Leopold Aschenbrenner's Situational Awareness thesis argues AGI will arrive by 2027 through compounding gains in compute scaling (~0.5 OOMs/year), algorithmic efficiency (~0.5 OOMs/year), and capability "unhobbling." His fund, Situational Awareness LP, grew from $225M to $5.5B in disclosed equity exposure in just over one year, returning +47% net of fees in H1 2025.
The critical insight: the binding constraint on AI progress is shifting from GPUs to power and physical infrastructure. By 2028, individual training clusters may require $100B+ in investment and consume power equivalent to a small U.S. state.
| Year | H100 Equiv. | Power | Est. Cost | Power Reference |
|---|---|---|---|---|
| 2022 (GPT-4) | ~10,000 | ~10 MW | ~$500M | 10,000 homes |
| ~2024 | ~100k | ~100 MW | $Billions | 100,000 homes |
| ~2026 | ~1M | ~1 GW | $10sB | Hoover Dam |
| ~2028 | ~10M | ~10 GW | $100sB | Small U.S. state |
| ~2030 | ~100M | ~100 GW | $1T+ | >20% U.S. electricity |
Key Insight: The scarce resource in AI is not GPUs—it is gigawatts of permitted, grid-connected power. Bitcoin miners own exactly this.
The Q4 2024 13F showed a semiconductor-heavy portfolio (NVIDIA, Intel, Broadcom). By Q4 2025, the fund pivoted wholesale to the infrastructure layer—power generation, data center real estate, and connectivity. NVIDIA puts worth $300M were sold; positions in Bloom Energy, CoreWeave, and Bitcoin miners were accumulated.
The Bitcoin Mining "Shortcut"
Bitcoin miners possess two assets AI data centers desperately need: secured land with proper zoning and existing power interconnection agreements that took years to obtain. Acquiring a miner is like taking over a bar that already has a liquor license—a shortcut through regulatory timelines that hyperscalers cannot replicate.
Mining companies could derive 70% of revenue from AI/HPC by end-2026, up from ~30% at start of year.
| Company | Ticker | Category | Est. Alloc. |
|---|---|---|---|
| Bloom Energy | BE | Power Gen | 15.9% |
| CoreWeave (Call) | CRWV | Compute | 14.0% |
| Intel (Call) | INTC | Semiconductors | 13.6% |
| Lumentum | LITE | Connectivity | 8.7% |
| CoreWeave | CRWV | Compute | 7.9% |
| Core Scientific | CORZ | Bitcoin Miner / Compute | 7.6% |
| IREN | IREN | Bitcoin Miner / Compute | 6.0% |
Top Position: Bloom Energy
The fund's largest holding (~16%) embodies the core thesis. Bloom's solid oxide fuel cells convert natural gas to electricity at data center sites, bypassing the congested grid entirely. The company reported $20B order backlog and 34% revenue growth in 2025, with 40% growth guided for 2026. The stock returned +1,280% over 12 months.
Core Scientific (CORZ) — AI Pivot Leader
Core Scientific is the most advanced conversion story. The company secured a $3.3B financing package and a landmark $10.2B, 12-year hosting contract with CoreWeave. Its Pecos, Texas campus is being repurposed from 300 MW of mining to 1.5 GW (1 GW leasable for AI)—a 5x expansion. AI/HPC generated 39% of Q4 2025 revenue.
IREN Limited (IREN) — AI Infrastructure King
IREN's $3.4B GPU cloud contract with NVIDIA combined with a $9.7B Microsoft agreement brings contracted revenue to $13B+. The company secured 5 GW of AI capacity—one of the largest single-provider deployments globally. 52-week return: +715%.
Cipher Digital (CIFR) — Hyperscaler Gateways
Cipher secured $9.3B in long-term leases with AWS and Google covering a fraction of its 4+ GW pipeline. Trades at ~10x NOI from existing leases. Nine "Buy" ratings; median target $24 implies ~68% upside.
| Company | Price | Mkt Cap | Fwd P/E | 52W Chg | AI Rev% |
|---|---|---|---|---|---|
| Bloom Energy | $261 | $74.2B | 63.2x | 1,280% | ~95% |
| IREN | $61.20 | $20.3B | 48.4x | 715% | ~60% |
| Cipher Digital | $20.55 | $8.4B | 45.0x | 552% | ~15% |
| Riot Platforms | $24.08 | $9.1B | N/M | 177% | ~10% |
| Core Scientific | $22.92 | $7.3B | 43.7x | 132% | ~39% |
| CleanSpark | $14.20 | $3.6B | N/M | 48% | ~5% |
| MARA Holdings | $12.94 | $4.9B | 3.6x | -19% | ~2% |
The Bull Case
Power is the new oil of the AI age. The investment case rests on: (1) hyperscalers announced $1.4T+ in computing partnerships but face 5-7 year power interconnection timelines; (2) miners already possess these scarce grid connections; (3) AI hosting economics significantly exceed Bitcoin mining returns; (4) revenue mix shifts from volatile crypto to stable long-term contracts should drive multiple expansion.
The Key Differentiator: Power Assets
The investment framework is simple: value the power, not the mining equipment. Grid interconnection agreements, land permits, and utility contracts that took years to secure are worth far more than ASIC rigs.
Risk Assessment
| Risk | Severity | Mitigation |
|---|---|---|
| Execution: Facility conversion delays | High | Focus on contracted-revenue (IREN, CORZ) |
| Capital Intensity | High | Monitor balance sheets and financing access |
| Power Market Volatility | Med | Prefer long-term fixed-price PPAs |
| Concentration Risk | Med | Track contract diversification |
| Bitcoin Price Decline | Med | Prioritize advanced AI revenue mix |
Mining companies with secured power, hyperscaler contracts, and conversion capability are critical infrastructure assets in the AI age—not speculative crypto plays. The window is narrowing (IREN +715%, Bloom +1,280% in 12 months) but with trillions in planned hyperscaler capex yet to deploy, the runway extends several years for positioned investors.
Standard Kepler Research | standardkepler.com