Interview, Fireside Chat
Should American Enterprises Work With Open-Source Chinese Models? | Only 10% of Neo-labs survive
- Smartest models are predicted to become the cheapest within two years, with growth rates potentially exceeding current estimates by an order of magnitude to reach valuations of $200 to $300 billion driven by data requirements.
- Within three years, 99% of workflows are expected to run on open models, while the remaining 1% of tasks—requiring frontier capabilities—will likely generate 30% to 40% of total economic value, with the performance gap between these segments widening.
- Model speciation will accelerate rapidly, creating numerous specialized internal models that businesses will keep private, accessible via software services in a few clicks without requiring in-house expertise.
- The industry faces high failure risks in the next 18 months, with 80% to 90% of new model startups expected to "die" or be acquired, whereas legal-focused startups may remain durable for 5 to 20 years due to persistent system requirements.
- Margin profiles for model labs are shrinking, posing risks if companies cannot match price increases with value, potentially forcing providers to enter the chip layer to manage massive debt as they scale to multi-trillion dollar valuations.
- Business decision-making is viewed as fundamentally reactive rather than planned, with future value concentrated in real-time responses to split-second information rather than static forecasts.
- Software interfaces are predicted to become fully custom and generated on demand within three to five years, eliminating the need for a specialized "priestly class" of two million software decision-makers.
- Microsoft is identified as a top-positioned hyperscaler due to infrastructure ownership and developer independence, while general knowledge work in tools like Excel and Jira is unlikely to sustain independent business value over five to ten years.
- Chinese open source models are assessed as having no unique security risks compared to American counterparts, differing primarily in bias, though specific national security documents should avoid them.
- Massive M&A waves for SaaS companies are anticipated, driven by the shift from static monopolies to markets requiring continuous scale, with all future hires at Factory expected to come via acquisitions of founding teams.
- Long-term growth will be driven by self-service adoption of cost-effective, locally runnable models and sectors like media clipping that capitalize on proprietary knowledge deltas.
- Economic expansion is expected to accelerate faster and become larger over time, though high infrastructure debt levels pose existential threats to providers lacking significant free cash flow.