Summary
Bricks & Bytes synthesis arguing that the gap between announced and actually-built AI data center capacity is the critical metric for assessing the buildout. Cites Sightline Climate showing only 5 GW of the 16 GW global 2026 pipeline is under construction; Wood Mackenzie data on transformer lead times; the Stargate Abilene cancellation; and Michael Burry’s depreciation thesis (~$176B in understated depreciation through 2028 if AI hardware’s actual economic life is 2-3 years vs. the 5-6 year industry standard). Identifies the binding constraint as having shifted from semiconductor availability to electrical grid capacity and equipment lead times.
Key Points
- 2025 capex: ~$400 billion committed to AI infrastructure.
- 2026 announced capex: $650 billion (60% YoY increase).
- Global 2026 pipeline status: 5 GW of 16 GW under construction (per Sightline Climate).
- Cancellation projection: Sightline Climate expects 30-50% of the 2026 pipeline to slip or cancel outright.
- Stargate Abilene: Oracle/OpenAI 600 MW expansion scrapped March 2026 after financing collapse.
- Transformer lead times (corroborates Wood Mackenzie): 128 weeks for power transformers; 144 weeks for GSUs.
- Equipment pricing: Up 70-150% since 2020.
- Electricity costs: Virginia +267% over 5 years; US natural gas +60% vs. 2024.
- Burry depreciation thesis: Hyperscalers depreciate AI hardware over 5-6 years when actual economic life is 2-3 years; ~$176B understated depreciation through 2028.
Newsletter Angles
- The argument is being made adjacent. Bricks & Bytes is a contemporary publication making structurally the same argument as the May 15 piece in development. This is good: the article doesn’t have to claim isolation. It can position as “here is the layer-by-layer evidence” while citing Bricks & Bytes as adjacent corroboration.
- The Burry depreciation angle is a separable thread. Worth a different piece on its own — if AI hardware’s economic life is 2-3 years (not 5-6), then the $176B depreciation gap intersects with the buildout-delay story in a way that compounds: capacity that takes 7 years to deliver but depreciates in 2-3 years has ROI math that doesn’t close. Possible Q3 paid-tier deep dive.
- The scope mismatch. Bricks & Bytes uses 5 GW / 16 GW global. Almost Half of US Data Centers That Were Supposed to Open This Year Slated to Be Canceled or Delayed — Futurism - 2026-04-02 uses 4 GW / 12 GW US. Both can be true (different scopes). Cite carefully: the article should use the US numbers as the primary headline since the substation/PPA story is US-specific.
Entities Mentioned
- Sightline Climate — analytics firm tracking pipeline status (deferred stub; recurring)
- Wood Mackenzie — transformer lead-time data source
- Oracle — Stargate Abilene cancellation party
- OpenAI — Stargate Abilene cancellation party
- Michael Burry — investor; depreciation thesis (deferred stub)
- Aren Deu — Bricks & Bytes author
Concepts Mentioned
- AI Buildout Grid Constraint — directly thematic
- Hyperscaler Hardware Depreciation Gap — Burry thesis (deferred stub; candidate concept page)
- AI DRAM Crisis — Bricks & Bytes’s “binding constraint shifted from semis to grid” framing implicitly references the prior chip-layer tightness this newsletter has covered
Quotes
“If one piece of your supply chain is delayed, then your whole project can’t deliver.” — Sightline Climate analyst (per Bricks & Bytes attribution)
“The gap between announcement and shovel is now the biggest signal for anyone building, financing, or investing in data center capacity.” — article assertion
Notes
Bricks & Bytes is a relatively new publication; treat as secondary synthesis of primary sources (Sightline Climate, Wood Mackenzie, Bloomberg). The Likens quote attributed by Bricks & Bytes to “a Sightline Climate analyst” is the same quote attributed to Andrew Likens of Crusoe Energy in Almost Half of US Data Centers That Were Supposed to Open This Year Slated to Be Canceled or Delayed — Futurism - 2026-04-02. The Futurism attribution (Likens, Crusoe Energy) is more likely the original; verify before quoting in any article.
The 5 GW / 16 GW figure (Bricks & Bytes) and 4 GW / 12 GW figure (Sightline via Bloomberg/Futurism) are reconcilable as global vs. US scopes, but the article in development should anchor on the US figures since the supporting evidence (PPAs, PUCO filings, Wood Mackenzie transformer data) is US-centric.
⚠️ Attribution discrepancy noted: same quote (“If one piece of your supply chain is delayed…”) attributed to two different people across Almost Half of US Data Centers That Were Supposed to Open This Year Slated to Be Canceled or Delayed — Futurism - 2026-04-02 (Andrew Likens, Crusoe Energy) and this Bricks & Bytes piece (unnamed Sightline Climate analyst).