pulse note desk

Burry's Parabolic Warning Is Late-Cycle Noise — AI's Cash Register Is Wide Open

Michael Burry sees dot-com echoes in every soaring chart. The numbers say this time the infrastructure is actually getting built — and paid for.

You've heard the pitch from Burry's latest Substack: reject greed, slash tech exposure, especially anything going parabolic. The market feels like the final months of 1999-2000, with non-stop AI chatter and momentum detached from reality. Fair warning from a guy who nailed the housing bust. But here's the punchline: Burry is fighting the last war while hyperscalers are wiring nine-figure checks for GPUs and data centers that are already generating real revenue.

Let's start with the scoreboard everyone is ignoring. Hyperscalers — Amazon, Microsoft, Google, Meta — have committed roughly $700-725 billion in combined 2026 capex, up massively from prior years. Microsoft alone guided $190 billion for the year, up 61%. Amazon is at $200 billion. These aren't PowerPoint dreams; they're line items tied to exploding cloud workloads. Amazon's AWS grew 28% in the latest quarter, its fastest pace in 15 quarters, while its in-house chip business hit a $20 billion run rate growing triple digits. Microsoft's AI business surpassed a $37 billion annual run rate, up 123% YoY.

Compare that to the dot-com era Burry references. Back then, the leaders chased eyeballs with negative unit economics. Today, NVIDIA reported fiscal Q4 2026 revenue of $68.1 billion, up 73% YoY, with data center at $62.3 billion, up 75%. Full-year revenue hit $215.9 billion, up 65%. Gross margins are in the mid-70s on real product shipments, not speculative land grabs. The Philadelphia Semiconductor Index (SOX) is up around 65-70% YTD in 2026 on this momentum, yet the buyers keep raising guidance instead of tapping the brakes.

That's the deadpan fact bomb: Burry correctly flagged housing once. Since then, his big-tech timing has been louder than profitable. His latest 13F shows a rotation into value names like Molina Healthcare and Lululemon rather than a concentrated short book against the AI leaders. Smart positioning for his style, but it doesn't disprove the tangible demand cycle unfolding in front of us.

The variant perception here is straightforward. Consensus hears "parabolic" and reaches for the 2000 playbook. The market is wrong because AI infrastructure has verifiable economics that the last bubble never did. GPU and memory demand is tied to actual training and inference workloads that hyperscalers are monetizing today. Software layer players are posting 30-60%+ revenue growth in recent prints — think coding agents and enterprise tools hitting nine-figure run rates in months, not years. Valuations on forward earnings for the leaders have actually compressed despite the price gains, with Morningstar noting the AI theme trading at its largest discount to fair value since 2019.

You don't need faith when the order book is this loud. Supply chain signals, shipment data, and backlog growth all point to multi-year digestion of this capex wave. Power and cooling are the real constraints, not demand. Companies like Equinix and component suppliers are seeing the spillover. This isn't vaporware — it's a capital cycle with positive gross margins and customers who print cash.

Of course, trees don't grow to the sky. The thesis has clear kill criteria. If Q2 or Q3 2026 earnings show major hyperscalers or NVDA/AMZN/MSFT cutting AI capex guidance or delivering under 20% YoY revenue growth in cloud/AI segments, the narrative shifts hard. A confirmed >25% drop in SOX or NVDA by August on customer pauses or weak shipment data would validate Burry's caution. Broad tech gross margin contraction over 300 basis points with rising inventory would signal the froth winning.

But right now, the data doesn't support ringing the bell on the whole theme. Selective parabolic names in the actual infrastructure stack keep running while the speculative froth elsewhere burns off. Burry's call is classic late-cycle noise — the kind of voice that sounds wise in every bull market but misses when the underlying unit economics have flipped from speculative to structural.

Reality is the punchline. The same crowd that doubted the capex numbers last year is now watching them get raised again while earnings compound. You don't fight momentum when the customers are the richest companies on earth wiring money for capacity they can't add fast enough. Position accordingly: own the picks and shovels with real traction, trim the pure hype, and watch the numbers, not the headlines.

key takeaways

  • Hyperscalers (AMZN, MSFT, GOOG, META) committed $700-725 billion in 2026 capex, with Microsoft at $190B (+61%) and Amazon at $200B.
  • NVIDIA reported fiscal 2026 revenue of $215.9 billion (+65% YoY), with data center revenue at $62.3B in Q4 alone (+75%).
  • AWS posted 28% growth — its fastest pace in 15 quarters — while Microsoft's AI business hit a $37B run rate (+123% YoY).
  • Unlike the dot-com era's negative unit economics, today's AI infrastructure shows high-70s gross margins and verifiable monetization.
  • AI leaders are trading at compressed forward valuations and their largest discount to fair value since 2019 per Morningstar.

faq

Why does the article claim Michael Burry's AI warning is wrong?

Burry is comparing today's AI boom to the dot-com bubble, but unlike 1999-2000 when companies chased eyeballs with negative economics, current AI spending is backed by exploding cloud revenue, massive real capex, and high-margin product shipments already generating billions.

How much are hyperscalers spending on AI infrastructure in 2026?

Combined capex from Amazon, Microsoft, Google, and Meta is projected at roughly $700-725 billion for 2026, with Amazon guiding $200 billion and Microsoft $190 billion (up 61%).

What were NVIDIA's key financial results for fiscal 2026?

NVIDIA reported full-year revenue of $215.9 billion, up 65% YoY. In Q4, total revenue reached $68.1 billion (+73%) with data center revenue at $62.3 billion (+75%).

Is the AI market showing signs of a bubble according to the article?

No. The article highlights real demand from training and inference workloads, strong backlog growth, supply chain signals, and forward earnings valuations that have actually compressed despite price gains.