Summarize this content to 2000 words in 6 paragraphs
Ben Hallen, the Dempsey endowed professor in strategy and entrepreneurship at the Foster School of Business at the University of Washington. (UW Photo)
Editor’s note: This is a guest post written by Ben Hallen, a professor of strategy and entrepreneurship at the Foster School of Business at the University of Washington.
We woke up this morning to AI-related stocks taking a hit as investors became concerned about the implications of DeepSeek, a China-based developer of a new large language model that has been the buzz of AI geeks the last few weeks.
Not only does DeepSeek’s R1 match the performance of cutting-edge models from OpenAI, Anthropic, Google and Meta, but here’s the kicker — it appears to have been developed at a fraction of the computing costs. To top it off, the company has started pulling back the curtain on how they achieved this feat, sending shockwaves through the industry.
To understand what this means for startups and entrepreneurs, we need to zoom out and adopt what strategists call an “ecosystem perspective.” Traditional strategy discussions often center on a single question: which company will dominate a market? But the ecosystem perspective, rooted in game theory, shifts the focus. It emphasizes how value — measured by the total price customers are willing to pay — is split across the stack of players delivering that technology or service.
Until now, soaring AI valuations have reflected the belief that much of the value in AI will be captured by the creators of general-purpose models. The assumption has been that the AI industry will follow “winner-take-all” dynamics, akin to the internet era, where juggernauts like Google and Meta dominated search and social media, capturing much of the value from the advertisements they delivered.
But DeepSeek’s innovation calls for a rethinking of the future of the AI ecosystem. While general-purpose models will remain pivotal — and likely become even more widespread — the barriers to creating them may be far lower than anyone expected. As competition among these models heats up and open-source models become more competitive, margins are poised to shrink, taking valuations for developers of foundational AI models down with them.
So, who stands to win in this new paradigm?
DeepSeek’s breakthrough suggests a shift reminiscent of the commercialization of electricity in the late 19th and early 20th centuries. Some key fraction is likely to be captured by the developers of domain-specific applications — in essence, the next generation of SaaS companies.
SaaS companies’ competitive advantages traditionally lie in a combination of better understanding customer needs, developing more and more features, and customer switching costs. Unique training data may also play a role here, though the advances of DeepSeek around how to train models indicate this may be less of an advantage than traditionally suspected.
Finally, it also seems likely that much of the value will be captured by end users — both consumers and companies that begin to use generative AI, further expanding AI’s broader economic and social impact.
