Our Take
Procurement moves are not product wins; this is supply-chain jostling under sanctions, not evidence that either company has solved the chip shortage.
Why it matters
ByteDance faces acute hardware constraints due to US restrictions on advanced chips. Domestic sourcing options remain limited and unproven at scale, making any viable supplier a strategic asset in the short term.
Do this week
Enterprise AI teams: audit your inference infrastructure dependencies now before supply constraints force costly mid-project pivots.
ByteDance explores domestic chip options under export pressure
ByteDance is in preliminary talks with Iluvatar CoreX, a Chinese AI chip manufacturer, to purchase processors for its AI operations, according to sources cited by Reuters. The discussions reflect ByteDance's need to secure hardware as US export controls restrict access to advanced semiconductors from Nvidia and other Western suppliers.
Iluvatar CoreX, based in Beijing, designs and manufactures AI accelerators intended for Chinese enterprises and research institutions. The company has positioned itself as an alternative to foreign chips, though it operates in a crowded field of domestic competitors including Huawei, Baidu, and Alibaba-backed efforts.
No deal terms, timelines, or chip specifications were disclosed in available reporting. The talks are described as ongoing and preliminary.
Supply scarcity is reshaping who builds what
ByteDance's exploration of Iluvatar CoreX chips underscores a structural shift in AI infrastructure procurement. US sanctions targeting advanced semiconductor exports to China have forced major Chinese tech firms to either reduce AI workload ambitions, source from a narrowing pool of domestic suppliers, or both.
The practical constraint is real. ByteDance runs TikTok's recommendation engine, Douyin's content systems, and other compute-intensive services. Nvidia H100 and H200 GPUs, which dominate large-model inference, are now subject to export licensing. Alternatives remain immature. Domestic chips from suppliers like Iluvatar CoreX exist but lack published performance benchmarks, production scale verification, or long-term reliability records at ByteDance's workload volume.
This is not ByteDance betting on Iluvatar CoreX because it is faster or cheaper. It is ByteDance rationing access to forbidden hardware and exploring the least-bad fallback.
Plan for fragmented supply chains
If you operate AI inference infrastructure in restricted jurisdictions or depend on Chinese suppliers, treat domestic chip sourcing as a real constraint, not a temporary regulatory inconvenience. Verify publicly available benchmarks before committing to procurement. Test reference implementations at realistic scale. Do not assume a supplier's domestic market success translates to your workload or reliability needs. Build inference redundancy across chip architectures sooner rather than waiting for a crisis to force it.