The billion dollar bill for artificial intelligence: how far will hyperscalers go with data centers

Combined capital expenditure from the world's fourteen largest data center operators is set to reach close to $750 billion in 2026, nearly double what was invested two years ago. Meta, Microsoft, Amazon and Alphabet lead this race, and the most recent example shows the scale of the bet: Meta's Hyperion project in Louisiana had its budget revised from $27 billion to more than $50 billion within months.

The volume of capital is not the only relevant figure. Data center capacity under construction worldwide already exceeds 23 gigawatts, and the US Department of Energy projects these facilities could account for up to 12% of US electricity demand by 2028. Power infrastructure, not just chips, has become the bottleneck of the AI race, which explains why turbine and electrical equipment makers such as Eaton and Caterpillar are also having a record year for orders.

The point worth watching for investors is the gap between corporate messaging and market reaction. Three of the four largest hyperscalers lost market value after their most recent earnings releases, even as they raised their investment forecasts. The question is not whether long term demand for artificial intelligence is real, but whether these outlays will pay off in the short and medium term. Meta itself signaled the most obvious answer to that discomfort: reselling idle computing capacity to other companies, including direct competitors in model development.

That capacity resale trend is worth watching closely. When a company spending $145 billion a year in capex starts building a cloud business to monetize excess capacity, it signals both confidence in future demand and real pressure for near term results. For capital allocators, the narrative of essential, recession resistant infrastructure coexists, in practice, with high valuations and risk concentrated among a small number of companies.

The lesson for Brazilian investors exposed to this theme, whether through US equities, multi-strategy funds or venture capital vehicles built around artificial intelligence, is to separate two questions that tend to get treated as one. That the technology will reshape the economy is no longer in doubt. But not every company building a data center will capture that value in equal measure, and the market has started pricing that distinction more strictly.