AI is becoming a financial engineering business

AI capex is shifting advantage from frontier models to balance sheets: hyperscalers have already invested $1.1T since 2023 and plan another $745B in 2025. Despite ongoing losses at OpenAI/Anthropic, cloud returns are emerging—Microsoft cloud rose 32% to $39.3B, AWS grew 37% to $42.2B, and Google cloud surged 82% to $24.8B—driving an estimated +$450B market-value jump for Microsoft in a day. The cycle remains risky as IBM shares fell ~25% after customers postponed software purchases to prioritize AI infrastructure, and Alphabet/Meta have seen free cash flow pressure from capex.

Analysis

The near-term winner set is the hyperscalers with the cheapest funding and the most monetizable installed base: MSFT, AMZN, and GOOGL. In this phase, capex is functioning less like a drag and more like a barrier-to-entry tax on everyone else; that should keep share gains concentrated in the firms that can finance utilization before pricing power is proven. The market is still underappreciating that the AI buildout is a balance-sheet contest, not just a model contest.

The clearest loser is legacy enterprise software where AI spend substitutes for renewal budgets before it creates new demand. IBM is the canary, but the second-order risk extends to slower-growth software names with weak platform lock-in and to any vendor relying on CIO discretionary spend; that budget is being pulled into data centers, power, and chips first. Over 1-3 months, watch for further revenue deferrals and weaker bookings rather than outright demand collapse.

Second-order beneficiaries sit in the financing stack: BX, GS, and MS can monetize data-center, structured credit, and private financing flow even if the operating economics of AI remain uncertain. NVDA and AVGO still win on volume, but if customers increasingly fund capex through long-dated structures, chip demand becomes more durable but also more financialized, which can compress upside if the market starts pricing in leverage and future overcapacity. The contrarian miss is that the current capex wave may be more bullish for cloud revenues than bearish for FCF in the next two quarters; overcapacity is a 6-18 month risk, not a near-term one.

Key Decisions for Investors

• Long MSFT / short IBM for 1-3 months: express the budget-substitution theme with asymmetric downside in IBM if software deferments continue; stop if IBM bookings stabilize or MSFT cloud growth decelerates materially.
• Long GOOGL / short META into the next earnings season: GOOGL has more direct cloud monetization and financing optionality, while META remains more exposed to free-cash-flow compression if capex keeps outrunning revenue.
• Tactical long NVDA on pullbacks, but size modestly: demand should remain supported by the financing buildout; risk/reward is weaker than in MSFT/GOOGL because the stock already discounts structural scarcity. Falsifier: any sequential slowdown in hyperscaler capex guidance.
• Watch-list long BX and GS on weakness if private-credit and structured-finance activity around data centers continues: this is a second-order fee pool, but only worth adding if deal volume and spreads confirm the trend.