• In a June 30 Substack post, "Big Short" fund manager Michael Burry disclosed a new short in Caterpillar (CAT) at $1,060.98, his first position in the name, arguing CAT's price-to-sales ratio is at a multi-decade high (CNBC, https://www.cnbc.com/2026/06/30/burry-shorts-caterpillar-after-it-nearly-doubled-in-ai-rally-of-2026.html).
• CAT had nearly doubled in 2026 on AI-infrastructure/data-center spending exposure; Burry simultaneously disclosed shorts in NVDA, AMAT, TSLA and the SOXX semis ETF, framing a broader AI-bubble fade trade.
• The disclosure hit social media and trading desks in late afternoon, adding to already-skittish AI-trade sentiment heading into July 4.

Caterpillar's AI-rally surge isn't about selling AI chips — it's about the power bottleneck behind AI data centers, and CAT is one of the largest US-listed "picks-and-shovels" beneficiaries of that buildout. The trade works through several channels:

1. Prime & backup power generation (the biggest thread) - AI data centers need gigawatts of reliable, fast-ramping power, and utilities can't deliver grid connections fast enough. Hyperscalers are increasingly using on-site gas generation as the bridge. - CAT owns Solar Turbines, one of the few Western makers of large industrial gas turbines (50+ MW-class machines) used for behind-the-meter data center power. Their large reciprocating engine platforms (G3520K, G3516, etc.) are sold in multi-unit arrays as prime or backup power for hyperscale campuses. - As Microsoft, Google, Amazon, and Meta scrambled to bring new capacity online through 2025–26, Solar Turbines' order book exploded and management has guided to multi-year backlogs.

2. The broader "electrification / power T&D" theme - The same data center buildout drove demand across the electrical infrastructure complex (transformers, switchgear, gas-insulated substations, turbines). Investors rotated from pure-play AI chip names (NVDA, AVGO) into liquid industrial "power" beneficiaries. CAT is the most recognizable US large-cap in that bucket — that's why Burry's other shorts (NVDA, AMAT, TSLA, SOXX) sit alongside it: it's the same thesis expressed on the supply side of the AI buildout rather than the chip side.

3. Construction & site-prep demand (smaller but real) - CAT's earthmoving equipment (dozers, excavators, motor graders, off-highway trucks) is used for land grading and site work on hyperscale data center campuses and on the gas-generation sites behind them — a modest tailwind for the Construction Industries segment.

4. Multiple expansion, not just earnings - CAT's underlying earnings growth from this has been real but modest. The bulk of the 2026 doubling came from P/S re-rating as a staid old-economy industrial was priced like an AI infrastructure name — exactly Burry's complaint. Before 2026, CAT had traded at roughly 1.5–2.5× sales for years; during the AI-power bubble it touched a level last seen decades ago.

Why Burry thinks it's stretched - His argument is that the market is pricing CAT for a sustained, decades-long data center power capex super-cycle, when in reality Solar Turbines' incremental upside is lumpy, lopsided to a handful of mega-deals, and competitive (with GE Vernova's LM6000/Simple Cycle turbines and Mitsubishi Heavy Industries as the main Western alternatives). If hyperscaler capex normalizes — or if AI workload efficiency gains reduce the per-query power requirement — CAT's "AI premium" compresses quickly.

So the connection is indirect but mechanical: CAT doesn't sell a single chip, but its turbines and engines are now one of the rate-limiting inputs to the AI buildout, and in 2026 the market decided to pay for that link at a multiple it hadn't used since the late 1990s.