Google's TurboQuant May Have Popped The Bubble. But It's a Good Thing
Speakers
Summary
- Google released TurboQuant, a compression algorithm that reduces LLM key value cache memory by 6x and delivers up to 8x speedup without loss, causing significant sell-offs in memory stocks (e.g., SanDisk -7%, Micron -18% over 5 days, SK Hynix -6%).
- Speakers argue the market overreacted; efficiency gains will likely increase AI consumption via Jevons paradox, making the development bullish for AI overall and specifically for Google, which Ejaaz bought more of.
- OpenAI sunsetted its AI video app Sora after 6 months, despite 5 million downloads in the first day and topping app stores, citing lack of focus, consumer behavior favoring passive consumption, and cash constraints for GPU projects like Stargate.
- OpenAI is pivoting to AGI deployment and world models for robotics, consolidating teams, which Ejaaz views as a bullish refocus after previous scattered efforts.
- SpaceX is rumored to file for a confidential IPO soon, with a potential public listing in June, valuation over $2 trillion, and raising $75 billion; Josh owns private shares and is eager to buy more.
- Apple has halted updates for AI vibe-coded apps on the App Store, stifling a revival in app development, which speakers see as a bottleneck for innovation; this may be strategic amid a deal with Google for Gemini model weights.
- Apple's deal with Google involves paying $1 billion annually for access to Gemini model weights, allowing fine-tuning, which is strategic for Apple's AI capabilities and Google's distribution.
- Meta is offering a $800 million comp package to its CFO conditional on achieving a $9 trillion market cap in 5 years, while laying off 20% in its metaverse division, leading to skepticism but hope from speakers.
- AI music generation via Google LIA 3 allows high-quality song creation with simple prompts, raising questions about the future of artistry but optimism for increased creativity.
- Speakers contrast Anthropic's focused approach on coding models with OpenAI's lack of focus, noting Anthropic's rapid updates as competitive.
Trade Ideas
Ejaaz stated that Google's TurboQuant algorithm is "very bullish Google stock" and he "bought a bunch more when this came out." TurboQuant enhances AI efficiency, reducing memory needs and costs, which may accelerate AI adoption and solidify Google's leadership in AI research and infrastructure. LONG because Google's innovation drives AI growth and positions it for long-term gains. Delays in scaling the technology or competitive responses from other tech giants.
Ejaaz said he is "bullish OpenAI after all of this" following the pivot to AGI deployment and world models. OpenAI is refocusing on core areas like LLMs, coding AI, and robotics after shutting down distractions like Sora, which could improve execution and competitiveness. LONG because strategic focus may lead to breakthrough products and market leadership in AGI. Continued lack of focus, cash burn, or strong competition from focused rivals like Anthropic.
Josh said he "currently own[s] SpaceX, the only private company I own" and is "ready to buy more" in the upcoming IPO. SpaceX is rumored to file for a massive IPO soon, with a potential $2+ trillion valuation and ambitious projects like AI chip factories in space, indicating high growth potential. LONG due to SpaceX's dominant position in space technology and expected market expansion. IPO delays, regulatory hurdles, or execution challenges in new projects.
