For more than a decade after the Global Financial Crisis, hyperscalers – such as Alphabet Inc. (NASDAQ:GOOGL), Meta Platforms Inc. (NASDAQ:META), Microsoft Corp. (NYSE:MSFT) and Amazon.com Inc. (NASDAQ:AMZN) – perfected the asset-light model.
Cash piled up. Margins expanded. And excess capital flowed back to shareholders through aggressive buybacks and steadily rising dividends.
U.S. hyperscalers became machines of financial efficiency — generating enormous free cash flow while keeping capital intensity relatively contained.
That era is ending.
According to Goldman Sachs analysts Ben Snider and Ryan Hammond, Wall Street consensus estimates now project hyperscaler capital expenditures will reach $667 billion in 2026 — up $127 billion since the start of the fourth-quarter earnings season and implying 62% year-over-year growth.
More striking than the headline number is what it represents.
Capex is now on pace to consume roughly 92% of hyperscaler cash flows from operations, a higher share than during the Dot Com Boom.
"This dynamic has dramatically reduced hyperscaler free cash flows," Goldman Sachs said.
In practical terms, nearly every dollar of internally generated cash is being reinvested back into infrastructure — primarily AI-related compute, data centers, networking and power capacity.
The Great Capital Allocation Shift
This marks a dramatic pivot in business models.
At the start of 2023, hyperscalers were allocating roughly 43% of their cash flows to buybacks. Today, that share has fallen to just 16%. Collective gross buybacks declined 15% year-over-year in 2025.
At the same time, some firms — including Oracle Corp. (NYSE:ORCL) and Alphabet — have tapped debt markets to help fund their expansion.
Instead of optimizing for shareholder distributions, management teams are optimizing for scale in AI.
This is not incremental spending. It is a full-scale reinvestment cycle.
Goldman expects modest further upward revisions to 2026 capex estimates, potentially pushing spending toward $700–$725 billion under upside scenarios.
Capex Growth Is Peaking — But Not Yet Slowing
The growth rate of capex is expected to begin decelerating in the second half of 2026.
This suggests hyperscalers will continue fortifying their AI infrastructure — but at a slower pace than the breakneck expansion seen so far.
Goldman argues that a deceleration in capex growth could actually serve as a catalyst.
Slower spending would provide visibility into a trough in free cash flow, potentially allowing investors to value these companies on earnings again rather than on reinvestment narratives.
"Decelerating capex growth should also signal a deceleration in the revenue and earnings growth of the infrastructure complex that has benefited from that investment spending," analysts added.
What This Means For Investors
For investors and shareholders, the trade-off is becoming clearer.
Every dollar funneled by big tech into AI data centers, chips and power capacity is a dollar not returned through buybacks or dividends in the near term.
The same companies that once defined capital-light efficiency are now redirecting nearly all of their free cash flow toward defending and expanding their AI dominance.
The implicit bet is strategic: spend aggressively today to protect monopoly-like positioning tomorrow.
But that raises the central question — will future AI-driven revenue and margins be large enough to justify the extraordinary scale of current investment?
In that sense, the AI boom is not just a growth story. It is a capital allocation story — and one that forces investors to decide whether hyperscalers are temporarily sacrificing returns or permanently rewriting the rules of Big Tech finance.
Image: Shutterstock
