AI Infrastructure Spending: From Cash Machine to Capital Guzzler
AI infrastructure spending is the large‑scale capital outlay by cloud and platform giants on chips, data centers, networking, and power systems required to train and run modern AI models, and it is now large enough to reshape cash flow patterns, shareholder payouts, and even the long‑term economics of the technology sector. Goldman Sachs estimates that Amazon, Alphabet, Microsoft, Meta and Oracle together will spend about USD 755 billion (approx. RM3.49 trillion) on capital expenditures in 2026, an 83% increase from 2025. That figure alone shows why AI is no longer a side project bolted onto the old Big Tech model. The cash that once funded reliable buybacks is now committed to GPUs, memory and concrete. Investors who treated these firms as growth stocks with automatic capital returns must now reassess what, exactly, they own.
The Quiet Collapse of the Big Tech Buyback Era
The surge in hyperscaler capex is forcing a visible reset in tech capital allocation. According to Goldman analysis cited by MarketWatch, hyperscaler buybacks fell by nearly two‑thirds in the first quarter, even as spending plans soared. Alphabet repurchased no stock in its latest quarter after buying back about USD 15.1 billion (approx. RM69.9 billion) in the same period a year earlier. Meta is pushing more cash into infrastructure, while Amazon has long stepped back from regular repurchases. At the same time, Amazon’s free cash flow for the 12 months through March 31, 2026 dropped to USD 1.2 billion (approx. RM5.6 billion) from USD 25.9 billion (approx. RM119.6 billion) a year earlier, as purchases of property and equipment rose by USD 59.3 billion (approx. RM274.0 billion). The old capital‑return engine is giving way to an AI‑first balance sheet.
Hyperscaler Capex Plans: AI Data Center Costs Take Priority
The new spending profile is specific: it is AI infrastructure, not generic growth capex. Amazon has laid out a roughly USD 200 billion (approx. RM924.0 billion) capital plan for 2026 tied to AI chips, data centers and robotics. Analysts guide Alphabet into the USD 175–185 billion (approx. RM808.0–RM854.0 billion) range, while Meta has lifted its 2026 capital expenditure outlook to USD 125–145 billion (approx. RM577.5–RM669.9 billion), citing higher infrastructure and memory costs. Microsoft’s spending sits above the USD 88.2 billion (approx. RM407.7 billion) it recorded in fiscal 2025. With Goldman now expecting total hyperscaler capex near USD 770 billion (approx. RM3.56 trillion) and roughly equal to cash flow from operations, debt markets are becoming part of the model. J.P. Morgan sees AI capital spending reaching USD 5.5 trillion (approx. RM25.41 trillion) by 2030, with USD 4.1 trillion (approx. RM18.94 trillion) financed through debt, turning AI into a long‑dated infrastructure bet.
Tech Sector Inflation: AI Data Centers Break the Deflation Habit
While capex reshapes balance sheets, AI data center costs are rewriting pricing trends across the tech sector. For more than two decades, software and electronic components mostly became cheaper year after year. That pattern is reversing. Recent data show that consumer prices for computer software and accessories rose 14.5% year‑on‑year in May, the largest annual increase since records began in 2000, while producer prices for electronic components jumped 27% over the same period. Memory stands out: DDR5 and DDR4 prices have climbed about 290% year‑on‑year, more than tripling in a single year. Analysts tie this surge to AI data center construction consuming vast amounts of chips and storage, tightening supply and causing tech sector inflation. Industry views suggest this pressure, combined with geopolitical risk, could extend through 2027, ending the long assumption that tech product costs only fall.
From Software Margins to Infrastructure Economics
The shift in AI infrastructure spending marks a structural change in how hyperscalers create value. For years, investors enjoyed software‑style margins paired with rising buybacks and, in some cases, dividends. Now these firms look more like infrastructure companies: heavy upfront AI data center costs, long payback periods, and tighter internal cash rationing. Free cash flow for the five largest hyperscalers is projected to drop 91% in 2026 to about USD 16 billion (approx. RM73.9 billion), even as net income is expected to rise 25% to USD 506 billion (approx. RM2.34 trillion). This accounting gap is the AI buildout in condensed form. Shareholders are exposed to a different risk: less technical support from buybacks and more dependence on future AI revenue to cover depreciation and debt service. The bet may still work, but the capital strategy and the investor contract have clearly changed.
