DeepSeek’s Huawei Chip Move: China’s AI Workaround in Action
DeepSeek turns to Huawei chips to bypass US export controls—what this means for AI training costs and global supply chains.
The story of the day
DeepSeek is accelerating its pivot to Huawei’s training chips, with deliveries expected as early as Q4 2026, per The Information. The move underscores a growing trend: Chinese AI labs are bypassing US export controls by adopting domestic silicon for model training—a far more demanding workload than inference. For builders, this means two things: (1) Huawei’s chips are now viable for frontier models, and (2) the cost of compliance (or evasion) is being baked into China’s AI stack. The $7.5B funding round DeepSeek is finalizing will likely fund more of these workarounds. Expect margin pressure on Nvidia’s high-end offerings in the region.
New tools & launches
Bitcoin-rs — Open-source Bitcoin full node in Rust, using AI for implementation verification against libbitcoinkernel. [HN]
Open Source AI Employees — MIT-licensed folder of plain-text routines for 8 business roles (SEO, sales, etc.), agent-agnostic. [HN]
Models & research
Thomson-1.0-Small — Proprietary LLM from Thomson Reuters, now listed on Featherless.ai. Target use case: legal/financial text analysis. No benchmarks provided. [HN]
Industry pulse
Nvidia CEO Jensen Huang dismissed existential AI risk by 2030 as "0% chance," per The Guardian. The remark contrasts with Anthropic’s recent safety warnings and underscores the divide between chipmakers (focused on infrastructure) and model builders (grappling with alignment). Builders note: the comment says more about Nvidia’s incentive to sell GPUs than about model risk.
Stat of the day
$7.5B — DeepSeek’s target for its second funding round, per The Information. The capital will fund Huawei chip adoption, a clear signal that China’s AI ecosystem is prioritizing supply chain resilience over cost efficiency.
Get the full analysis — subscribe to Gradient Pro.
