Nvidia's China Woes Deepen As TikTok Parent ByteDance Eyes AI Chips From Iluvatar CoreX, Baidu: Report

ByteDance is reportedly in talks to buy AI chips from Iluvatar CoreX and Baidu, with Iluvatar projected to supply at least 50,000 chips this year for AI inference tied to Doubao chatbot expansion. If completed, Iluvatar would become ByteDance’s third major domestic GPU supplier after Huawei and Cambricon, reinforcing the shift toward Chinese chip vendors amid U.S. export restrictions. The article also notes ByteDance’s separate Malaysia AI buildout and Nvidia’s efforts to market new AI silicon to Chinese customers.

Analysis

The important read-through is not just incremental demand for Chinese accelerators; it is that inference is becoming the first monetizable workload where domestic silicon can win on cost, supply assurance, and power efficiency. That matters because inference demand scales with user growth, so the next leg of capex is likely to be less cyclical than training and more resilient to export-control pressure. For BIDU, this supports a longer runway for AI product monetization and reduces the risk that its chatbot traffic growth is bottlenecked by foreign supply constraints.

For NVDA, the near-term hit is less about lost unit sales and more about margin dilution from forced product customization and a shrinking share of a strategically important ecosystem. Once a local supplier wins sockets for inference, switching costs rise quickly because software stacks, deployment tooling, and procurement relationships harden over 2-4 quarters. The more subtle second-order effect is that every domestic chip shipped into China reinforces an alternative vendor ecosystem, which compounds into lower future attach rates for networking, CPU, and software layers.

The overseas build-out is a tell that Chinese buyers are treating offshore capacity as a pressure valve, not a replacement, which keeps global demand for advanced AI infrastructure elevated even if China onshore mix shifts away from NVDA. The contrarian point is that the market may be underestimating how much of NVDA’s China exposure is now being repriced from a growth story into a policy option value trade; that typically compresses multiples before it hits revenue. The risk to the bearish NVDA view is that product adaptation like a China-targeted CPU can partially offset share loss if it preserves developer lock-in and keeps ASPs from collapsing.

Key Decisions for Investors

• Long BIDU 3-6 month horizon: AI infrastructure optionality plus inference-scale demand should support multiple expansion if execution remains visible; use pullbacks to build, with downside protected by the broader search/ad cash flow base.
• Short NVDA vs long semicap basket in China-sensitive names over 1-3 months: the asymmetric risk is margin compression and share leakage in China, while the offset from global AI demand limits outright downside, making this a cleaner relative-value expression.
• Buy NVDA downside hedges 2-4 months out: finance puts or put spreads around the next China-policy headline window, since the market may be slow to price the cumulative effect of domestic supplier substitution.
• Long domestic China AI supply chain beneficiaries on weakness, including Cambricon-related proxies where accessible, for a 6-12 month horizon; the trade works if inference demand continues compounding and procurement stays localized.
• Avoid chasing the headline on BIDU into strength; better to wait for confirmation that chip supply translates into active inference capacity utilization, because the first move is often a sentiment trade and the second move depends on monetization.