Intel CEO Lip-Bu Tan said at a DBS-hosted forum in Singapore on October 12, 2026, that AI demand is unlikely to slow significantly over the next two to three years. According to CNBC, he also raised concerns about the cost of capital and returns on investment. Even if demand for AI computing capacity is strong, there is no guarantee that companies can raise the funds to expand it, or that the conditions will be in place for usage to generate revenue. At a time when hardware supply remains tight, the question the semiconductor chief raised was what will sustain investment beyond this point.
What Does "Two to Three Years" Actually Cover?
CNBC's report on the forum describes Tan's outlook as follows: driven by the spread of AI applications and demand for computing, a major slowdown is unlikely over the next two to three years. This is not a promise about when shortages will ease for every type of semiconductor. The category is too broad to use as a forecast for component lead times.
Broadcom CEO Hock Tan, who appeared alongside him, said demand for computing infrastructure could outstrip supply until 2029, and possibly 2030. At the same time, he believes constraints on individual components have room to improve. Even taken together, the two executives' views do not lead to the conclusion that every component will be equally scarce for the same period.
The official DBS program listed the question for the two CEOs' session as where AI will create the next trillion dollars in value. How will companies that increase supply, and those that use the resulting equipment, generate value? That is why demand strength and profitability were discussed at the same event.
Even With Enough Equipment, Can Investments Be Recouped?
Intel's Tan named capital and the cost of capital as the immediate challenge, and said that over the long term "people want to see return on investment." This is a Japanese-language rendering of remarks reported by CNBC. Broadcom's Tan also pointed to fundraising as a potentially major constraint, treating strong demand for computing equipment and the difficulty of sustaining investment as two sides of the same picture.
The cost of capital has to be considered separately from the purchase price of equipment. Borrowing carries interest, and taking on equity means investors expect a commensurate return. Even if component supply grows, the burden of financing does not necessarily lighten in the same way.
Moreover, once the equipment is in hand, the task remains of recovering the cost through its use. Being able to run large volumes of AI processing is one thing; having that processing translate into customer payments or operational efficiency is a separate stage. Supply shortages can delay deployment, and the cost of capital raises the expenses a business must bear. Investment returns then test how useful the deployed AI actually was.
Seen this way, a bullish view on demand and caution about profitability can coexist. Orders for equipment do not guarantee profits for the buyers. Even if AI capabilities improve, the time it takes to embed them in operations and the running costs can lead to differences in investment outcomes.
How DBS Reads "Economic Value"
DBS, which hosted the forum, has also published concrete figures as a user of AI. According to its official announcement on September 30, in 2025 it deployed more than 2,000 AI models across more than 430 use cases, generating roughly S$1 billion in economic value from data analytics and AI/machine learning. This is the company's own tally, and it indicates the scale at which AI use has expanded from experimentation to ongoing operational use.
However, the S$1 billion cannot be read as AI revenue or net profit. It covers not only generative AI but also data analytics and AI/machine learning. Because the announcement does not break down the investment amount or the calculation method, converting it into a return on investment is also difficult.
Between "value was created" and "the money invested was recovered" lies information that needs to be checked. Evaluations change depending on how far costs were included, over what period results were measured, and what was compared against the scenario without AI. DBS's figure is useful for gauging the scale of use, but it is not a metric that can be applied directly to the profitability of other companies or data center investments.
To judge investment recovery, one needs to match the results obtained in operations against the costs required, not just the number of models deployed. Gains on the supply side, such as increased semiconductor shipments, do not reveal whether users can keep investing.
When Cheaper AI Changes Demand and Profitability
In a Reuters article distributed by AsiaOne, Intel's Tan also touched on the low token costs and rapid adoption of Chinese AI models. He called the performance of DeepSeek, Kimi and GLM "good enough" and said the market will choose the model suited to each use. This is his own assessment, not the result of independent testing that settles how the models rank.
A token is the unit a model uses when processing text. If its cost falls, it becomes easier to hold down spending for the same processing volume. But whether quality can be maintained when switching to a cheaper model, and how much total operating cost can be cut, must be verified separately. Lower unit prices cannot simply be linked to reduced semiconductor demand or the end of supply shortages.
When tracking AI investment, it is worth looking at how user companies measure and report their results alongside equipment expansion plans. If they can show contributions to operational improvement or revenue with costs and time periods aligned, companies will be able to base their next capital spending not only on expectations of demand but on actual results from use.