- The AI infrastructure boom has a new problem, and it’s not chips or electricity—it’s money. Broadcom CEO Hock Tan said at the DBS Asia Future Forward Conference in Singapore that capital access could become a bigger constraint than semiconductor shortages, memory limitations, or power availability. Intel CEO Lip-Bu Tan agreed, pointing to rising borrowing costs and growing questions about whether the hundreds of billions being poured into data centers will actually generate returns. For two years, the industry cycled through supply bottlenecks: chip shortages, memory crunches, power constraints. Each time, the bottleneck eased as production ramped. But capital is different. If borrowing costs stay high, the economics of AI infrastructure projects become difficult to justify, and deals simply don’t get done.
- Rising bond yields are the culprit. With 10-year Treasurys above 5%, borrowing to fund multi-year data center buildouts is expensive. A company that planned a $10 billion data center project with 4% financing now faces 6-7% costs. That extra 2-3 percentage points, compounded over years, meaningfully changes project returns. If a data center takes three years to build and five more years to generate positive cash flow, the weighted cost of capital matters enormously. For AWS, Google, Microsoft, and other infrastructure buyers, the calculus gets harder when rates are high. They may still proceed (because demand is real), but they’ll be more selective about which projects get funded.
- The near-term outlook is still strong. Both Hock Tan and Lip-Bu Tan expect demand for compute to exceed supply through 2029 or 2030. The AI adoption wave is real, and companies need infrastructure. But that demand is dependent on continued strong returns from AI investments. If the next generation of AI models disappoints—if they don’t deliver the productivity gains companies expected—then demand drops and a bunch of half-built data centers suddenly become white elephants. That’s the risk nobody wants to talk about yet.
- Supply constraints are actually easing. Hock Tan expects memory shortages to ease significantly over the next year, wafer supplies to improve by 2028-2029, and power infrastructure to come online as needed. The industry solved each bottleneck by throwing money and innovation at it. But capital constraints can’t be solved that way. If capital becomes scarce, projects get delayed or cancelled, and the entire timeline slips. For infrastructure investors expecting to deploy capital into AI data centers at high returns, that timing risk is now the real constraint.
What Happened?
Broadcom CEO Hock Tan warned at the DBS Asia Future Forward Conference in Singapore on Monday that access to capital could become the biggest constraint on AI infrastructure growth, potentially surpassing historical bottlenecks like semiconductor and memory shortages or electricity availability. Intel CEO Lip-Bu Tan echoed the concern, citing rising costs of capital and investor questions about whether companies making enormous AI infrastructure investments will generate sufficient returns. Both executives noted that the industry has repeatedly overcome supply-side bottlenecks over the past two years through expanded production capacity. Hock Tan anticipated that memory shortages could ease within a year, wafer constraints could improve by 2028-2029, and power infrastructure could come online as needed. Despite capital concerns, both remained bullish on near-term demand, with Hock Tan stating that compute demand would likely exceed supply through 2029 or 2030. The warnings reflect growing pressure on project financing from higher bond yields, which increase the cost of funding multi-year data center construction and chip purchases.
Why It Matters?
Capital constraints represent a different kind of bottleneck than previous supply-side shortages. Chips can be produced faster with more fabs and investment. Power can be added by building infrastructure. But if the cost of capital rises, projects simply don’t get funded, regardless of technical feasibility. With 10-year Treasury yields above 5% and rising, the cost of borrowing for long-term infrastructure projects has increased significantly. A company financing a data center project now pays 200-300 basis points more than it would have two years ago. Over a five to ten-year project timeline, that compounds into a meaningful reduction in expected returns. For mega-cap tech companies (Amazon, Google, Microsoft), the higher rates are a nuisance but manageable. For smaller infrastructure investors and emerging markets companies, the capital crunch could be severe. The CEOs’ comments also highlight an emerging credibility crisis in AI ROI. If companies have been overestimating the returns from AI infrastructure investments, or if the productivity gains materialize slower than expected, funding dries up fast. The fact that two major semiconductor executives are publicly warning about capital now suggests that behind closed doors, infrastructure investors are getting nervous about deal economics.
What’s Next?
Monitor capital raising announcements from infrastructure investors and tech giants over the next quarter. If Amazon, Google, Microsoft, and OpenAI continue announcing $20-30B infrastructure commitments despite higher rates, that signals confidence in ROI. If the pace of announcements slows, it validates the capital crunch concern. Watch for infrastructure projects getting delayed or scaled back. If construction timelines for data centers slip, or if planned capacity additions get reduced, that confirms capital constraints are biting. Track bond yields and credit spreads. If yields fall back below 5% or corporate spreads tighten, that relieves financing pressure and accelerates infrastructure deployment. If yields spike above 5.5%, expect more caution. Monitor Fed communication. If the Fed signals rate cuts ahead, that improves infrastructure economics immediately. If the Fed hints at further hikes, expect capital to dry up. Watch infrastructure deal flow. Private equity firms have been exploring AI data center investments. If that deal activity slows or IPO plans get postponed (like Firmus), that validates the financing concerns. Finally, pay attention to whether AI productivity gains materialize as expected. If companies start disclosing disappointing ROI on their AI investments in Q3 and Q4 earnings, that will make capital even scarcer because the fundamental case becomes questionable.
Affected Tickers and Coins: AVGO | INTC | AMZN | GOOGL | MSFT
Source: Market reporting
