Former OpenAI researcher’s hedge fund pivots to AI infrastructure focus amid industry expansion

Leopold Aschenbrenner, a former researcher at OpenAI who left the lab’s superalignment team to found Situational Awareness LP, has repositioned his San Francisco‑based hedge fund around the physical backbone of large‑scale artificial intelligence, according to a report by CryptoBreaking and contemporaneous filings. The fund’s latest regulatory disclosures for the fourth quarter of 2025 show a dramatic enlargement of its U.S. equity exposure to roughly $5.52 billion across fewer than 30 positions, a sizable increase from the low‑hundreds‑of‑millions scale reported at the beginning of 2025.

Rather than following the crowded trade in consumer‑facing AI applications, Situational Awareness has concentrated capital on capacity: power generation, high‑density data centres and specialised hardware vendors that enable high‑performance computing. Top positions cited in the filings and press coverage include CoreWeave, Bloom Energy, Intel, Lumentum and Core Scientific, reflecting a thesis that ownership of energy and real‑estate constrained compute capacity will underpin value creation as AI workloads scale.

The fund’s amended Schedule 13D names a near‑double‑digit stake in Core Scientific , about 28.76 million shares representing roughly 9.4% of the company with shared voting and disposition authority , illustrating an activist bent that could shape how a prominent bitcoin miner pivots toward hosting AI and HPC customers. According to the CryptoBreaking account and the amended filing, Situational Awareness appears to be using both equity ownership and governance rights to press for a transformation of mining sites into multipurpose, megawatt‑dense compute facilities.

Industry developments beyond the filings lend context to that strategy. CNBC reported CoreWeave’s multibillion‑dollar commitment to build a 100‑megawatt data centre in Lancaster, Pennsylvania, with room to expand to 300 MW, a project designed to serve growing demand for enterprise AI compute. The business case for on‑site, reliable power is underscored by a partnership disclosure from Bloom Energy and CoreWeave, in which Bloom’s fuel cell systems are being installed to provide on‑site power for a high‑performance data centre in Volo, Illinois, with commissioning planned for the third quarter of 2025. Those commercial ties help explain why a fund focused on infrastructure would hold both energy and data‑centre names.

Situational Awareness’s portfolio also shows elevated exposure to companies bridging crypto mining and broader compute hosting , including IREN, Cipher Mining, Riot Platforms, Bitdeer and Applied Digital , a pattern reported by Cointelegraph. Market participants and analysts have increasingly treated large‑scale miners and their grid‑connected sites as latent AI hosting capacity: assets whose value depends less on short‑term hash price dynamics and more on long‑term contracted compute and power arrangements. Core Scientific’s past expansion of its CoreWeave deal, disclosed in earlier reporting, exemplifies how such arrangements can translate into multi‑year revenue streams tied to infrastructure capacity rather than pure hashing economics.

The fund has also taken an explicit short position in Infosys, reflecting a view that advances in large‑language models and AI coding tools will compress demand for traditional outsourcing services. That trade highlights how Situational Awareness is pairing long bets on the physical enablers of AI with selective shorts on firms it perceives as vulnerable to automation and platform substitution.

Regulatory filings tracked by 13f.info and other database services show the fund’s path of rapid scaling: earlier 13F disclosures recorded materially smaller assets under management and a different position mix, suggesting a deliberate and concentrated repositioning into the compute‑and‑power complex. The degree of concentration , heavy weights in a handful of infrastructure and energy names , amplifies both upside if the thesis plays out and downside if long‑dated contracts or grid‑capacity constraints fail to materialise as expected.

Looking ahead, continued disclosure cycles and any further Schedule 13D amendments will be key indicators of whether the $5.5 billion posture is maintained or adjusted. Market watchers will also be monitoring announcements of long‑term HPC contracts, additional data‑centre builds, and regulatory or grid‑policy developments that could alter the economics of running megawatt‑scale compute facilities. If miners increasingly monetise capacity through hosting and enterprise contracts, the metrics used to value those businesses will shift away from hash rate and toward contracted megawatts, uptime and power cost differentials.

The repositioning by a prominent former AI researcher into the capital‑intensive layers that enable model training and inference signals a broader reappraisal among investors: that the next phase of AI growth may be decided as much by who controls reliable, scalable power and space as by chip makers and model developers. Whether that bet is rewarded will depend on a mix of commercial contract execution, grid resilience and the evolving interplay between crypto‑native infrastructure and mainstream AI demand.