Exclusive-Prices of Nvidia’s B300 server at $1 million in China on US curbs, sources say

Nvidia B300 server prices in China have nearly doubled to about 7 million yuan ($1 million) each from roughly 4 million yuan late last year, driven by tighter U.S. export curbs and a crackdown on chip smuggling. The article also says U.S.-approved H200 shipments to China remain stalled, while demand from Chinese AI companies is strong and pushing rental prices as high as 190,000 yuan a month on one-year contracts. The news is supportive for pricing power and demand in AI hardware, but it also highlights supply constraints and regulatory risk in China.

Analysis

The market is no longer pricing Nvidia’s China exposure as a simple revenue line item; it is pricing a scarcity regime. When access to frontier inference hardware becomes constrained, the economic value migrates from the chip vendor to whoever can secure, assemble, and deploy compute at scale, which can temporarily benefit gray-market intermediaries, server integrators, and cloud-like rental pools more than the OEM itself. The bigger second-order effect is on Chinese AI operators: higher capex barriers should favor the most capitalized platforms and compress smaller model labs into tenancy/rental structures, accelerating consolidation in domestic AI.

For NVDA, the short-term signal is bullish on pricing power but not necessarily on clean China monetization. A tighter export environment can raise the shadow price of its best products, but it also increases substitution pressure toward domestic accelerators and raises the risk that China demand becomes more volatile, more indirect, and more politically fragile over the next 1-2 quarters. If enforcement keeps tightening, the market may eventually treat China as an option value story rather than a dependable growth leg, which is a negative multiple input even if headline scarcity keeps supporting sentiment.

BABA is the more interesting second-order beneficiary: the company can monetize compute demand through cloud and AI services without needing to hold restricted hardware directly, which reduces sanctions optics while capturing the rental premium embedded in scarcity. AMD is a relative loser because the narrative here reinforces that China buyers are willing to pay for the highest-end performance, but AMD still lacks the same must-have status in frontier inference and therefore gets less pricing lift from the constraint. The contrarian risk is that the market overestimates how durable the premium is; if the H200 approval issue resolves or domestic Chinese accelerators reach acceptable inference efficiency, the rental bubble can compress quickly.

The time horizon matters: in the next 2-6 weeks, the trade is mostly sentiment and supply tightness; over 3-9 months, it becomes a substitution and policy story. Any evidence of eased export conditions, renewed approved shipments, or a meaningful drop in token demand growth would reverse the scarcity thesis faster than expected. Conversely, continued enforcement plus strong Chinese AI usage would extend the premium, but that also increases the odds of a regulatory backlash against intermediaries and a sudden air-pocket in gray-market pricing.

Key Decisions for Investors

• Maintain a tactical long NVDA position for 2-6 weeks, but hedge with downside protection via out-of-the-money puts; the trade is on scarcity/margin optics, not durable China unit growth, so trim aggressively if policy headlines improve.
• Go long BABA vs short AMD for a 1-3 month pair trade: BABA can capture AI demand through cloud monetization and compute rentals, while AMD is less levered to the scarcity premium and more exposed to a normalization in China procurement.
• Consider a short-duration long on NVDA server-supply chain beneficiaries outside the chip itself, funded by partial short exposure in China-sensitive hardware names; the cleaner alpha is in distribution and infrastructure friction, not the OEM headline.
• Sell volatility in NVDA only after a confirmed pullback in gray-market pricing or an H200 shipment resolution; until then, realized volatility should stay elevated and short vol is unattractive on policy headline risk.
• For a risk-controlled expression of the domestic-China substitution theme, use a 3-6 month BABA call spread rather than outright shares; upside comes from captive AI demand and cloud rent economics, while downside is limited if the scarcity premium normalizes.