Thesis: AI Infrastructure Innermost Loop

The structural argument

Four interlocking supply chains are simultaneously hitting escape velocity, with demand at each layer driving the next:

1. Frontier labs (OpenAI, Anthropic, xAI, DeepMind) build the models
2. Chips (NVIDIA, AMD, Broadcom, Micron, TSM) train and serve them
3. Data center infrastructure (Vertiv, GE Vernova, Bloom, Oklo, Fluence) houses the chips
4. Power (Talen, Constellation, Vistra) feeds the data centers

The bottleneck has migrated from chips to power. $156B of US data-center projects are blocked or stalled on local power-infrastructure opposition. That makes the power layer the highest-asymmetry public play because it's the named constraint.

This is structurally different from prior tech cycles because four supply chains are accelerating simultaneously instead of one at a time.

Full reference: [[06-reference/2026-05-12-diamandis-innermost-loop-ai-infrastructure-thesis]].

Demand-side anchors (the disqualifying conditions)

This thesis holds as long as ALL three anchor metrics remain on trajectory. If any one breaks materially (>30% miss to projection for 2+ consecutive quarters), the thesis archives.

Position-sizing logic by layer

The four layers do NOT carry equal forward leverage. Sizing should reflect:

Layer 4 — Power (highest forward leverage):

• Named bottleneck in the thesis
• Tickers: TLN, GEV, OKLO, BE, FLNC
• 12-mo returns: TLN +73%, GEV +158%, OKLO +178%, FLNC +200%, BE +1,647% (extended)
• Position-sizing call: BE is extended (avoid top-tick); TLN / GEV / OKLO / FLNC have more runway. Equal-weight basket of those four is the cleanest exposure.

Layer 3 — Data center infra (mid leverage, late entry risk):

• Tickers: VRT and adjacent
• 12-mo: VRT +256%
• Position-sizing call: small starter position only; the trade is "trend continues but entry was 12 months ago."

Layer 2 — Chips (lowest forward leverage, most picked-over):

• Tickers: MU, INTC, AMD, TSM, AVGO, NVDA
• 12-mo avg: ~320%
• Position-sizing call: NVDA and AVGO are the structural holders (highest market caps, deepest moats). MU and INTC are momentum-rich; don't chase. Position should reflect "long-term substrate exposure" not "the trade."

Layer 1 — Frontier labs (private secondaries):

• Anthropic ($1T secondary, May 2026), OpenAI ($852B → likely $1T IPO), xAI ($230B)
• Access: Forge Global, Hiive, or direct allocations. Accreditation required. $50k-100k minimums.
• Position-sizing call: only relevant if the founder is qualified + willing to lock capital for years. If so, Anthropic at current secondary is the cleanest entry given recent compute resolution via xAI partnership (see [[06-reference/2026-05-12-stratechery-spacex-anthropic-xai-musks-two-companies]]).

The "I missed it" question

Most of these tickers are up 100-1600% over 12 months. The honest framing is NOT "I'm getting in at the bottom" — it's "this is a multi-year buildout where the trend continues even if I missed the first leg." Three reasons this defense holds for now:

1. The demand-side anchors are still accelerating, not decelerating
2. The power-layer bottleneck specifically has not been resolved — capex is being committed faster than power is coming online
3. Q1 2026 AI-startup funding doubled all of 2025 in three months — capital flow is intensifying, not slowing

If the demand-side anchors flatten, this defense breaks and the thesis transitions to "exit / lock in gains" mode.

Open candidate list

(See ~/rdco-vault/01-projects/investing/candidates/ for per-ticker pre-decision research as it accumulates.)

Initial candidates from this thesis:

• TLN (power, nuclear)
• GEV (power, turbines)
• OKLO (power, advanced nuclear)
• FLNC (power, storage)
• NVDA (chips, substrate)
• AVGO (chips, custom ASICs)
• Anthropic secondary (frontier lab, private)

Next decisions

No active capital-deployment decision queued yet. When founder is ready to deploy:

1. Pick a sizing for the Layer 4 basket
2. Spawn /decisions/<date>-investing-innermost-loop-power-basket.html on HQ for the actual deployment gate
3. Log the position in positions/ after execution

Related

• [[06-reference/2026-05-12-diamandis-innermost-loop-ai-infrastructure-thesis]] - full Diamandis source synthesis
• [[06-reference/2026-04-30-not-boring-scarce-assets-abundance-driven-scarcity]] - Packy abundance-driven scarcity framework (validates the bottleneck-migration thesis)
• [[06-reference/2026-05-12-stratechery-spacex-anthropic-xai-musks-two-companies]] - frontier-lab compute supply landscape
• [[01-projects/investing/README]] - project home