Intel $INTC CEO Lip-Bu Tan warns that the AI computing power shortage will continue through 2029, but funding may be what truly determines the fate of this trillion-dollar industry.
On October 12, two semiconductor industry heavyweights took the stage together at the DBS Asian Insights Summit in Singapore.
Intel $INTC CEO Lip-Bu Tan and Broadcom $AVGO CEO Hock Tan.
Lip-Bu Tan’s remarks sent a signal that Wall Street should take seriously:
Demand for AI computing power may remain strong at least through 2029, but what will determine how far this super-investment cycle can go is no longer just chip supply—it’s the cost of capital and investment returns.
He discussed five key trends, each of which could influence the direction of AI investment over the next few years.
First, the shortage of AI computing capacity could persist through 2029.
Lip-Bu Tan expects the shortage of AI computing resources to last at least until 2029, with no significant slowdown in the industry likely over the next two or three years.
Why?
Because the bottlenecks facing AI infrastructure keep shifting.
At first, there was a shortage of $NVDA GPUs.
Then came tight HBM (high-bandwidth memory) capacity from suppliers such as $MU and Samsung.
Later, data centers began running up against constraints from power supply, cooling systems, and the time needed to build out the power grid.
Now, access to financing and the cost of capital have become new constraints.
This shows that growth in the AI industry can't be achieved simply by producing more chips; that won't solve every problem.
Second, financing could replace chips as the biggest bottleneck for AI.
Lip-Bu Tan noted that the cost of capital is rising, and the market is increasingly focused on whether massive AI investments can generate sufficient returns.
This may be the most important investment signal in the entire discussion.
Tech giants such as $MSFT, $AMZN, $GOOGL, and $META are investing enormous sums in building AI infrastructure.
But AI data centers aren't traditional assets that can be used indefinitely once they're built.
GPUs become outdated, servers need maintenance, power must be supplied continuously, and model training and inference keep consuming computing resources.
If revenue from AI applications fails to keep pace with capital expenditure, companies' free cash flow could come under pressure.
More importantly, some AI infrastructure projects rely on debt financing.
When financing costs rise, projects that once seemed to offer attractive returns can suddenly become uneconomical.
So in the future, investors can't just look at how much AI capital expenditure is growing; they also need to watch how much cash flow these investments ultimately generate.
Third, the AI industry can't rely forever on the people selling the shovels to make money.
Lip-Bu Tan offered a very important assessment:
The most direct beneficiaries in the AI supply chain today are mainly chip, memory, server, and other infrastructure providers.
Including $NVDA, $AVGO, $INTC, and Samsung.
But if infrastructure providers are the only ones in the industry that keep making money, while downstream application companies fail to earn adequate returns, this growth model will be difficult to sustain over the long term.
What will ultimately determine the future scale of the AI industry is whether OpenAI, Anthropic, and other AI application companies can turn computing power into real productivity and commercial revenue.
He believes that if businesses can adopt AI widely and earn a return on their investments, the semiconductor industry could grow from $1 trillion to $2 trillion or even $3 trillion.
The reasoning behind this is very important.
The long-term ceiling for demand for AI chips depends not only on how much tech giants are willing to invest, but also on how much economic value society as a whole can ultimately create with AI.
Fourth, Chinese open-source AI models may be changing the global AI pricing landscape.
Lip-Bu Tan specifically mentioned Chinese AI models such as DeepSeek, Kimi, and GLM.
He said these models are already good enough, and their Token costs—the cost of the units used to measure the text processed by a model—are low, enabling rapid adoption.
He also emphasized that he is a strong supporter of open-source technology.
There's a trend here worth paying attention to.
Businesses deploying AI in the future may not all need to buy the most expensive closed-source model services.
If open-source models can handle most commercial tasks at a lower cost, businesses may switch to more economical solutions.
This could have two different effects.
Lower-cost open-source models could put pricing pressure on providers of expensive, closed-source models.
But for the AI infrastructure industry as a whole, lower model usage costs could also encourage more businesses to deploy AI, boosting demand for inference computing.
In other words, models getting cheaper doesn't necessarily mean demand for chips will fall.
On the contrary, if AI usage grows fast enough, overall demand for computing power could continue to expand.
Fifth, quantum computing could become an important factor in the next phase.
Lip-Bu Tan also suggested that quantum computing could gradually enter mainstream applications over the next three to five years.
Although this timeline remains subject to significant technical and commercialization uncertainties, it reflects how the semiconductor industry is exploring new directions beyond traditional computing architectures.
Competition in future AI infrastructure may involve not only GPUs, ASICs (application-specific integrated circuits), and CPUs (central processing units), but also new computing technologies and software systems that improve hardware utilization.
From an investment perspective, I think the most important takeaway from Lip-Bu Tan's remarks this time is that they reveal the structural transformation underway in the AI industry.
Over the past few years, the market has focused mainly on who can produce the most chips.
Going forward, the market may focus increasingly on who can build infrastructure at the lowest cost of capital, and who can truly turn AI computing power into profits.
This means the investment thesis for AI is shifting from simply expanding computing capacity toward capital efficiency and commercial returns.
Upstream suppliers such as $NVDA, $AVGO, and $MU may continue to benefit from strong demand over the next two or three years.
But for the AI industry as a whole, the next round of valuation expansion may ultimately depend on whether revenue from applications can catch up with infrastructure investment.
The biggest risk to the AI revolution may not be a sudden disappearance of demand for computing power, but rather that demand remains strong while capital markets are no longer willing to provide funding without limit.
Do you think the biggest investment opportunities in the AI industry from 2027 to 2029 will still be concentrated in infrastructure companies like $NVDA and $AVGO, or will they gradually shift to AI application companies that can actually generate profits?
