Fireside Chat, Interview
Jeff Seibert: Why OpenAI Will Become an Infrastructure Play | E1085
- OpenAI is expected to evolve into an infrastructure provider similar to AWS, offering base capabilities and fine-tuning services while avoiding direct competition in vertical SaaS categories like Notion or Salesforce, even as LLMs become commoditized and startups building superficial wrappers face elimination.
- Apple is predicted to pioneer on-device AI using custom silicon to run sufficiently large LLMs directly on iPhones by leveraging performance levels described as "outlandish," potentially displacing OpenAI for users who can run models locally.
- The industry will see a split between proprietary models and a "very close second" open source equivalent that is "just as good" for many use cases, alongside a "second tier" player emerging as an "Android to OpenAI" for specialized, fine-tuned models.
- Adoption of AI in enterprises is forecast to follow cloud infrastructure trajectories, with companies eventually accepting third-party data centers for training as trust issues are resolved through clear guidelines, leading to AI implementation services becoming a major category within the next few years.
- Technological evolution is expected to be "radically faster" than prior tech waves due to the lack of new hardware requirements or UX changes, with a common popular open source base LLM and fine-tuning tools likely to emerge soon, followed by industry-specific specialization.
- Compute costs and query speeds are projected to drop significantly and become much cheaper over the next three years, while pricing models will likely remain consistent with existing standards such as per-seat or consumption-based fees.
- High-quality clean data will remain difficult to acquire and increasingly restricted through stricter rate limits and API controls, necessitating a shift in R&D focus toward maintaining performance in specific use cases while reducing model size.
- The search engine market faces an existential threat where Google must cannibalize its core search business to integrate AI, likely by combining ML teams and rapidly releasing products such as Gemini in Q1 to avoid obsolescence.
- The investment landscape for early-stage companies is expected to be harsh, with the vast majority of VC dollars resulting in total loss, secondary market sales becoming the most successful outcome for angels, and many founders needing to recap, trim expenses, and lengthen runway to address valuation discrepancies.
- Consumer app traction is viewed as unsustainable compared to enterprise software, while startup success will depend on decisive decision-making focused on customers rather than competitors, with most decisions being reversible "Type Two" moves requiring answers within 24 hours.
- Productivity gains from AI are anticipated to allow people to accomplish significantly more in less time, driven by the "lump of labor fallacy," with useful AI applications expected to materialize in less than a year rather than the longer timelines previously believed.
- Real workflow automation for massive industries is expected to be disrupted by new upstarts rather than incumbents like Intuit, while specific ventures like Digits are projected to become the de facto accounting platform by 2033 using a real-time, object-oriented, AI-driven approach to finance.
- Macro-level risks include the likelihood of runaway climate change occurring in less than 10 years, with society likely being too late to accept this new reality, contrasting with the rapid adoption of AI technologies.