute in the near term. An experimental survey published on 12 November assessed 107 studies first made public between 2024 and 6 August 2026. Under its standard for practical advantage, which requires independent replication and comparison with the best available classical methods, none of the included studies qualified. That finding concerns a particular body of research assessed against a specified standard of evidence. It should not be extended into a rejection of quantum machine learning’s future.[7][9]

These limitations have not diminished the importance I attach to quantum computing. They show that the potential of a new computing paradigm still has to be realised through working systems, suitable algorithms and real tasks. If quantum computing eventually changes AI’s computational foundations, the industrial structure built around GPUs may not simply carry over. Nvidia could remain important and could successfully participate in the new system, but its lead today cannot guarantee its advantage in the next stage.

Returning to Nvidia’s market value, what concerns me throughout is the relationship between growth in intelligent capability and growth in demand for computing power. Its extraordinary expansion over roughly the past decade shows how successfully it seized the opportunities of this period. If future AI can achieve more with fewer resources, or overcome existing limits through a new computational mechanism, I would regard that as welcome progress. Whether such progress would also allow Nvidia to maintain its past growth trajectory is a separate judgement.

Sources (checked as at 15 November 2026)

[1] CompaniesMarketCap, NVIDIA market capitalisation history. The figures for the end of 2016, the end of 2022 and 13 November 2026 come from the site’s historical series; the multiple is calculated from those figures.
https://www.companiesmarketcap.com/nvidia/marketcap/

[2] Jensen Huang, AI Is a 5-Layer Cake, 10 March 2026, NVIDIA Blog.
https://blogs.nvidia.com/blog/ai-5-layer-cake/

[3] Nobel Prize committee, press release for the 2024 Nobel Prize in Physics, 8 October 2024.
https://www.nobelprize.org/prizes/physics/2024/press-release/

[4] NVIDIA Corporation Shareholder/Analyst Call, 7 January 2025, S&P Global Market Intelligence transcript, page 14.
https://seekingalpha.com/attachments/download/40411

[5] Kif Leswing, Nvidia CEO says he was wrong about timeline for quantum, surprised his comments hurt stocks, CNBC report, republished by AOL on 21 March 2025. The Quantum Day event took place on 20 March.
https://www.aol.com/nvidia-ceo-huang-says-wrong-162357231.html

[6] NVIDIA Expands Open Source CUDA-Q Platform for Fault-Tolerant Quantum Computing, 14 September 2026, official Nvidia announcement.
https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Expands-Open-Source-CUDA-Q-Platform-for-Fault-Tolerant-Quantum-Computing/default.aspx

[7] OECD Digital Economy and Society Outlook 2026, “Powering AI progress: The future of compute”, 5 November 2026; see especially the discussion of new computing paradigms and quantum computing.
https://www.oecd.org/en/publications/oecd-digital-economy-and-society-outlook-2026_022ed474-en/full-report/powering-ai-progress-the-future-of-compute_b4007d1b.html

[8] IBM Quantum Nighthawk Demonstrates New Levels of Speed and Quality, with Plans to Scale Beyond 1,000 Qubits in 2027, 11 November 2026, official IBM announcement.
https://newsroom.ibm.com/2026-11-11-ibm-quantum-nighthawk-demonstrates-new-levels-of-speed-and-quality,-with-plans-to-scale-beyond-1,000-qubits-in-2027

[9] Erasmo Rivas and colleagues, Quantum Machine Learning: An Experimental Survey, Quantum Reports, 12 November 2026.
https://doi.org/10.3390/quantum8040123