Goldman Sachs estimates capex from Meta, Microsoft, Amazon and Alphabet will reach $5.3 trillion by 2030, more than Japan’s GDP, with total AI infrastructure spend hitting $7.6 trillion
Summary:
- Goldman Sachs estimates combined capex from Meta, Microsoft, Amazon and Alphabet will reach $5.3 trillion by end-2030, surpassing the GDPs of Japan, the UK, India and France
- Total industry spending on data centres, power and computing could reach $7.6 trillion over the next five years
- The four hyperscalers are on track to spend up to $725 billion on capex in 2025 alone, more than double the $360 billion spent in 2024
- Goldman expects private markets to play a growing role in funding the AI buildout, with private data centre construction having accelerated materially in recent years
- Goldman describes the data centre buildout as a multi-year investment cycle; investors have raised concerns about long-term returns on the scale of spending involved
Goldman Sachs has sharply revised up its estimates for artificial intelligence infrastructure spending, projecting that just four technology companies will collectively deploy more than $5 trillion in capital through the end of the decade, a sum that would outrank the entire economies of Japan, the United Kingdom, India and France.
The bank’s updated hyperscaler capex forecasts, published Tuesday, put combined spending from Meta, Microsoft, Amazon and Alphabet at $5.3 trillion by 2030. Across the broader industry, total investment in data centres, power infrastructure and computing capacity could reach $7.6 trillion over the same five-year window.
The scale of annual spending is already accelerating faster than most projections anticipated. The four companies Goldman tracked are expected to commit up to $725 billion in capital expenditure this year, more than double the $360 billion they spent in 2025. The pace of increase suggests the investment cycle is still in its expansion phase rather than approaching any natural plateau.
Goldman described the data centre buildout as a multi-year investment cycle and noted that private construction activity in the sector has accelerated materially over the past few years. The bank said it expects private markets to take on a larger share of the funding burden going forward, a structural shift that would redirect a meaningful portion of AI infrastructure returns away from public equity markets and toward private capital vehicles.
The return on investment question has not disappeared. Investors have repeatedly flagged concern about the magnitude of spending relative to the commercial revenues AI products have so far generated, and the gap between capital deployed and monetisation achieved remains a live debate. Goldman’s framing implicitly addresses that concern by treating the current period as the early phase of a durable cycle rather than a speculative surge, though the bank stops short of quantifying when or how the returns will materialise.
What the numbers do establish clearly is that the four companies involved have made a collective decision to spend at a scale that has no precedent in corporate history.
—
The $5.3 trillion figure resets the scale of the AI infrastructure trade in ways that matter for capital markets beyond the tech sector. Power, construction, and industrial supply chains are the direct beneficiaries, and Goldman’s broader $7.6 trillion total data centre estimate confirms that the investment cycle has years of runway. The doubling of annual capex from $360 billion in 2025 to up to $725 billion this year is the more immediately actionable number, as it speaks to current earnings cycles for suppliers and utilities rather than a 2030 endpoint. The shift toward private markets as a funding mechanism is a structural signal worth watching: it suggests the hyperscalers expect to move faster than public capital markets can efficiently accommodate, and it will direct a growing share of AI infrastructure returns away from listed equities
