Our Short Thesis For Nvidia, The Market Isn’t Pricing in the Downside Risk

When Perfection Is Already Priced In

Why Nvidia may now have more downside than upside comes down to one thing.

After eight months of strong results and relentless earnings momentum, the stock has barely moved even with repeated revenue and earnings beats.

That tells us the market is no longer reacting to strong execution alone. Nvidia is still delivering, but investors already expect that. The bar is no longer high, it is extreme.

Even today, Nvidia announced cumulative revenue of US$1T from its Blackwell and Rubin chips. That is an enormous number by any standard. But if the stock does not move on news like that, it usually means the market has already priced in a huge amount of future success.

That is the real question. Why is the market not budging? In our view, it is because Nvidia is no longer being valued on whether growth is strong. It is being valued on whether growth can keep beating some of the most aggressive expectations the market has ever placed on a company.

At this point, the risk is not the business. The risk is the valuation.

What are the Best ASX Stocks to invest in right now?

Check our buy/sell tips

A quick snapshot of the NVIDIA story

Nvidia has been a core position in our portfolio since 2021, when it was still a sub-US$600B market cap company largely known for its leadership in gaming GPUs and PC hardware.

What made the story so powerful was that Nvidia was already starting to transition beyond gaming into something much bigger. CUDA, its software platform, gave the company the ability to tailor its GPUs to a wide range of engineering and compute-heavy tasks, which became the foundation for its dominance in AI and accelerated computing.

Around that time, Nvidia had only just passed US$1B in revenue, but the direction of travel was becoming clear. The company was moving from being seen as a hardware business into one of the most transformational technology platforms in the market.

In 2021 and 2022, Nvidia released Hopper, which at the time was its most powerful chip. Since then, even Hopper has been eclipsed by the scale and capability of Blackwell and Rubin, which shows just how quickly the company has continued to push the frontier.

Why AI Chips Are So Expensive Right Now

The transistors used in today’s next-generation AI chips are now only a few atoms wide. That matters because Moore’s Law, the idea first observed in the 1960s that chip performance would keep improving as transistors got smaller, has broadly held true up until the late 2020s. We are now reaching the point where further shrinking is becoming much harder, creating major engineering challenges and pushing semiconductor capex to increasingly unsustainable levels. That is a big reason why we have seen Mag 7 capex rise so sharply.

Each new generation of chips is becoming exponentially more expensive to develop and manufacture than the one before it. Those costs ultimately get passed through to customers.

Nvidia’s AI GPUs are also still relatively early in their production lifecycle. Normally, as cumulative volume rises, unit costs begin to fall. But Nvidia has remained capacity constrained, which means that cost-down curve has been slower than the demand curve.

Because only a small number of foundries, particularly TSMC, can manufacture Nvidia’s most advanced chips, and because demand continues to exceed supply, Nvidia has been able to charge premium pricing. That is why the economics are so powerful. The H100 SXM costs roughly US$3,320 to manufacture but sells for multiples of that, which helps explain why Nvidia’s gross margins of 85% to 88% sit so far above AMD’s 65% to 68% and Intel’s 58%.

The Forces That Could Drive Prices Down and Compress Nvidia’s Margins Post-2026

The single biggest structural threat to Nvidia is that its largest customers, the hyperscalers, are now designing and building their own chips. That matters because the biggest buyers of Nvidia hardware are no longer just customers, they are increasingly becoming competitors.

JPMorgan expects custom chips developed by companies like Google and Amazon to account for 45% of the AI chip market by 2028. The strategic reason is clear. Hyperscalers do not want to remain dependent on a single supplier that effectively controls the market. Nvidia is now in the position where it has to compete with its own customers.

The cost of Nvidia’s AI chips is a major part of that shift. For cloud providers, renting out Nvidia GPUs can be highly lucrative in demand terms, but the margin profile is not always as attractive when the hardware is so expensive. That creates a strong incentive to develop custom silicon that is cheaper, more tailored to internal workloads, and more profitable over time.

What we know is that hyperscaler capex is exploding and is expected to reach US$600B, with Amazon among the largest spenders. What we are seeing now is the early stage of that buildout, but the bigger risk for Nvidia sits later in the decade. As hyperscalers improve their internal chip design capabilities, the need to rely on outsourced chip suppliers could start to fall.

The Bull Case That Complicates This Thesis

What would break this thesis comes down to one core idea, even if hyperscalers reduce their reliance on Nvidia, the total addressable market may still be expanding so quickly that Nvidia can keep growing anyway.

That is the real counterpoint. Nvidia may lose some share of hyperscaler capex over time, but if AI infrastructure, inference, robotics, autonomous systems, enterprise compute, and edge applications all continue scaling, the company could still access a much larger pool of demand across multiple markets.

Nvidia also still holds a dominant position, with an estimated 70% to 95% market share, supported by the CUDA ecosystem it has built over nearly 20 years. That includes more than 4 million developers, over 3,000 optimised applications, and deep integration across every major AI framework.

That is more than just switching costs. It is a massive base of accumulated human expertise and software investment that has taken years to build. Even if customers want to reduce dependence on Nvidia, replacing that ecosystem is far more difficult than simply swapping out a chip.

The investor’s takeaway for NVDA

The takeaway is that in the near term, if earnings margins compress even slightly, the market could react badly and push the share price lower.

At the same time, Nvidia’s addressable market is now so large that margin pressure does not automatically break the long-term story. If revenue continues to expand strongly, that can help offset some of the impact from lower gross margins.

There is also the added upside of new growth markets such as Omniverse and robotics. If Nvidia can translate its AI leadership into those areas, it could open another wave of growth that helps support the business beyond the current chip cycle.

So for us, the more likely outcome is that any operational margin compression in the near term creates volatility, but over the longer run Nvidia can still remain a very strong growth story.