Leopold Blows Up, OpenAI Drastically Cuts Prices, Microsoft’s Best Day

Summary

Reed Albergotti and Alex Kantrowitz start with the blowup around Leopold Aschenbrenner’s Situational Awareness fund, treating it as both a leverage problem and a cultural signal about how tech’s expected-value, EA-influenced worldview collides with market timing. They then connect the episode to the broader AI infrastructure trade: OpenAI cut prices aggressively, while the market continues to reward names tied to chips, memory, cloud, and data centers. Nvidia’s $5 billion strategic investment in Safe Superintelligence is framed as circular on the surface but still rational if it keeps Nvidia embedded in future demand. The second half focuses on earnings and valuation: Microsoft’s huge stock jump, Amazon’s strong AI-era infrastructure position, Google’s cloud and TPU strategy, and Meta’s unclear path through AI. The episode closes with a skeptical take on Apple’s long-term moat if the phone becomes less central and AI assistants shift the lock-in away from hardware.

Notes

• Leopold Aschenbrenner’s Situational Awareness hedge fund reportedly suffered a major drawdown and sold off much of its public equity book after being overlevered on AI-related positions.
• The story is framed as a finance event, but also as a tech-culture one: builder/accelerationist instincts versus effective-altruist expected-value trading.
• Reed says the core failure mode was overleverage, not necessarily a wrong long-term thesis.
• Alex says the fund’s biggest holdings were concentrated AI bottlenecks: Nebius, other neoclouds, SanDisk, Micron, and CoreWeave.
• The short book reportedly leaned into software names such as Adobe, extending the bet that infrastructure would outperform application software.
• Alex says the position was reportedly 4x leveraged, which turned a bad month into a cascading liquidation risk.
• The fund’s own update said it was not shut down or liquidated, and that it would keep operating as a hybrid public-private vehicle.
• Leopold’s interim estimate was stark: month-to-date net performance down 67%, but year-to-date still up 80%.
• Alex uses the episode to highlight duration mismatch: correct thesis, wrong timing, catastrophic outcome.
• Reed says compute bets can be partially self-healing because the hardware can be repurposed and sold into strong demand.
• OpenAI cut prices on GPT-5.6 Terra and Luna shortly after launch, with Terra down 20% and Luna down 80%.
• Reed breaks the AI market into three layers: GPU/chip vendors, model providers, and data-center service providers.
• He argues that simply owning racks is not a great long-term business without value-added services or leverage on the hardware layer.
• Reed says frontier-model businesses can still be good because the market is huge and the models improve through harnesses and repeated usage.
• He rejects pure zero-sum thinking, arguing frontier-model companies may not dominate the economy but can still be durable businesses.
• Reed says growth-mode AI companies can lower prices without becoming commodities; the move is about market share.
• He notes pricing pressure could slow IPOs and complicate the pitch to public investors.
• Nvidia’s $5 billion investment in Safe Superintelligence was cast as a bet on an unshipped company with no product yet.
• Reed says the obvious concern is circular financing, but the deal still makes strategic sense if it creates a future Nvidia customer.
• Reed argues the investment is low risk because even a failed startup may still leave behind valuable GPU infrastructure.
• He identifies the main downside risk to the GPU thesis as an algorithmic breakthrough that dramatically cuts compute needs.
• Reed says a world with much lower compute requirements would help civilization, but hurt Nvidia, OpenAI, Anthropic, and data-center builders.
• Altman’s reported agent plans are described as background multi-agent coordination plus solving harder math problems.
• Reed says that sounds like the same multi-agent pattern already visible in Codex-style workflows.
• Reed says the next wave of AI tools will reduce human check-ins and let agents run longer on their own.
• Reed gives a concrete example: he used AI to build a neighborhood flooding data portal with a database, backend, photos, and videos.
• Alex and Reed agree that one-shotting a full product would be a qualitatively new capability.
• Microsoft’s rally was tied to confidence that AI capex will not overwhelm cash generation.
• Azure grew 43% in the reported quarter, and Microsoft promised it would not go negative free cash flow next year.
• Reed says Microsoft is winning because the market wants an efficiency story, and Microsoft has been telegraphing one for a while.
• Reed says the AI market is still being priced on weekly vibes because fundamentals are too early and too unclear.
• Reed explicitly endorses the data-center buildout as productive capital spending rather than waste.
• Amazon’s data-center and service stack is positioned to capture long-run AI workload growth.
• Google Cloud revenue grew 82%, and the market later erased the earnings-day decline.
• Reed thinks Google’s stock reaction is disconnected from fundamentals and colored by perceptions that it is behind in model quality and coding.
• Reed says Google Cloud is a useful AI-era hosting layer even if Google’s models are not the market favorite.
• Google is also selling TPU chips and trying to shift some data-center capex burden onto customers through partnerships.
• Reed’s comparison is that Google’s market treatment is irrational relative to AWS, because both are infrastructure businesses.
• Meta fell 10% after AI costs rose, and Alex says the company’s strategy remains murky.
• Reed says Meta has not shown a clear path through the AI era and is simply trying to rent out compute like a generic hyperscaler.
• Alex says analysts effectively pushed Zuckerberg to become a hyperscaler, but he resisted dumping compute for short-term profit.
• Alex speculates Meta may be waiting for a safe AI companion product, but Reed thinks that market would be too small to matter.
• Reed says AI companions are a niche business relative to Facebook, Instagram, and WhatsApp.
• Apple’s memory outlook is pressured because AI data centers are pulling memory supply and may push up device costs.
• Reed argues the phone itself is becoming less important as AI assistants become the primary interface.