Leopold Aschenbrenner, a 24-year-old former OpenAI researcher fired in 2024, published a 165-page manifesto arguing that the real bottleneck on AGI would be electricity — not algorithms, not chips. He then created a hedge fund, bet $878 million of it on Bloom Energy, and watched that position grow to approximately $2.73 billion after a 1,400% rally in one year. His Q1 2026 13F filing confirms the thesis: the AI buildout is an energy trade first, and a software trade second.
🔍 THE BOTTOM LINE
A 24-year-old saw what most of Wall Street missed: the AI race isn’t won by the best model — it’s won by whoever can find the power to run it. Aschenbrenner’s $878M bet on Bloom Energy proved his thesis in the most undeniable way markets have: with money.
The Situational Awareness Thesis
Aschenbrenner’s 2024 essay, Situational Awareness, laid out a simple but terrifying trajectory:
| Year | Largest AI Training Cluster | Power Required | Equivalent To |
|---|---|---|---|
| 2022 | GPT-4 cluster | ~10 megawatts | Small substation |
| 2024 | Leading cluster | ~100 megawatts | City block |
| 2026 | Current front-runner | 1 gigawatt | Hoover Dam |
| 2028 | Projected | 10 gigawatts | Most US states’ total generation |
| 2030 | Single training cluster | 100 gigawatts | 20% of all US electricity |
By 2030, a single training run would cost over a trillion dollars in electricity alone. And that’s just training — inference requires multiples on top.
Meanwhile, US electricity production has grown barely 5% in a decade. The grid was not built for this.
Bloom Energy: The Bet That Proved It
Aschenbrenner began accumulating Bloom Energy (NYSE: BE) in the mid-teens through 2025. Bloom makes solid oxide fuel cells that generate power directly at data center sites — bypassing the overloaded grid entirely.
By the end of Q1 2026, Situational Awareness LP held:
| Holding | Value (Q1 2026) |
|---|---|
| Bloom Energy | $878.7M → ~$2.73B |
| CoreWeave | $556.1M |
| Iren Limited | $401M |
| Core Scientific | $389.1M |
| Applied Digital | $320M |
| Riot Platforms | $142.2M |
| Cleanspark | $104.5M |
| AMD | $20.2M |
| Taiwan Semi | $7.6M |
The entire portfolio reads as a single bet: AI compute infrastructure, not AI model companies.
Bloom Energy’s Q1 2026 earnings showed revenue up 130% year-over-year, driven by an Oracle deal to deploy 2.8 gigawatts of fuel cells across AI data centers. The stock has climbed 1,400% in the past year.
The rest of the portfolio follows the same logic:
- CoreWeave ($556M) — GPU cloud infrastructure
- Iren Limited ($401M) — Bitcoin mining → AI compute conversion play
- Core Scientific ($389M) — Mining co-location pivoting to AI hosting
- Applied Digital ($320M) — Data center developer
- Riot Platforms + Cleanspark ($247M combined) — Power infrastructure that can be redirected to AI compute
Every single one is a power and compute play. Zero pure model companies.
What This Means for the AI Industry
Aschenbrenner’s thesis has shifted from fringe provocation to market consensus faster than almost anyone expected. The implications are stark:
The grid can’t keep up. Half of US data centers planned for 2026 are already delayed or cancelled due to power constraints. The industry is hitting a wall that more efficient chips can’t solve — because the wall isn’t compute, it’s electrons.
On-site generation is the only answer. Bloom Energy’s fuel cells, natural gas peakers, and small modular nuclear reactors are the hot assets because they solve the grid problem. The Oracle-Bloom deal for 2.8 gigawatts is the shape of things to come.
The winners aren’t the AI companies. The trillion-dollar AI buildout is an infrastructure trade. Companies that own power generation, data center real estate, and compute fabric will capture more value than any model developer — because they own the bottleneck.
What About NZ?
New Zealand has a unique advantage here. With ~80% renewable generation and some of the cheapest electricity in the OECD, NZ could become a natural hub for AI inference workloads — particularly as training shifts to places that can stomach gigawatt-scale power consumption.
The catch? Transpower’s grid capacity is already strained. NZ’s data center boom (CDC, Datagrid, Microsoft’s hyperscale plans) will need distributed generation or grid upgrades on a scale not seen since the Think Big projects of the 1980s.
One to watch: if Aschenbrenner’s “100 GW by 2030” projection holds, countries with abundant renewable energy and stable grids become the Saudi Arabia of AI.
🔍 THE BOTTOM LINE
A 24-year-old ex-OpenAI researcher published an essay, bet his reputation and his investors’ capital on the most unfashionable trade in tech — fuel cells — and is now sitting on one of the best-performing portfolios of the AI era. Situational Awareness LP has grown from $225 million to reportedly $13.7 billion in assets in a little over a year.
The lesson isn’t that Bloom Energy is a good stock. The lesson is that the AI industry’s next bottleneck isn’t intelligence — it’s physics.
❓ Frequently Asked Questions
Q: Is Leopold Aschenbrenner a 24-year-old running a $13B fund? Yes. He was a researcher at OpenAI, published the Situational Awareness essay in 2024, and launched Situational Awareness LP, a long/short equity hedge fund focused on AI infrastructure. The fund’s Q1 2026 13F shows equity holdings of ~$3.86 billion in long positions across 26 stocks.
Q: What is Bloom Energy and why does it matter for AI? Bloom Energy makes solid oxide fuel cells that generate electricity on-site at data centers. They don’t need the grid — they convert natural gas or hydrogen into power at the facility. For AI data centers facing wait times of 4-7 years for grid interconnection, on-site generation is a competitive necessity.
Q: Does Aschenbrenner’s portfolio predict the AI winners? His portfolio predicts who profits from the AI buildout, not which AI companies win the model race. By betting on power generation (Bloom), compute infrastructure (CoreWeave), and data center real estate (Applied Digital), he’s essentially short the idea that model improvements alone will determine value creation. So far, that thesis has beaten the market.
Q: How does this affect NZ? NZ’s renewable energy surplus and stable grid make it an attractive location for AI inference processing. However, current grid capacity would not support training-scale clusters without significant investment. NZ could position itself as a “green AI” hub — inference processed with 100% renewable energy — which carries a carbon premium for ESG-conscious enterprises.
Q: Is Aschenbrenner’s fund open to retail investors? No. Situational Awareness LP is a private investment fund (typically accredited/qualified investors only). However, many of its public stock holdings — Bloom Energy, CoreWeave, AMD, Taiwan Semi — can be purchased by retail investors through standard brokerage accounts.
📰 SOURCES
- Milk Road AI: Aschenbrenner portfolio breakdown
- Benzinga: 24-Year-Old Turns $875M Into $2.2B Via Bloom Energy Stock
- Anna Coulling: Situational Awareness LP Q1 2026 13F Analysis
- 13fai.com: Leopold Aschenbrenner Portfolio
- Bankless: Aschenbrenner Files 13F Quarterly Disclosure
Disclaimer: This article is for informational purposes only and does not constitute investment advice. Past performance does not guarantee future results.
