The top AI tech bros operate as if today’s massive head start guarantees permanent market control. They command enviable capital, compute, talent, and user bases. But confusing today’s strongest position with ownership of tomorrow’s battlefield is a classic strategic blunder.
As Sun Tzu observed in The Art of War, terrain matters. An advantage depends on the conditions around it. Today’s high ground can become tomorrow’s vulnerability.
Dominant incumbents have committed hundreds of billions to centralized data centers, massive GPU clusters, and token-by-token generative LLMs. Those investments carry assumptions about what customers will keep needing, including centralized computing, continuous network access, and enormous all-purpose models.
Meanwhile, cracks/openings are appearing:
- Specialized inference chips can make more AI tasks practical on local devices, reducing latency and dependence on cloud computing.
- Systems built for narrow tasks can use structured decision-making instead of generating every answer token by token.
- Local AI offers privacy, control, and offline use that cloud services can’t always provide.
Sun Tzu’s advice to avoid an opponent’s strength applies here. Building a slightly bigger data center is an expensive way to fight an incumbent. Smaller competitors have a better chance finding opportunities that changes in hardware and software have opened up.
A battleship dominates deep water, but it cannot navigate a shallow stream.
Massive infrastructure investments can make incumbents reluctant to change course when smaller models, alternative architectures, or local hardware prove “good enough” for everyday tasks. They can afford to adapt. Whether they’ll welcome products that undercut the economics of their own investments is another question.
The future of AI is likely to include many kinds of specialized intelligence alongside the giant all-purpose models. The strategic question is what happens to today’s advantages when customers no longer need so much of what the giants have built.
RW
.
.
.