Greg Jensen, Bridgewater's co-chief investment officer, wrote in a client letter that compute demand continues to significantly outpace supply, driving hyperscalers to invest even more rapidly "to try to someday get ahead of the demand." He noted that these four companies have already curbed share buybacks more aggressively to help fund the capital expenditure surge, and that growing reliance on outside capital is a warning sign.

Bridgewater estimates that tech investment contributed roughly 50 basis points to U.S. GDP growth in 2025 and could add around 100 basis points this year. But Jensen warned that this spending boom may lift inflation in technology equipment and push up electricity prices regionally. A severe stock market correction, he said, could undermine growth and limit these companies' ability to raise capital, drawing a comparison to the Dot-com bubble in 2000 (while noting current conditions are far smaller in scale).

AI Capex: 2025 vs 2026 (Bridgewater estimates)
Meta	$135B capex
Microsoft	~$80B+
Alphabet	~$75B
Amazon	~$100B+
Source: Bridgewater Associates client letter (Feb 2026), Reuters. Bars show relative 2025 → 2026 growth. Individual company figures approximate.

Meta's $100 Billion Bet on AMD

One day after the crash, on Tuesday February 24, Meta and AMD announced a deal that could be worth more than $100 billion. Meta will purchase up to 6 gigawatts worth of custom AMD Instinct GPUs across multiple generations. The first 1-gigawatt tranche, powered by custom MI450-based GPUs and 6th Gen EPYC "Venice" CPUs, ships in the second half of 2026.

In exchange, AMD issued Meta a performance-based warrant for up to 160 million shares of AMD common stock at $0.01 per share. That is approximately 10% of AMD's outstanding shares. But the shares vest only as Meta hits purchase milestones, and the final tranche requires AMD's stock to reach $600. On the day of the announcement, AMD traded at around $211. The stock jumped 7% on the news.

This deal is nearly identical to AMD's October 2025 agreement with OpenAI: same 6-gigawatt scale, same 160-million-share warrant structure. Combined, OpenAI and Meta have committed to 12 gigawatts of AMD GPU compute. For AMD, which commands less than 10% of the AI chip market compared to Nvidia's approximately 90%, this is a major shift. AMD CEO Lisa Su called it one of the "most transformational deals" in the company's history.

Meta • $135B capex in 2026 • 30 data centers planned • Nvidia deal signed Feb 10 • AMD deal signed Feb 24	6 GW of GPUs ⇄ 160M shares (~10%)	AMD • Custom MI450 GPUs • Venice + Verano CPUs • Helios rack architecture • Same deal with OpenAI (Oct 2025)
Source: AMD press release, CNBC, TechCrunch (Feb 24, 2026). Shares vest at milestones; final tranche requires AMD stock at $600.

The Circular Financing Problem

There is a pattern emerging that some analysts find concerning. Meta buys $100 billion worth of AMD chips. AMD gives Meta 10% equity. AMD's stock jumps on the announcement. Meta's equity stake becomes more valuable. Meta uses the AI infrastructure to build products, generating revenue that funds more chip purchases. The same dynamic played out with OpenAI and AMD in October.

Jay Goldberg, senior analyst at Seaport Research Partners, noted that what can look like a virtuous cycle of investment can also tip into circular financing, where the same dollars flow between companies to create the appearance of demand. Chris Miller, professor of international history at Tufts University, pointed out that AMD commands less than 10% of the AI chip market and that its software ecosystem is far less developed than Nvidia's.

The timing matters too. Meta signed a deal to deploy millions of Nvidia chips and equipment just days before announcing the AMD partnership. Meta's total planned AI infrastructure investment exceeds $600 billion over the next several years. Mark Zuckerberg framed the AMD partnership as a step toward "personal superintelligence." Gil Luria of D.A. Davidson said the deal gives Meta a diversified supply chain and reduces its dependence on Nvidia's pricing power.

Key Insight: AMD has now signed identical 6-gigawatt, 160-million-share deals with both OpenAI and Meta. That is 12 GW committed from two customers. If these milestones are met, AMD could generate "double-digit billions of revenue per gigawatt," according to an AMD spokesperson. But if the AI spending cycle slows before delivery, AMD is left with massive production commitments and two major customers holding dilutive warrants.

What the "AI Scare Trade" Actually Hit

Monday's selloff was not confined to AI companies. It hit the companies that AI might replace. The Citrini Research report specifically modeled scenarios where AI agents eliminate payment processing fees (threatening Visa and Mastercard), "vibe-coded" apps displace DoorDash and Uber Eats, and automated systems replace consulting and IT services work.

The damage was broad: IBM fell 13.15%, DoorDash dropped over 6%, American Express fell about 7%, Mastercard dropped nearly 6%, and Visa fell 4.5%. CrowdStrike lost nearly 10%. Microsoft declined about 3%. The Kobeissi Letter noted on X that "every time a new Claude-based AI tool emerges, stocks in that industry are erasing over $100 billion worth of market cap."

DoorDash co-founder Andy Fang acknowledged the shift, posting on X: "We definitely believe agentic commerce will be transformative to the industry. The ground is shifting underneath our feet, and the industry is going to need to adapt to it."

Key Insight: The market is now pricing AI as a two-sided trade. The companies building AI infrastructure (Nvidia, AMD) gain. The companies that AI could automate away (payments, consulting, delivery, legacy IT) lose. Bridgewater's $650 billion figure is not just a spending number; it is a transfer of value from one set of industries to another. The open question is whether the receiving side can generate returns fast enough to justify the transfer.

Is This 2000 Again?

Jensen himself addressed the comparison. He said a severe stock market correction could limit companies' ability to raise capital, similar to the Dot-com bubble, but added that recent moves are far smaller in scale. The key difference: in 2000, the spending was mostly on speculative internet companies with no revenue. In 2026, Alphabet, Amazon, Meta, and Microsoft are among the most profitable companies in history. They have the cash flow to sustain this, at least for now.

The risk is not that these companies go bankrupt. The risk is that $650 billion in annual spending creates a dependency loop: the spending itself contributes roughly 100 basis points to GDP growth, which supports stock prices, which enables more capital raising, which funds more spending. If any link breaks (a market correction, slower AI revenue growth, or a regulatory shock), the feedback loop reverses.

Lori Calvasina, head of U.S. equity strategy at RBC Capital Markets, wrote that investors have been "ruminating on a number of AI concerns in early 2026, with concerns about cash flow, capex levels, and whether a number of different industries will survive the AI era." She added that existential concerns for industries beyond software "have seemed overblown."

Key Insight: Bridgewater's note frames the core tension clearly. AI infrastructure spending is simultaneously the largest source of U.S. GDP growth support and the largest concentration of downside risk. The four companies driving it have the profits to sustain the pace today. The question is whether the applications built on top of this infrastructure will generate returns before investor patience runs out.

Tuesday brought a partial rebound. AMD jumped 7% on the Meta deal. Some software stocks recovered. But the underlying math has not changed: four companies are spending $650 billion per year on infrastructure for a technology whose application-layer revenue remains unclear at that scale. The "dangerous phase" Bridgewater describes is not a prediction of collapse. It is a recognition that the margin for error is shrinking as the bets get larger.

ResearchAudio.io Sources: Reuters/Bridgewater • CNBC (AMD-Meta) • Bloomberg/Yahoo (Citrini selloff) • NBC News