Skip to main content

Nvidia's Empty Throne & The AI Boom Expiry Date

Damien Klassen
by Damien Klassen
November 20, 2026

The AI boom has reached a point where strong earnings, enormous infrastructure spending and growing adoption are increasingly being weighed against questions about competition, financing and political intervention. Nvidia sits at the centre of much of that debate, but I think investors risk focusing too narrowly on the company itself when assessing whether the broader AI cycle can continue.

Nvidia's latest results were impressive, but one strong earnings announcement doesn't tell us much about how the business will perform under a new leadership structure, how competitive pressures will evolve or whether current profit margins are sustainable. The more interesting question isn't whether Nvidia is performing well today. It's whether the conditions underpinning that performance will still exist several years from now.

That distinction matters because the AI investment cycle isn't dependent on a single company, a single customer or even a single source of funding. It is a much broader ecosystem, with different players benefiting at different stages and facing very different risks.

Nvidia's competitive advantage is more than its chips

Nvidia's dominance is often framed around its graphics processing units (GPUs), but its competitive advantage extends well beyond the hardware. Its software ecosystem effectively acts as the operating system for AI data centres, helping customers integrate chips and other components into functioning computing systems. That creates a degree of lock-in that makes competing with Nvidia more difficult than simply producing a faster chip.

The company currently holds more than 90% of the market, although that share is unlikely to remain unchanged indefinitely. Amazon Web Services (AWS), Google and other major technology companies are developing their own chips, while semiconductor competitors such as AMD and Intel are also seeking to capture a larger portion of the market.

Competition will almost certainly intensify. But that doesn't necessarily mean Nvidia's business will deteriorate. Semiconductor markets typically consolidate around a relatively small number of major players, rather than fragmenting into hundreds of equally competitive businesses. Nvidia could conceivably lose market share over the next five to ten years, perhaps falling to 60–70%, without experiencing a significant decline in revenue.

The reason is simple: market share and market size are two different things. If the total market for AI computing expands substantially, Nvidia could sell more chips even while competitors take a larger slice of the pie.

The more immediate concern is profitability. Nvidia's margins are unusually high, partly because of supply constraints and its strong pricing position. As competition increases and supply conditions improve, those margins could eventually compress.

Yet even a scenario in which Nvidia's margins fall significantly and its earnings begin to resemble those of a more conventional semiconductor business within three to five years may not necessarily be catastrophic for shareholders. At current valuations, much of that deterioration may already be priced in, leaving room for further growth or shareholder returns to provide upside.

The Jensen Huang factor

Nvidia's competitive position also raises a less easily quantified question: how much of its success depends on Jensen Huang?

Huang has built a company that operates with an unusually direct, almost startup-like management structure despite its enormous scale. With around 60 people reportedly reporting directly to him, he has maintained an influence over operations, strategic decisions, financing and investor communication that is difficult to replicate.

That concentration of leadership creates both an advantage and a risk.

Huang's ability to make decisions quickly has been particularly valuable in navigating supply constraints. His early efforts to secure memory supplies from companies such as Micron, Samsung and SK Hynix reportedly helped Nvidia manage shortages and protect profitability.

A successor may not approach the business in the same way. The next generation of management may be more operationally focused, potentially resulting in a more conventional corporate structure. That could make Nvidia more stable and easier to manage, but it may also reduce the speed at which the company responds to new opportunities or makes strategic decisions.

Huang's financing approach has also been relatively aggressive, helping accelerate the expansion of the AI ecosystem. A more conservative successor could reduce the pace of investment and expansion.

I don't think that automatically makes a leadership transition negative. But it does mean investors shouldn't assume that Nvidia will continue operating exactly as it has in the past simply because its latest earnings remain strong.

China, politics and the risk of restrictions

Nvidia's exposure to China adds another layer of uncertainty. The political debate surrounding AI chips and export restrictions remains difficult to predict, with the possibility of tighter controls, limited enforcement or policy announcements that ultimately have little practical impact.

The consequences would depend heavily on how restrictions are implemented.

If Chinese customers or smaller neocloud providers are prevented from purchasing Nvidia chips, some demand could shift towards larger cloud providers such as Microsoft or Google. Nvidia's overall sales might not necessarily decline dramatically, but its customer base could become more concentrated.

That concentration could create its own risks. Meanwhile, hyperscalers could benefit from lower chip costs if supply becomes more readily available, while restrictions could encourage China to accelerate the development of domestic alternatives.

The bigger concern would be a more substantial crackdown that reduces access to technology across the AI ecosystem. That could weaken demand for US semiconductor suppliers, disrupt open-source AI development and accelerate the development of competing Chinese technologies.

For now, political restrictions are a risk to monitor rather than a settled outcome. The potential disruption is real, but the eventual economic impact will depend on the extent to which policy changes affect actual purchases, infrastructure investment and technological development.

AI infrastructure spending: what could slow it down?

The AI boom is supported by enormous capital expenditure (capex), but that spending isn't unlimited. Political opposition, electricity constraints, financing conditions and construction delays could all affect the pace of new data-centre development.

Texas is one example of the political and regulatory pressures surrounding the expansion. Its relatively inexpensive energy, available land and business-friendly environment have helped attract data-centre projects, but growing concerns about electricity and water consumption could introduce new restrictions.

Some proposed measures, however, may be less disruptive than they initially appear. Water-recycling requirements could add costs without materially changing the economics of a large data centre. Restrictions on connecting new facilities to the electricity grid may also have limited consequences if developers are already installing their own power systems while waiting years for grid access.

The more important constraint may be financing.

AI infrastructure investment is currently supported by three broad sources: debt raised by hyperscalers, equity financing for major AI companies and debt financing for smaller neocloud providers.

Google, Amazon and Microsoft still have access to debt markets, despite borrowing costs rising relative to government bonds. Major AI companies such as OpenAI and Anthropic continue to raise capital, while smaller neocloud providers rely more heavily on debt and may be more vulnerable to financing difficulties.

A slowdown among neocloud providers wouldn't necessarily end the infrastructure boom. If a smaller company can't finance a data centre, a larger competitor may be able to take over the project.

The consequences of slower spending could also be uneven. Nvidia and Broadcom may face weaker sales if infrastructure construction slows, while hyperscalers could benefit from lower chip costs and greater access to computing capacity.

The real test: AI demand

Ultimately, the AI investment cycle depends on whether businesses and consumers continue using AI, not just on how much money companies spend building infrastructure.

AI training remains an important source of demand, with companies including OpenAI, Anthropic, Google, Meta, SpaceX and Chinese developers competing to build increasingly capable models. That competition has created an arms race in which businesses spend heavily to avoid falling behind.

But training isn't the only source of demand. Inference, which refers to using trained AI models to respond to queries and perform tasks, may become increasingly important as AI adoption expands.

The distinction matters. Training spending could slow as companies become more selective about developing new models, without necessarily signalling the end of the broader AI cycle. If inference demand continues growing, the need for computing infrastructure may remain substantial.

One metric worth watching is GPU spending rather than the price of individual AI tokens. Token prices may fall as models become more efficient, but the number of tokens being generated could increase dramatically. Falling prices alone therefore don't necessarily indicate declining demand.

The more meaningful warning sign would be a sustained decline in actual AI usage and the associated spending on computing capacity.

This is also where today's AI boom differs from the late-1990s dot-com cycle. While parts of the technology boom involved companies raising money and spending it within the same ecosystem without developing sustainable business models, current AI investment is largely directed towards building infrastructure that serves real users.

That doesn't mean today's investments will all be profitable. But the underlying question is whether the broader economy continues generating demand for AI services, rather than whether investment is simply circulating between companies.

Valuations also look different. Nvidia is trading at roughly 10 times next year's earnings and seven times the following year's earnings, compared with around 18 times for the broader market, according to the figures discussed. The debate is therefore less about an extreme valuation multiple and more about whether its earnings can be sustained.

Investment Takeaways

The AI boom doesn't need Nvidia to maintain its current market share or margins to remain viable. The broader market could continue expanding even as competition intensifies, and Nvidia may still generate substantial revenue growth despite becoming less dominant.

Investors should distinguish between risks to Nvidia and risks to the AI cycle as a whole. Leadership changes, margin compression, export restrictions and competitive breakthroughs could affect Nvidia's earnings without necessarily ending AI infrastructure investment.

The more important indicators to monitor are actual AI usage, the sustainability of inference demand, access to financing for major AI infrastructure projects and whether political restrictions materially delay construction.

A portfolio holding both Nvidia and hyperscalers may also provide exposure to different outcomes. Nvidia benefits from selling computing infrastructure, while hyperscalers may benefit if chip costs fall or infrastructure becomes more readily available.

Ultimately, the AI cycle should be judged by the durability of demand and the economics of the businesses involved, rather than by a single earnings announcement, political headline or changing price of AI services.