AI & ComputeLOT 261120 · 4 min read

AI Build-Out Hits $800 Billion a Year as Debt Fear Grows

Hyperscalers will spend $800 billion on AI infrastructure in 2026, up 94 percent, while borrowing hits $214 billion and investors question when revenue arrives.

By Tom Whitfield799 words

Key takeaways

  1. Alphabet, Amazon, Meta, Microsoft, and Oracle plan $800+ billion in 2026 AI capex, up 94 percent from 2025.
  2. Brookings projects over $10 trillion in U.S. AI investment from 2025 to 2032, averaging 3.6 percent of GDP per year.
  3. The big four hyperscalers borrowed a combined $214 billion in 2026, up from $83 billion in all of 2025.
  4. Sona identified 176 circular AI financing deals worth $3.6 trillion in the past three and a half years.
  5. Nvidia lost roughly $1.7 trillion in market value after an alleged Chinese espionage operation forced out CEO Jensen Huang.
Boom or Bubble? Understanding AI and the U.S. Economy
Boom or Bubble? Understanding AI and the U.S. Economy — FotoRieth / pixabay

The five largest U.S. hyperscalers plan to spend $800 billion or more on data centers, chips, and AI infrastructure this year, a 94 percent jump over 2025, and investors are starting to ask who pays for it. Alphabet, Amazon, Meta, Microsoft, and Oracle have yet to see meaningful returns on those sums, and the debt financing the build-out is getting more expensive.

The scale of the spending has no modern precedent. A September 2026 Brookings Institution report projects U.S. investment in AI and related equipment will exceed $10 trillion between 2025 and 2032, an average of 3.6 percent of GDP per year. Roger W. Ferguson Jr., a distinguished fellow for international economics at the Council on Foreign Relations, said the AI boom "may be one-and-a-half to three times bigger, as a percentage of GDP, than was true in other technology booms." The last time a new industry took such a large share of the economy was the railroad boom of the late nineteenth century.

How much is AI actually adding to growth?

A lot, by most measures, though less than the headline numbers suggest. U.S. stock portfolios have roughly doubled in value since 2022, and household wealth gains lifted consumer spending. In 2026, the ten largest AI stocks in the S&P 500 accounted for about 50 percent of the index's gains, according to Goldman Sachs.

Official GDP figures understate the effect, because most AI infrastructure relies on imported equipment that gets subtracted from the calculation. Excluding that discount, Bloomberg estimates U.S. growth in 2025 would have been 3.1 percent, against the 2.3 percent officially reported. Bloomberg's rival Economist, for its part, estimates the build-out has created around one million U.S. jobs and propped up data center construction in an otherwise weak construction sector.

The gains have not spread evenly. Lower-income households missed the stock rally and continue to face high inflation, energy prices, and slow wage growth. Productivity gains remain patchy: some studies find growth in software development and finance, others find none at all. A September 2026 McKinsey study estimates AI could displace between six million and sixteen million American workers over the next decade.

Is the industry a bubble?

Some features look like one. In third-quarter earnings, Alphabet, Amazon, and Meta all posted negative free cash flow. The S&P 500 dropped 4.4 percent after the U.S. midterms, as investors worried about regulation under a Democratic Congress. Nvidia lost roughly $1.7 trillion in market value after an alleged Chinese espionage operation forced out CEO Jensen Huang in early November after thirty-three years at the helm.

"The leading frontier AI labs have a 'whatever it takes' approach to spending to win the AI race," Rebecca Patterson, a CFR senior fellow, said. "The growing use of debt to finance the AI build-out, alongside questions about the timing and degree of revenue, is pushing up the cost of debt for many AI-related firms." Alphabet, Amazon, Meta, and Microsoft have borrowed a combined $214 billion so far in 2026, up from $83 billion in all of 2025.

Circular financing adds risk. Sona, a London-based asset manager, counted 176 AI-related circular deals worth $3.6 trillion in the past three and a half years — companies buying and selling infrastructure among themselves and lending against future purchase commitments, a pattern some compare to pre-2008 housing finance. Yet Anthropic still raised a record $81.3 billion in its mid-November IPO, valuing the Claude developer above $2 trillion. Ferguson cautioned that some companies "will probably go under," but noted AI inputs are multipurpose and adoption has outpaced the PC and the internet.

What could slow the build-out?

Power, land, water, and labor. Backlogs on imports of memory chips and grid equipment run two to ten years. "The challenge is training the people," CFR fellow Vinh Nguyen said. "We have a limited number of electricians and skilled talent in building data centers and manufacturing chips." In 2025, roughly 90 percent of key inputs — chips, servers, networking gear — came from East Asia, and Trump-era tariffs on semiconductors, aluminum, copper, and steel raise the cost of importing what U.S. manufacturing cannot yet supply.

Where does China fit?

Export controls on chip sales to China remain hard to enforce, and Chinese models have closed the performance gap despite far smaller investment. "The Chinese will beat us at pricing and will have comparable capabilities eventually," Nguyen said, while placing the U.S. edge in AI safety and trust. Patterson put the bottom line plainly: if Beijing wants to slow Washington down, it only has to pause critical mineral exports, and building supply-chain resiliency "will take years, not months."

cfr.org (Source)

Filed under

Share this article:

More from Tom Whitfield

Tom Whitfield

Show full bio

Staff writer covering companies, deals and strategy in Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.

1,131 articles

Read also

« Previous article