Intel CEO says Chinese AI models are cheap and spreading fast

The latest:

Chinese frontier AI models are spreading quickly because their token costs run lower than rivals’, Intel Chief Executive Lip-Bu Tan said Monday in Singapore. Speaking at a panel organised by Singapore lender DBS, Tan described the performance of models including DeepSeek, Kimi and GLM as good enough, and said he expected markets rather than vendors to sort out which systems suit which tasks.

Details:

• The cost argument: Tan tied the speed of adoption directly to price, saying the token costs of Chinese frontier models have been lower. Token costs are the per-unit charge for processing text through a model, and they set the running cost of deploying AI at scale across enterprise workloads.
• The models named: He singled out DeepSeek, Kimi and GLM, three Chinese model families, as performing well enough. Tan did not rank them against specific US competitors, nor did he give adoption figures or name customers deploying them.
• The open-source position: Tan said he is a strong believer in open source, calling it “the best way to drive efficiency.” The remark places the head of one of the largest US chipmakers alongside a development model that Chinese labs have used aggressively to distribute their systems.
• Who decides: Rather than predicting a winner, Tan said markets will determine which models benefit different use cases the most. He did not say whether Intel is building products around any particular model family.
• The quantum timeline: Tan said quantum computing could become mainstream within the next three to five years. He gave no detail on Intel’s own quantum roadmap, and did not define what mainstream adoption would look like in commercial terms.
• Software and infrastructure: He added that software will play a role in driving the efficient use of some of the infrastructure that the shift will require. Tan did not specify which infrastructure categories he meant or quantify the efficiency gains available.
• The venue: The comments came at a panel in Singapore organised by DBS, a regional lender, rather than at an Intel product event. No new company announcement accompanied the remarks.

Background:

Chinese labs including DeepSeek have released frontier-class models under open or permissive licences, competing on price per token rather than on exclusive access. That approach contrasts with the closed, subscription-based distribution favoured by several leading US developers.

Between the lines:

Tan’s framing matters because of who is making it: an Intel chief executive crediting Chinese models with being cheap enough and good enough, while backing open source as the most efficient path. Combined with his view that markets, not vendors, will pick winners, the remarks read as a bet on commoditised model access rather than on any single dominant system.

What’s next

Watch whether Intel ties product or software announcements to open-source model support, and whether Tan’s three-to-five-year quantum timeline is repeated at the company’s next earnings call or investor briefing.