The bill is coming due. Goldman Sachs calls it an intact trade. I call it a forced march. On May 2, Anshul Sehgal, the global co-head of Fixed Income, Currency and Commodities at Goldman Sachs, signaled that the megacap tech earnings cycle confirms the AI trade is far from over. This narrative is comforting for the institutional desk. It is less so for the analyst looking at the widening gap between capital expenditure and realized net income. The market is addicted. It needs the hit of the latest generation of silicon to keep the hallucination alive.
The FICC Perspective on Silicon
Why is a fixed income and commodities head commenting on tech earnings? The answer lies in the plumbing. The AI trade has migrated from a speculative equity play to a massive infrastructure and debt story. We are no longer just talking about software margins. We are talking about the massive issuance of corporate debt to fund data center clusters that consume more power than mid-sized European nations. Sehgal’s commentary suggests that the liquidity flows in the bond markets are still supporting this expansion. Per recent Reuters reporting on global finance trends, the appetite for high-grade corporate debt remains robust, specifically for firms that can prove they are building ‘productive’ AI assets.
But productivity is a lagging indicator. The current reality is a frantic arms race. Microsoft, Alphabet, and Meta are locked in a prisoner’s dilemma. If one slows down their GPU procurement, they risk losing the foundation of the next decade’s computing. This is not a choice. It is a survival mechanism. The fixed income desk sees this as a steady demand for capital. The investigative eye sees it as a potential debt trap if the revenue growth does not accelerate to match the depreciation of these rapidly aging chips.
The Capex Trap in Numbers
The numbers from the first quarter of this year are staggering. We are seeing capital expenditure growth rates that double or triple the rate of revenue growth. This is the definition of a low-return environment in the short term. The hope is that the ‘inference’ phase of AI will eventually lower costs and drive consumer adoption. However, as of early May, the enterprise spend on AI tools is still largely in the pilot phase. The infrastructure is being built for a city that has yet to be populated.
Q1 2026 Megacap Tech AI Capital Expenditure vs Revenue Growth Percentage
The Technical Bottleneck of Inference
Training a model is expensive. Running it is where the real cost lies. The industry is currently shifting from the training phase to the inference phase. This requires a different kind of architecture. The H100s and B200s that dominated the 2024 and 2025 landscape are being joined by custom ASICs designed for efficiency. Yet, the energy grid is not keeping up. In Northern Virginia and the Dublin corridor, data center permits are being delayed by years. This physical constraint is the one thing Goldman’s FICC desk cannot fix with a swap or a hedge.
The cost of electricity has become a primary commodity concern for tech giants. This is why we see Amazon and Microsoft investing directly in nuclear energy projects. They are no longer just software companies. They are energy utilities. According to Bloomberg’s latest market analysis, the correlation between tech stock performance and the price of copper and electricity futures has reached an all-time high this month. The ‘AI trade’ is now a physical infrastructure trade.
Comparative Financial Metrics Q1 2026
| Company | AI Capex (Billions USD) | Year-over-Year Growth | Free Cash Flow Margin |
|---|---|---|---|
| Microsoft | 16.2 | 35% | 28% |
| Alphabet | 14.5 | 42% | 24% |
| Meta | 11.2 | 38% | 21% |
| Amazon | 18.5 | 30% | 12% |
The Liquidity Threshold
Goldman Sachs is right about one thing. The trade is intact because the money is still there. The megacaps are sitting on a mountain of cash that allows them to ignore the immediate ROIC (Return on Invested Capital) requirements that would sink a smaller firm. But this is a game of attrition. The ‘intact’ trade relies on the assumption that a ‘killer app’ will emerge to justify the trillion dollar spend. If that app remains a more efficient chatbot or a better ad-targeting algorithm, the valuation multiples will eventually contract.
We are watching the credit spreads on tech-issued bonds. If the market begins to price in a longer wait for AI profitability, those spreads will widen. For now, the narrative of ‘limitless potential’ holds the line. But narratives are fragile. They shatter when the interest payments on the infrastructure exceed the growth of the top line. The market is currently betting that this point is years away. The data suggests the window is much tighter.
The June Milestone
Attention now shifts to the June 15 Federal Reserve meeting. The cost of capital remains the ultimate arbiter of the AI trade. If the Fed maintains its current stance, the pressure on the megacaps to show real, non-speculative revenue from AI will intensify. Watch the 10-year Treasury yield. If it crosses the 4.8% threshold, the ‘intact’ trade will face its first true stress test of the year. The infrastructure is built. The chips are spinning. Now, the world has to actually use it.
