Fireside Chat, Interview
What DeepSeek Means For The Future Of AI | Tech Veterans Weigh In
- DeepSeek R1 signals a strategic shift from "scale up" in massive data centers to "scale out," where "gazillions" of small models run on endpoints like phones, enabling a Total Addressable Market (TAM) expansion of 10x to 100x as revenue shifts from the model layer to applications across every device.
- Industry maturity will move beyond parameter counts toward metrics defined by application performance, with models judged on their ability to provide "truth" and source citations, supported by "thousands of routine tests" similar to ImageNet.
- The sector is expected to evolve from the "first breakthrough" of model creation to a focus on "distribution" and "apps," where the "real money" lies in stateful, complex systems and AI-native versions of software like PowerPoint, Canva, and Word that replace legacy formatting commands.
- Enterprise adoption will act as a primary differentiator, driven by the necessity of integrating security features such as single sign-on, Role-Based Access Control (RBAC), and granular user filtering from the outset.
- DeepSeek's rise to 35% of OpenAI's Daily Active Users (DAUs) in a short timeframe illustrates a "zero" calendar time for technology diffusion, suggesting that global connectivity renders export controls on chips and software ineffective against rapid worldwide innovation.
- Despite the "wake-up call" provided by new competitors like DeepSeek, established players including OpenAI, Anthropic, and NVIDIA face no existential crisis, provided they accelerate their pace, as the competitive landscape will resemble the internet's "coopetition" where big players compete with partners and newcomers.
- Future infrastructure investment is predicted to mirror the early internet fiber build-out but with greater stability due to cloud balance sheets, avoiding a "glut crash," while the U.S. must prioritize rapid domestic research investment over restrictive containment policies to "win" the competition.
- Innovation is expected to remain decentralized and unpredictable, emerging from "pockets of people" such as algorithmic hedge funds and universities, with most significant advancements likely remaining undocumented in history for five years.
- The Chinese internet's advantage in structured data and its "great researchers" are projected to allow China to execute well in AI for the long term, despite U.S. export controls which are described as "wrong-headed" and historically futile compared to the isolation of encryption or gaming consoles.
- The market is anticipated to follow historical internet patterns where apps matter more than raw models, leading to the emergence of entirely new categories for creativity and coding, while developers create diverse pricing models for specific scenarios.