Bridgewater Warns Big Tech's $650 Billion AI Infrastructure Spending in 2026 Enters "More Dangerous Phase" as Compute Demand Outpaces Supply

US technology giants Alphabet, Amazon, Meta, and Microsoft are expected to collectively invest approximately $650 billion to scale up AI-related infrastructure in 2026, according to analysis by Bridgewater Associates, marking a sharp jump from $410 billion in 2025 that the world's largest hedge fund warns carries significant downside risks.

In a letter to clients, Bridgewater co-chief investment officer Greg Jensen said the artificial intelligence boom has entered a "more dangerous phase," marked by exponentially rising investments in physical infrastructure and growing reliance on outside capital to fund the unprecedented buildout.

"Compute demand continues to significantly outpace supply, driving hyperscalers to invest even more rapidly to try to someday get ahead of the demand," Jensen wrote. The four companies have already curbed share buybacks more aggressively to help fund the surge in capital expenditure, preserving cash for infrastructure investments rather than returning it to shareholders.

The scale of spending, Jensen emphasized, is creating significant downside risks if anything went wrong. Without continued access to capital markets and sustained investor confidence in AI returns, the massive infrastructure commitments could become financial liabilities rather than competitive advantages.

Beyond stock markets, Jensen noted that tech investment spending remains a significant "upward pressure for US growth." Bridgewater estimates tech investment added about 50 basis points to US GDP growth in 2025 and could provide around 100 basis points of support in 2026 as construction, chip production, and power infrastructure expand to meet AI demands.

However, the spending boom may also lift inflation in technology and communications equipment and push up electricity prices in some regions. Data centers require enormous power capacity—a single AI facility can consume as much electricity as a small city—creating localized strain on electrical grids and potentially driving up costs for other users.

Jensen cautioned that a severe stock market correction could undermine growth and limit companies' ability to raise capital, drawing parallels to the Dot-com bubble collapse in 2000 when inflated valuations and excessive capital deployment led to widespread failures and a prolonged market downturn.

However, he added that current market moves remain "far smaller" in scale than the late 1990s technology bubble. The key difference: today's infrastructure investments support actual deployed AI systems generating measurable productivity gains in specific workflows, whereas much Dot-com era spending funded speculative business models with no clear path to profitability.

The Bridgewater analysis specifically highlighted risks for AI model developers including Anthropic and OpenAI. Jensen noted that both companies will need major product breakthroughs to secure backing for massive final fundraisings ahead of potential IPOs.

Without a credible path to outsized profits, these companies could struggle to justify lofty valuations and heavy capital demands, particularly as competition from well-funded incumbents like Google, Microsoft, and Meta intensifies. The analysis suggests that the next 18-24 months represent a critical window for AI startups to demonstrate sustainable business models before capital availability potentially tightens.

Bridgewater's note highlighted financial behavior that's starting to look different from prior technology investment cycles. Companies are curbing share buybacks more aggressively to preserve cash for capital expenditure, and there's growing reliance on outside capital—including debt financing and strategic partnerships—to keep the buildout moving.

Meta's announcement Tuesday of a potentially $100 billion AMD chip deal exemplifies this pattern. The agreement includes performance-based warrants that could give Meta up to 10% ownership in AMD, effectively using equity instruments to finance hardware purchases and align chip supplier incentives with customer success.

The broader signal for the AI ecosystem is clear: the platform layer is doubling down on infrastructure first, and the pressure is shifting to the application layer to prove it can generate returns sufficient to justify the massive capital deployment.

While hyperscalers build out compute capacity at unprecedented scale, software companies and AI application developers face mounting expectations to demonstrate that AI capabilities translate into measurable business value—not just productivity improvements, but actual revenue growth and margin expansion.

For investors, the Bridgewater warning suggests careful evaluation of which portfolio holdings are built on assumptions that may not survive a shift in AI investment sentiment. Companies positioned as infrastructure providers (chipmakers, data center operators, power utilities serving AI facilities) may see more durable demand than application-layer software vendors whose AI features prove less differentiated than anticipated.

The analysis also raises questions about sustainability. Can $650 billion in annual AI infrastructure spending continue growing at similar rates through 2027 and beyond? Or does 2026 represent a peak investment year before deployment catches up with capacity and growth rates normalize?

Jensen's characterization of the current phase as "more dangerous" suggests Bridgewater sees meaningful probability that AI infrastructure investment growth is not sustainable at current trajectories, making timing and positioning critical for investors navigating the next 12-24 months.