Intel chief executive Lip-Bu Tan gave Nikkei a brief assessment of how far China’s chip capabilities trail those of the United States. Asked whether China could catch up within five to 10 years, he said: “Today, I think it’s not that far behind.” Nikkei Asia published the exclusive interview, held in Tokyo on 7 October, on 16 October 2026.
Tan has also spoken about China’s AI models. At a panel organised by DBS in Singapore on 12 October, he said the token costs of Chinese frontier models have been lower and adoption has been “very fast”, Reuters reported. We checked what those models cost today against the top models from Anthropic and OpenAI.
Business Pill · OPEN WEIGHTS
A short explainer on open-weight models: renting a model versus running it yourself. It teaches the general idea only and says nothing about any company in this story.
The key insight: As we read it, Tan is making two separate points: on chips, he told Nikkei China is “not that far behind”; on models, he said at the DBS panel that Chinese models are “good enough” with lower token costs, and the labs’ own price lists show that gap.
Nikkei Asia describes Tan as relatively cautious when asked about the competitive threat from China’s chip industry. Whether a company is from the U.S. or China, he said, “they all have different strengths and weaknesses, and then I think more important is what is best for the consumer, what is best for the enterprise to use.”
Most of the interview was about Intel. Tan said the Trump administration does not have “any policy” to use its minority stake to interfere in Intel’s strategy. “They don’t have any policy. They don’t have any influence on the governance of the company,” he told Nikkei.
He still wants Washington’s backing for contract chipmaking. “From a foundry’s point of view, Intel had to look at the longer-term horizon, and it’s always very important to have very strong government involvement supporting,” he said. He added that the business could become a growth driver “if we do it right,” but “it takes time.”
Output price per million tokens for the top general model on each lab’s own API pricing page, standard rate, short prompts, read 19 October 2026: Claude Fable 5.1 (Anthropic) $50, GPT-6 Astra (OpenAI) $50, Kimi K3 (Moonshot AI) $15, GLM-5.3 (Z.ai) $4.40, DeepSeek V4 Pro (DeepSeek) $3.96 at its peak rate, half that off-peak. The multiples are our arithmetic. List prices only; tokenizers and quality differ.
Nikkei Asia’s own post of the interview, 19 October 2026. The article it links is the one we read in full: Lip-Bu Tan on the US government’s stake, Intel’s foundry plans and China, where he said: “Today, I think it’s not that far behind.”
Intelโs CEO says Chinaโs semiconductor technology is not far behind the US.
In an interview with Nikkei. Asked if China could catch the US in five to ten years, he said it is not that far behind even now.
He rated the research strength of Tsinghua University and Peking University as on par with MIT and Stanford.
On AI he named Moonshot AIโs Kimi, Zhipu AIโs GLM, and DeepSeek, saying they are not technically behind Anthropic or OpenAI.
He also noted Chinese AI models are much cheaper to use. $INTC
— Semiconductor Insider (@semiconductorsx) Oct 19 01:43
A summary of the interview posted by @semiconductorsx, 19 October 2026. It includes the Tsinghua, Peking and AI-model remarks that Chosun Ilbo attributes to Nikkei’s Japanese-language version; we have not read that version, so those lines are the poster’s and Chosun’s account, not ours.
On China’s AI Models
At the DBS panel, Tan said the performance of Chinese models such as DeepSeek, Kimi and GLM is “good enough”, according to Reuters. “I’m a strong believer of open source, that’s the best way to drive efficiency,” he said, adding that markets will decide which models suit which uses.
He went further in Nikkei’s Japanese-language version of the interview, according to Korea’s Chosun Ilbo. In its English report, Chosun says Tan rated research at Tsinghua University and Peking University on par with MIT and Stanford. Naming Kimi, GLM and DeepSeek, he said: “They are not technologically behind U.S. firms like Anthropic or OpenAI.”
Chosun also reports that Tan said Chinese AI models are significantly cheaper to use than their American counterparts. We could not read the Japanese original, and Chosun notes its English text was machine-translated.
What the Models Cost
We read each lab’s own API pricing page on 19 October 2026 and took the top general model listed, at the standard rate for short prompts. Anthropic lists Claude Fable 5.1 at $10 per million input tokens and $50 per million output tokens. OpenAI lists GPT-6 Astra at the same $10 and $50.
The Chinese labs Tan named list lower prices. Moonshot AI charges $3 and $15 for Kimi K3. Z.ai charges $1.40 and $4.40 for GLM-5.3. DeepSeek charges $1.32 and $3.96 for DeepSeek V4 Pro at its peak rate, and half that off-peak.
Our arithmetic: a $50 output price is 12.6 times DeepSeek V4 Pro’s $3.96 and 3.3 times Kimi K3’s $15. These are list prices only. Anthropic says its newer tokenizer produces about 30% more tokens for the same text, so a price per token is not a like-for-like cost per task. For one task-level test, see our 15 October report on Sakana’s AI Scientist.
Intel’s Own Position
Nikkei Asia reports that the U.S. government converted federal funding into an $8.9 billion investment last year, for a 9.9% stake later diluted to around 8.2%. Intel has also received around $7 billion from SoftBank Group and Nvidia, in exchange for equity.
Tan said both investors could “make a lot of money” and potentially earn “five to six times the return.” He said Intel was “80% done” with its restructuring, and identified AI-related advanced chip packaging and application-specific integrated circuits, or ASICs, as other potential areas of growth. Intel shares trade at around $100, roughly four times their level when he took over in March 2025, Nikkei Asia reports.
Update, 19 October: JPMorgan Lifts Its Intel Rating to Neutral
On 19 October JPMorgan upgraded Intel to neutral from underweight, CNBC reported in its roundup of the day’s analyst calls. CNBC quoted the bank: “We are upgrading INTC to Neutral from Underweight as the stock approaches our $85 price target where risk/reward is much more balanced as the stock has pulled back ~36% from its peak in June.”
The Structural Read
As we read it, the price gap is the measurable part of Tan’s claim: on the labs’ own pages, the top Chinese models he named list output prices of $3.96 to $15 per million tokens, against $50 for Claude Fable 5.1 and GPT-6 Astra.
As we read it, the quality part is the contested part: Tan called the Chinese models “good enough” at the DBS panel, and Chosun reports he told Nikkei they are not technologically behind Anthropic or OpenAI; list prices say nothing about that.
As we read it, his open-source remark ties the two together: “I’m a strong believer of open source, that’s the best way to drive efficiency,” he said, adding that markets will decide which models suit which uses.
Lip-Bu Tan, chief executive of Intel, to Nikkei Asia, asked whether China could catch up within five to 10 years
PRICE IS THE VISIBLE GAP On each lab’s own pricing page, read 19 October 2026, the top general models from Anthropic and OpenAI list $50 per million output tokens; Kimi K3 lists $15, GLM-5.3 $4.40 and DeepSeek V4 Pro $3.96 at peak.
QUALITY IS THE CLAIM TO TEST Tan called the Chinese models “good enough”, according to Reuters; a per-token price does not measure quality or the tokens a task needs.
INTEL’S OWN POSITION Nikkei Asia reports Intel is rebuilding its foundry business with a U.S. government stake of around 8.2% and around $7 billion from SoftBank Group and Nvidia.
The framework puts the pattern this way: “Close the SCARCE layer (keep proprietary) + Open the ABUNDANT layer (commoditize).”
As we read it, the price lists show which layer each side is pricing as scarce: the US labs charge $50 per million output tokens for their top models, while DeepSeek, Moonshot AI and Z.ai list theirs at $3.96 to $15. The framework was published in March 2026 and does not discuss these labs.
What Is Not Established
We have not read the Japanese-language Nikkei interview; the Tsinghua, Peking and Anthropic-or-OpenAI remarks come from Chosun Ilbo’s machine-translated English report. The English Nikkei Asia article we read in full does not contain them.
The price comparison uses list prices for one model per lab. It does not measure quality, speed or the number of tokens each modelneeds for a task, and labs change prices often. None of the reports we read cite specific prices from Tan.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This piece is part of this week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
Intel’s chief executive says China is “not that far behind” in chips, and that Chinese models are good enough and cheaper. On the labs’ own price lists, the top models from Moonshot AI, Z.ai and DeepSeek cost between $3.96 and $15 per million output tokens, against $50 for the top models from Anthropic and OpenAI. For more on DeepSeek’s pricing, see its search API launch.
A note on sourcing. This piece is based on Nikkei Asia’s exclusive interview with Lip-Bu Tan (published 16 October 2026), read in full; Reuters’ report of his 12 October DBS panel remarks, read in full through a syndicated copy; and Chosun Ilbo’s English report of 19 October on Nikkei’s Japanese-language interview, which we did not read. Prices come from each lab’s own API pricing page, read on 19 October 2026; the multiples are our arithmetic. Nothing here is a forecast, and nothing here is financial or investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
