Nvidia’s Asian Supply Chain Costs Hit 90% as Nuclear Bet and Earnings Test Loom

Nvidia's Asian supplier dependency hits 90%, robotics and data-center chips compete for scarce TSMC capacity, while China revenue drops to zero and market cap falls to $4.82 trillion.

The numbers tell a stark story of dependency. Nvidia’s production costs tied to Asian suppliers have ballooned to 90%, up from 65% just a year ago. TSMC fabricates the chips, SK Hynix and Samsung supply the high-bandwidth memory, and Foxconn and Quanta handle server assembly. But this isn’t simply a tale of growth — it’s a structural shift that is reshaping the company’s entire hardware strategy.

The Robotik Squeeze

Nvidia’s push into Physical AI — hardware for robotics and autonomous vehicles — is tightening the bottleneck. The new Jetson Thor platform for robotics relies on the same 3-nanometer manufacturing capacity at TSMC as the Blackwell data-center GPUs. Two product lines are now competing for scarce supply, driving up costs and creating shortages of LPDDR5X memory. The strain has already forced Nvidia to discontinue older Jetson processors.

The automotive SoC DRIVE AGX Thor, aimed at the vehicle segment, faces the same capacity constraints. LG is weighing a strategic partnership in the Physical AI ecosystem, while Boston Dynamics and Amazon Robotics are already onboard. But the resource conflict between data-center and robotics hardware is a problem that won’t resolve itself — and it will likely dominate the conversation when Nvidia reports next quarter.

China: Zero Revenue, Black Market Premium

CEO Jensen Huang has acknowledged what analysts long suspected: server sales to China have effectively dropped to zero. US export restrictions have blocked official shipments, and Huang believes the curbs have only accelerated China’s drive to build its own semiconductor industry. Yet on the Chinese secondary market, B300 AI servers still trade at prices reaching $1 million per unit — a stark reminder that demand hasn’t disappeared, just gone underground.

The Market Shuffle

Nvidia’s market capitalization has slipped below $5 trillion, currently sitting at $4.82 trillion. The stock trades around €169, roughly 7% below its 52-week high set in late April. The shares gained 14% in April, but that pales next to AMD’s 74% surge over the same period — a comparison that has investors questioning Nvidia’s valuation.

The competitive landscape is shifting. Alphabet’s Google Cloud revenue topped $20 billion, sending its stock up 10% and lifting its market cap above $4.6 trillion. Options traders now see a 53% probability that Alphabet overtakes Nvidia in market value by mid-May. Amazon is seeing strong demand for its own chips, and Google has started selling its custom processors directly to customers. A Wall Street Journal report that OpenAI missed internal revenue targets sent Nvidia’s stock down more than 6% in two days.

Despite the pressure, Nvidia’s profitability remains formidable. The gross margin stands at 71%, with an operating margin of 60%. There’s no sign of a demand slump yet.

A Nuclear Hedge

Away from the chip wars, Nvidia is securing its energy future. The company has partnered with Oklo and the Los Alamos National Laboratory to develop small modular nuclear reactors for data centers. Nvidia is using its own AI models to accelerate reactor design and improve the validation of nuclear fuels. It’s a vertical integration play — ensuring the power supply needed to keep its computing infrastructure running long-term.

The May 20 Reckoning

All eyes are on May 20, when Nvidia reports first-quarter results. Management has guided for revenue of roughly $78 billion, a figure that completely excludes the China data-center business. Analysts expect revenue growth of 79%. Historically, Nvidia’s stock has fallen after most recent earnings reports, with an average decline of 3%. The options market is pricing in a move of up to 10% by the end of May.

Memory costs are expected to consume about 30% of cloud providers’ data-center spending this year — a headwind that hits not just Nvidia but the entire AI hardware sector. How the company resolves the capacity tug-of-war between its data-center and robotics lines will be a key question when the numbers come out.