The consensus reads the AI boom as a demand story with a financing footnote. AI infrastructure ROI uncertainty, however, was the dominant theme when Broadcom CEO Hock Tan and Intel CEO Lip-Bu Tan took to the stage at the DBS Asia Future Forward Conference, where both men said plainly what the financial hype cycle has been reluctant to accommodate: the business case for AI remains unproven, the cost of capital is rising, and nobody, including the people building the infrastructure, is entirely sure what they are doing.
The panel was moderated by DBS CEO Tan Su Shan, according to The Business Times. The audience was not a casual one: some 700 senior business leaders, entrepreneurs and investors were expected at the inaugural conference, according to PR Newswire. That is the room in which Hock Tan chose to compare a large class of AI investors to casino gamblers.
The Casino Metaphor Was Not Throwaway
Hock Tan’s precise formulation, as reported by Bloomberg on the panel discussion, was that ‘a lot of investors have no clue what they are investing in’ and that they ‘set very extreme expectations only to find out that it’s almost like they are in a casino.’ He applied the same scepticism to OpenAI directly. Its annualised $50 billion in revenues at the end of September, he argued, does not demonstrate a sustainable, profitable business model, given that the costs of AI and the capital expenditures of the buildout are generating large losses. ‘So frankly, it doesn’t tell anything at this point except that there is a lot of potential revenue growth out of providing frontier generative AI models,’ he said.
The more structural problem, in Hock Tan’s framing, is that AI model developers need customers who generate genuine returns from deploying AI, rather than users who are, in his word, ‘wasting money’ on tokens as they experiment. His analogy, as reported by Bloomberg: ‘When you are a kid you go to a video arcade, you play. Your parents give you $10 of tokens, you finish it in 10 minutes and you haven’t learned anything. That’s what’s happening in enterprises today.’
AI Infrastructure ROI Uncertainty Runs Into Rising Capital Costs
On the infrastructure side, Hock Tan said his presumption is that the buildout boom ‘might continue’ through 2029 and ‘probably into 2030,’ though he qualified that heavily: ‘that depends to a large extent on things we are learning as we go along in this process.’ The industry has navigated ‘a series of bottlenecks’ over the past two years, he said, from memory shortages to wafer constraints (which could ease by 2028 or 2029) to power availability, per CNBC’s reporting on the panel. Demand for compute infrastructure currently exceeds supply, and that imbalance could persist through 2029 and potentially into 2030, he added.
But the harder constraint may be financial rather than physical. Financing the AI infrastructure needed to power these models ‘might be the biggest constraint,’ Hock Tan said. Intel’s Lip-Bu Tan agreed without equivocation: ‘If there’s any challenge in the front, it’s capital and the cost of capital,’ he said, according to CNBC. ‘And then the other part is in the longer run, people want to see the ROI.’
The bond market is already pricing in that anxiety. SpaceX’s 10-year senior unsecured investment-grade notes, sold in June with a coupon of 5.875%, were trading at a yield of 7.20% by the end of last week, according to Wolf Street. That spread compression (running in the wrong direction) is being described as a template for future issuance of similar AI-adjacent bonds. The cost of funding the buildout has risen materially in a short window, which makes the still-undemonstrated ROI considerably more precarious to achieve.
Whether demand holds up depends on business adoption rates and the returns companies actually realise from AI investments, both CEOs acknowledged, according to The Straits Times’ reporting on the panel. If advanced AI model performance disappoints, or capability improvements begin to plateau, demand for computing infrastructure could slow, Hock Tan added, per CNBC. Companies will need to demonstrate meaningful ROI through improved efficiency and new business models, Lip-Bu Tan said.
The Host’s Own Numbers Complicate the Picture
There is an irony in the conference setting that deserves a moment. DBS, whose CEO was moderating the panel on AI ROI uncertainty, reported internally that it deployed more than 2,000 AI models across over 430 use cases in 2025 and generated approximately SGD 1 billion in economic value from data analytics and AI/ML, according to PR Newswire. That figure is a rare, concrete, institution-specific claim of AI returns, precisely the kind of evidence that Hock Tan argued the broader market still lacks.
The contrast is useful rather than contradictory. One large, operationally sophisticated institution has quantified returns from AI deployment. The question the two CEOs were really raising is whether that example scales, and whether the infrastructure spending being laid down in anticipation of scale is being funded at a cost that the eventual returns can justify. Neither man was willing to guarantee it. Hock Tan’s closing note on what he might have got wrong when looking back: ‘I think we’ve probably got everything wrong. It’s a good way, humble way to realize how little we really know about the technology we are creating.’ Lip-Bu Tan set his own bar somewhat higher, saying he would consider himself a ‘damn good CEO’ if he got 70% right. The AI infrastructure ROI uncertainty embedded in both answers is not a talking point. It is the stated position of two of the industry’s central builders.
