Key Takeaways
- • Pershing Square argues Wall Street is underestimating the long-term payoff from hyperscaler AI capex at Microsoft, Meta, Amazon and Alphabet, even though analyst models already reflect current spending levels.
- • Pershing Square estimates the four companies could spend a combined $740 billion on capital expenditures in 2026, up from roughly $140 billion in 2023, with consensus pointing to more than $1 trillion in 2027.
- • The firm's math assumes one gigawatt of AI data-center capacity costs $40 billion to $50 billion and generates $15 billion to $20 billion in annual revenue, implying a 15% to 25% pre-tax return on invested capital.
- • If those assumptions hold, Pershing Square says hyperscaler revenue growth could exceed consensus by 5 to 10 percentage points annually through 2029, leaving revenue 15% to 35% above current estimates after three years.
- • The return assumptions are unproven; low utilization or falling AI pricing could turn the same capex into value destruction, so investors should monitor utilization, cloud growth, incremental margins and free cash flow.
Summary
Pershing Square, the investment firm led by Bill Ackman, contends that Wall Street may be underestimating the long-term earnings potential of the large AI spending programs at Microsoft (MSFT), Meta (META), Amazon (AMZN) and Alphabet (GOOG). According to the firm, analyst models fully incorporate today's elevated capital expenditures but assign too little value to the future revenue and earnings those investments could generate.
The scale of hyperscaler spending has grown rapidly. Pershing Square estimates the four companies could spend a combined $740 billion on capital expenditures in 2026, compared with roughly $140 billion in 2023, and consensus estimates point to more than $1 trillion in 2027. The firm characterizes much of this spending as offensive rather than defensive, arguing that it is aimed at expanding cloud demand, improving advertising, accelerating software adoption and creating new AI products rather than merely protecting existing businesses.
Pershing Square's return framework rests on several assumptions. It estimates that building one gigawatt of AI data-center capacity with the latest-generation chips costs roughly $40 billion to $50 billion, and that this capacity could generate around $15 billion to $20 billion in annual revenue and $7.5 billion to $10 billion of incremental operating income, implying a 15% to 25% pre-tax return on invested capital. If these assumptions prove correct, the firm says annual hyperscaler revenue growth could exceed current consensus by five to 10 percentage points through 2029, leaving revenue 15% to 35% above today's estimates after three years.
The firm points to early signs of monetization, including stronger AWS growth at Amazon, AI-driven advertising improvements at Meta, and opportunities across Microsoft Azure, Copilot and GitHub. It also cites advantages held by the hyperscalers, including distribution, customer relationships, proprietary data, security and balance-sheet strength.
The article notes that these returns are assumptions rather than guarantees. If AI demand disappoints, utilization stays low or pricing falls faster than expected, the same capex programs could destroy value. Key indicators to watch include AI infrastructure utilization, cloud growth, incremental operating margins, free cash flow, and whether consensus revenue estimates for 2027 through 2029 begin moving higher.
The scale of hyperscaler spending has grown rapidly. Pershing Square estimates the four companies could spend a combined $740 billion on capital expenditures in 2026, compared with roughly $140 billion in 2023, and consensus estimates point to more than $1 trillion in 2027. The firm characterizes much of this spending as offensive rather than defensive, arguing that it is aimed at expanding cloud demand, improving advertising, accelerating software adoption and creating new AI products rather than merely protecting existing businesses.
Pershing Square's return framework rests on several assumptions. It estimates that building one gigawatt of AI data-center capacity with the latest-generation chips costs roughly $40 billion to $50 billion, and that this capacity could generate around $15 billion to $20 billion in annual revenue and $7.5 billion to $10 billion of incremental operating income, implying a 15% to 25% pre-tax return on invested capital. If these assumptions prove correct, the firm says annual hyperscaler revenue growth could exceed current consensus by five to 10 percentage points through 2029, leaving revenue 15% to 35% above today's estimates after three years.
The firm points to early signs of monetization, including stronger AWS growth at Amazon, AI-driven advertising improvements at Meta, and opportunities across Microsoft Azure, Copilot and GitHub. It also cites advantages held by the hyperscalers, including distribution, customer relationships, proprietary data, security and balance-sheet strength.
The article notes that these returns are assumptions rather than guarantees. If AI demand disappoints, utilization stays low or pricing falls faster than expected, the same capex programs could destroy value. Key indicators to watch include AI infrastructure utilization, cloud growth, incremental operating margins, free cash flow, and whether consensus revenue estimates for 2027 through 2029 begin moving higher.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.