Intel CEO Lip-Bu Tan expects AI demand to stay strong for another two to three years. But at the DBS Asia Future Forward Conference in Singapore on October 12, he warned that financing costs and uncertain returns remain key challenges. His message: continued appetite for computing does not settle whether the spending pays off.
Tan spoke on a panel with Broadcom CEO Hock Tan, who placed financing ahead of familiar physical constraints. The Broadcom chief said capital could become a bigger obstacle to AI expansion than shortages of semiconductors, memory chips or electricity, even while demand for computing infrastructure exceeds supply.
Lip-Bu Tan’s warning covered both the ability to fund expansion and the eventual value of that investment. He identified capital and its cost as near-term challenges, then pointed to the longer-term requirement to show a return on investment: what companies get back from the money they spend.
If there’s any challenge in the front, it’s capital and the cost of capital, and then the other part is in the long run, people want to see the ROI
The Intel chief was not forecasting an imminent collapse in activity. He cited growing adoption of AI applications and continued demand for computing power as reasons for his two-to-three-year outlook. That confidence sat alongside his financial warning, rather than replacing it.
Hock Tan described an industry that has repeatedly encountered supply bottlenecks over the past two years and then worked through them. He said today’s memory shortages might not persist a year from now. Wafer shortages could ease by 2028 or 2029, while new infrastructure could improve electricity availability.
Those improvements would not necessarily close the overall gap between computing demand and supply. The Broadcom chief said that imbalance could last through 2029 and potentially into 2030. His forecast allowed for individual shortages to ease while the wider infrastructure buildout continued to struggle to meet demand.
There was also a limit to that confidence. According to Wolf Street’s account, which cited CNBC’s panel reporting, Hock Tan said computing demand could slow if advanced AI models disappoint or gains in their capabilities begin to level off. Continued expansion was a forecast, not an unconditional promise.
Lip-Bu Tan also pointed to competition over the price of using AI. He said Chinese frontier models had lower token costs—the usage costs he highlighted—and were being adopted very quickly. He named DeepSeek, Kimi and GLM as models already sufficient for some tasks.
His assessment was use-specific: those models were “good enough” for some uses, not a declaration that they lead across every task. Tan said the market would ultimately determine which models fit different applications. He also endorsed open source as a way to drive efficiency.
The question of value also reached the businesses selling AI. Hock Tan argued that revenue growth alone did not establish a sustainable, profitable business model. In discussing OpenAI, he pointed to the costs of AI and infrastructure investment as reasons not to treat sales growth as proof of profitability.
For Lip-Bu Tan, the next test was what companies could demonstrate from adoption. He said meaningful returns would need to come through improved efficiency and new business models. That sets a different measure from demand for chips or computing capacity: whether the organizations using AI can turn their spending into business value.
Reader comments
Newest comments first. Replies stay oldest first.