An Aschenbrenner-Style AI Infrastructure Investment Strategy for 2028

Damnang · Damnang’s Substack · May 05, 2026 at 07:21 · ⏱ 23 min read  | Read on Substack ↗
Summary
The article argues that Leopold Aschenbrenner's infrastructure bottleneck thesis for AI investing has been validated, but the specific bottlenecks have shifted from power generation to power distribution, cooling, optical bandwidth, HBM, and advanced packaging. For traders, this means the highest-return opportunities are now in companies that own these physical bottlenecks — not in GPU cloud providers or model companies — but the author explicitly presents this as a thought experiment, not a recommendation.
  • Aschenbrenner's fund grew from $254.8M to $5.52B in 13F book value, but $1.76B was new capital and options accounted for ~30% of the book, not pure alpha.
  • Hyperscaler combined CY2026 CAPEX guidance is $745-775B, up from $256B in 2024, and the IEA says it now exceeds global upstream oil and gas investment.
  • Power transformer lead times are ~128 weeks and high-voltage switchgear 2-3 years, making power distribution equipment the binding bottleneck in 2026.
  • HBM is sold out across all three vendors for CY2026; HBM3E contract prices are up ~20% YoY with a 5-6x premium vs DDR5.
  • Co-packaged optics (CPO) arrived early — NVIDIA integrated it into post-Rubin networking at GTC 2026 — pulling forward optical component demand.
  • DeepSeek showed equivalent GPT-4 performance at 1/18th inference cost, but Jevons' paradox is winning: AI data center power demand grew 50% in 2025.
  • The author's thought experiment basket: core = Eaton, Powell, Corning, Micron, Amkor; satellites = Vertiv, Bloom, Modine, Lumentum — explicitly not a recommendation.
  • CoreWeave and IREN are critiqued as not owning physical bottlenecks; their GPU cloud business is exposed to GPU price declines and customer concentration.
Read time 23 min
Length 23,891 chars
Category finance
Trade Ideas
Damnang Substack author, Damnang’s Substack
Powell Industries is described as a pure-play power distribution equipment specialist with an all-time high backlog of $1.6B and its first data center megaproject booked in Q1 FY2026. The article posi
Powell Industries is described as a pure-play power distribution equipment specialist with an all-time high backlog of $1.6B and its first data center megaproject booked in Q1 FY2026. The article positions it as the most direct recipient of the switchgear bottleneck. Risk: Small-cap volatility; customer concentration; execution risk on megaprojects.
Damnang Substack author, Damnang’s Substack
Amkor is the only U.S.-listed name with direct exposure to advanced packaging. Article notes TSMC outsourcing ~240K wafers to Amkor and SPIL, and Amkor tripling CAPEX to $2.5-3.0B to absorb demand. Ar
Amkor is the only U.S.-listed name with direct exposure to advanced packaging. Article notes TSMC outsourcing ~240K wafers to Amkor and SPIL, and Amkor tripling CAPEX to $2.5-3.0B to absorb demand. Arizona fab adds CHIPS Act subsidy and geopolitical value. Risk: Heavy CAPEX cycle; customer concentration risk (TSMC); technology transition risk if packaging methods change.
Damnang Substack author, Damnang’s Substack
Corning is flagged as a different kind of bottleneck: fiber optic cable that lasts decades and scales with AI cluster size. The $6B long-term deal with Meta and 36% Optical Communications revenue grow
Corning is flagged as a different kind of bottleneck: fiber optic cable that lasts decades and scales with AI cluster size. The $6B long-term deal with Meta and 36% Optical Communications revenue growth confirm hyperscalers are locking in fiber supply years ahead. Risk: Commodity fiber pricing pressure; slower-than-expected cluster deployment.
Damnang Substack author, Damnang’s Substack
Micron holds pricing power in HBM, sold out for CY2026 with 5-6x premium vs DDR5. The article highlights its divestment of consumer memory to go all-in on AI data center, structurally changing its ear
Micron holds pricing power in HBM, sold out for CY2026 with 5-6x premium vs DDR5. The article highlights its divestment of consumer memory to go all-in on AI data center, structurally changing its earnings profile toward long-term contracts and supplier pricing power. Risk: Cyclical memory downturn risk; HBM4 transition may create temporary oversupply if demand wavers.
Damnang Substack author, Damnang’s Substack
Article identifies Eaton as a core bottleneck holder because it covers the full power path from grid to chip, with data center orders up +200% YoY and a new factory in Nebraska. The thesis is that pow
Article identifies Eaton as a core bottleneck holder because it covers the full power path from grid to chip, with data center orders up +200% YoY and a new factory in Nebraska. The thesis is that power distribution equipment (switchgear, distribution panels) is the binding constraint in 2026. Risk: If efficiency gains or regulatory pushback slow AI buildout, demand could moderate. Large-cap multiple compression also a risk.
Damnang Substack author, Damnang’s Substack
CoreWeave is criticized for not owning physical bottlenecks; it leases GPUs and has 62-72% revenue concentration on Microsoft with $25B in debt-financed GPU procurement. Article quotes Jim Chanos: 're
CoreWeave is criticized for not owning physical bottlenecks; it leases GPUs and has 62-72% revenue concentration on Microsoft with $25B in debt-financed GPU procurement. Article quotes Jim Chanos: 'return on capital is roughly 0%.' It is positioned as an optionality play, not a core holding. Risk: GPU price declines, customer churn, high leverage, and competition from hyperscaler in-house clouds all threaten the business model.
More from Damnang’s Substack

This newsletter, published May 05, 2026, features Damnang discussing POWL, AMKR, GLW, MU, ETN, CRWV. 6 trade ideas extracted by AI with direction and confidence scoring.

Speakers: Damnang  · Tickers: POWL, AMKR, GLW, MU, ETN, CRWV