Intel, Broadcom CEOs Say Capital, Not Chips, Is Emerging as AI's Biggest Constraint

Hock Tan said Broadcom is working through wafer and memory shortages and predicted that the shortage would ease by 2028 or 2029. He added that the imbalance between demand for computing capacity and the industry's ability to meet it could persist into 2030. Intel disclosed last month that it has capacity to meet only 50% of orders for its central processing units, which are used by AI agents to run tools, SiliconANGLE reported.

Capital, however, may be a harder problem. Hock Tan said access to capital "might be the biggest constraint" facing the sector. Lip-Bu Tan echoed that view, saying that if there is a challenge ahead, "it's capital and the cost of capital," and that over the longer run, "people want to see the ROI."

Hock Tan also discussed the economics of AI model developers. He said OpenAI is still losing money because it spends heavily on training, but the company's potential revenue growth from frontier generative AI models is significant. Broadcom helped OpenAI design its first custom inference chip earlier this year. Hock Tan said AI labs should acquire more customers that can realize a return on inference spending. He compared some enterprise AI revenue to a child buying arcade tokens: "Your parents give you $10 of tokens, you finish it in 10 minutes and you haven't learned anything."

He also commented on AI stock volatility, saying many investors "set very extreme expectations only to find out that it's almost like they are in a casino." The Financial Times reported last month that Anthropic's gross margins exceed 80% when excluding model training costs and revenue shared with go-to-market partners, and that the company is on track to generate positive adjusted operating income for the second quarter in a row, according to SiliconANGLE. Reuters reported that Hock Tan said Anthropic and OpenAI have shown that deploying the latest models as AI personal assistants can generate substantial revenue.

Lip-Bu Tan said open-source language models from Chinese labs are proving "good enough" for many budget-conscious users, adding that he is "a strong believer of open source" because it is the best way to drive efficiency. He also predicted that quantum computing could become mainstream within three to five years, according to Reuters. Intel is developing quantum processors that can be made with standard chipmaking equipment, and in January Argonne National Laboratory deployed a 12-qubit computer developed by the company. Intel has said it hopes to develop quantum systems with hundreds of qubits.

Even with capital concerns, investors continue to fund AI and robotics startups. Bloomberg reported that DeepSeek plans to lead a seed round of several tens of millions of dollars in robot AI startup Winverse AI, which would value the company at about $400 million post-money. Meituan's Long-Z Investments and other investors have made commitments, according to the report. Winverse AI was founded in 2026 by Hu Han, formerly chief scientist of Tencent's vision-language and embodied foundation model team, and develops AI for robots rather than general chat applications. The deal has not been finalized and terms may change. Hu Han, DeepSeek and Long-Z did not respond to Bloomberg's requests for comment.

Wuwen Xinqiong, a Chinese AI infrastructure company, announced on Oct. 12 that it completed a 2.3 billion yuan Series C round, bringing total funding since its founding in 2023 to 4.3 billion yuan. The company, spun out of technology transfers from Tsinghua University and Shanghai Jiao Tong University, organizes computing resources across chip types and regions, provides large-model inference services and offers industry solutions to enterprises. It said its platform reaches 37,000 petaflops of compute, covers 16 mainstream chip types and supports mixed training on domestic and imported chips. Its enterprise infrastructure agents include platform management, intelligent computing operations and the KernelMind operator generation tool. It works with model companies including Zhipu and Moonshot, with industry solutions in gaming and entertainment, healthcare, legal and energy.

Wang Xingxing, founder of Unitree Robotics, said on Oct. 12 at the opening of the 26th China International Industry Fair that his biggest mistake over the past decade was not believing in AI enough. He said his confidence in AI is increasing each year and that global AI technology updates every one or two months. After GPT-6 was released, his team tested it on different robots with good results, changing how embodied intelligence practitioners view technical progress. He predicted the tipping point for embodied intelligence could arrive quickly, possibly within one or two years. Before large-scale factory deployment, robots still need to improve generalization and efficiency, though he expects these problems to be solved in the coming years. In terms of deployment order, he said commercial services may advance faster than industrial applications, with home scenarios later.

Editor's Summary

Intel and Broadcom CEOs said at a Singapore financial conference that capital access and cost of capital are emerging as major constraints on AI expansion, even as chip shortages ease and demand remains strong. Hock Tan highlighted the costly training economics of AI labs and investor volatility, while Lip-Bu Tan pointed to open-source models and quantum computing. At the same time, funding continues to flow into AI infrastructure and robotics, with DeepSeek reportedly backing Winverse AI and Wuwen Xinqiong closing a 2.3 billion yuan round.