Jensen Huang keeps showing up in China. That detail alone tells you more about Nvidia’s strategy than any earnings slide. A CEO of a company this size does not spend travel time on a market he has written off, and the Reuters report that Nvidia appears close to approval for China chip sales pushed the stock up on Friday. The market read it as a door swinging open.
I read it differently. The door may open. The room behind it has been rearranged.
Two numbers that do not fit together
Hold two facts side by side. First, China’s potential reopening to Nvidia’s AI market could lift NVDA, and analysts stay optimistic about long-term growth. Second, Nvidia’s share of AI chips in China is projected to fall from 40% to 8% by 2026, with Huawei’s scaling cited as the cause.
Both can be true. That is the part worth sitting with. Regulatory access and market share are different variables, and the past two years decoupled them. Access was restricted, so Chinese labs did what engineering organizations always do under constraint: they rebuilt around the constraint. Huawei did not need to match Nvidia on raw silicon performance. It needed to be good enough, available, and not subject to an export approval process.
So Monday’s price action, whichever direction it goes, is a bet on access. The 8% figure is a statement about substitution that already happened.
Why the software stack matters more than the chip here
From where I sit as someone who spends most of her time on agent architecture rather than trading screens, the interesting question is not whether Nvidia can ship GPUs into China again. It is whether the surrounding software layer still pulls developers in once alternatives exist.
Nvidia’s actual moat was never only the die. It was leadership in GPUs plus the software stack required to build AI applications. That stack is sticky because it is where accumulated engineering habit lives: kernels people already tuned, training pipelines people already debugged at 2am, inference paths with known failure modes. Switching costs in AI infrastructure are mostly measured in institutional memory, not dollars.
But institutional memory has a half-life. If a large developer population spends two years writing against a different toolchain, some of that habit transfers permanently. Reopening access does not automatically reverse it. You get the customers who never left and the ones who tolerated the alternative without committing. You do not automatically get back the teams who rebuilt their stack and shipped on it.
This is the part I think gets underweighted in the stock conversation. Market access is a binary that flips on a regulatory decision. Developer preference is a slow-moving distribution that flips over multiple product cycles.
What the calendar actually gives us
There are dated events ahead, which is more than most of this discussion offers:
- Nvidia earnings on May 20, 2026
- A GTC keynote from Jensen Huang in Taipei on June 1
- The Ineffable Intelligence news involving the London-based AI lab from Wednesday
Earnings will give actual revenue composition. The Taipei keynote will give product direction. Between them, you learn more than any weekend of speculation produces. Anyone telling you with confidence how NVDA opens Monday is describing sentiment, not information.
The demand picture is bigger than one country
TSMC, which manufactures Nvidia’s chips, now sees the global semiconductor market exceeding $1.5 trillion by 2030, up from a previous estimate of $1 trillion. That revision is the most substantive signal in the current set of facts. It is a supply-chain participant with direct order visibility raising its own long-range forecast by 50%. Foxconn also reported strong earnings.
Which reframes the China question. If total demand is growing that fast, losing share in one large market is painful but not structural. The discussion around whether Nvidia reaches a $10 trillion valuation by 2030 leans on high cloud demand and stack leadership, not on any single region. China matters as upside, and as a signal about how much of the world’s compute buildout Nvidia gets to participate in directly.
What I would actually watch
If you care about where agent infrastructure goes rather than where the ticker closes, three things deserve attention over the next quarter.
Whether approval, if it comes, covers current-generation parts or throttled variants. The performance tier determines whether returning hardware serves frontier training or only inference and smaller workloads. Those are different businesses with different margins.
Whether Chinese labs publish work that is clearly architected around domestic silicon. Papers and open releases reveal hardware assumptions faster than any market report. Attention implementations and memory layouts tell you what people are optimizing for.
Whether the 8% projection gets revised. Forecasts made under one policy regime rarely survive a change in that regime unexamined.
Monday will be noisy. The 40-to-8 trajectory and the $1.5 trillion revision are the two lines actually worth tracking, and neither resolves this week.
đź•’ Published: