A hedge fund managed by Leopold Aschenbrenner, former OpenAI researcher and author of the widely-circulated "Situational Awareness" memo on AGI timelines, disclosed put positions valued above $1 billion against Oracle in a Q2 2026 13F filing. The fund also holds short positions in Nvidia and Intel, according to the regulatory disclosure.

Aschenbrenner left OpenAI in 2023 after internal disagreements over safety protocols and compute allocation. His fund, which operates without a public brand, manages capital estimated in the low single-digit billions. The Oracle puts represent the largest single position in the filing by notional value. The Nvidia shorts are structured as a combination of direct equity shorts and put options with strikes clustered between $95 and $110, suggesting a thesis tied to near-term earnings compression rather than a long-term technology disavowal. Intel positions are smaller and appear tactical.

The timing matters because Oracle's cloud infrastructure revenue growth decelerated to 21% year-over-year in its March quarter, down from 49% the prior year, while its AI-linkedCapEx commitments surged past $8 billion annually. Aschenbrenner's thesis, inferred from the strike prices and position sizing, appears to center on a wedge between Oracle's infrastructure buildout costs and decelerating customer adoption of its GPU cloud offerings. Oracle has committed to expanding its GPU cluster capacity to compete with AWS and Azure, but its customer base skews enterprise SaaS rather than frontier model labs. If model training runs plateau or shift toward inference-optimized architectures, Oracle's GPU bet becomes a stranded asset story. The Nvidia short likely hedges general sector exposure but also reflects skepticism that data center GPU demand can sustain $120 billion annual run-rates once the current wave of H100 and H200 deployments completes.

Allocators should note that Aschenbrenner's edge is not proprietary compute access or venture deal flow. It is a ground-level view of which AI workloads actually require cutting-edge infrastructure and which can migrate to cheaper, older silicon or inference-specialized chips. If his thesis holds, the next twelve months will show Oracle's cloud margins compressing as it finances underutilized capacity, and Nvidia's data center revenue growth decelerating below the 40% consensus estimate for fiscal 2027. Both companies report quarterly earnings in late May and late August respectively. Watch Oracle's remaining performance obligations and deferred revenue line items. A sequential decline there, even with revenue growth, signals customer hesitation. For Nvidia, watch the mix shift between H100/H200 and the new Blackwell architecture. If Blackwell adoption is slower than guided, it confirms the inference-optimization trend that makes Aschenbrenner's bet logical.

The 13F does not disclose the fund's long positions in private AI companies, but Aschenbrenner has publicly argued that the real alpha in AI investing is in application-layer companies that will benefit from collapsing inference costs, not the infrastructure providers. If he is correct, the next cycle belongs to software margin expansion, not hardware capital intensity.