Every tech revolution triggers the same panic.
Railroads? Monopoly doom. Electricity? Corporate stranglehold. Internet? Platform death trap.
Some fears came true. Most missed the plot.
AI is now hitting its concentration phase. OpenAI, Google, Microsoft, Anthropic—hundreds of billions flowing into models, chips, data centres, talent. At the foundation level, the game is viciously capital-intensive. Most startups? Already priced out.
So the question lands hard: Is the future just five AI tools—and a graveyard?
History says: not quite
Zoom out—something Ray Dalio won’t shut up about for good reason.
Every major innovation wave follows the same arc: power centralizes, then explodes outward.
- Cloud infrastructure? AWS, Azure, GCP won, then SaaS went supernova on top.
- Smartphones? Apple and Google own the OS, but millions of apps eat the world anyway.
- Electricity? Grids consolidated but entire industries spawned to use that juice.
The fatal mistake: confusing owning the pipes with owning everything that flows through them.
Big Tech is crushing the infrastructure war. That doesn’t hand them the application war.
What’s actually going down
AI funding and compute are concentrating hard at the stack’s base: models, chips, data, centers. Enterprise LLM spending already hits billions, funneling to maybe five players.
Looks apocalyptic for everyone else.
It’s not.
This is cloud computing circa 2008: brutal consolidation below, wild experimentation above.
Where smaller AI tools still win, and keep winning
Real tech value never came from raw infrastructure. It comes from context.
Smaller AI companies dominate when they:
- Own a specific workflow (law, healthcare, education, design, logistics)
- Embed into how people actually work
- Mix AI with proprietary data, trust, and distribution channels big platforms can’t touch
- Ship UX that giants won’t prioritize
Foundation models don’t understand your hospital’s protocols, your law firm’s risk paranoia, your weird creative workflow. That knowledge lives in applications, not base models.
Hence the explosion of copilots, vertical tools, AI-native products—even as model providers shrink to a handful.
Regulation: annoying, but prevents total domination
Policy matters more than founders want to admit.
EU’s AI Act? Compliance hell that slows everything—but also blocks abusive lock-in. US regulation? Looser, favoring scale, but keeps scrappy startups alive through sheer chaos.
Competition watchdogs are already circling API access, app stores, data policies. Translation: governments expect concentration and are trying (badly) to stop it from becoming total.
Dalio principle: imbalances correct. When power concentrates too much, counter-forces emerge. Always.
The uncomfortable truth about what’s coming
The future isn’t “millions of equal AI companies.” It also isn’t “five tools, game over.”
The actual endpoint:
- Small number of global AI infrastructure providers
- Massive, chaotic ecosystem of dependent tools riding on top
- High platform risk + constant switching, bundling, disruption
This is uncomfortable because it’s fragile. Pricing shifts, policy changes, model updates can nuke entire categories overnight. Many startups die fast. Some scale explosively. Few become durable.
That volatility isn’t a flaw—it’s late-cycle innovation working as designed.
What this means if you’re building or using AI
Users: AI won’t feel monopolized. You’ll use a few big-tech copilots—and dozens of niche tools baked into your daily life.
Builders: The lesson is brutal but clear. Trying to beat Big Tech at training massive models? Wrong game, wrong decade. Using those models to solve real, painful, specific problems better than anyone? That’s where the money lives.
The edge
Big Tech isn’t killing smaller AI companies.
It’s killing the fantasy that everyone can compete at the same layer.
The stack’s hardening at the bottom, cracking wide open at the top. That’s not innovation’s end—it’s the start of a nastier, more Darwinian phase.
And if history—and Dalio’s cycles—teach anything:
The moment power looks most locked down is usually when new forms of value start forming quietly at the edges, ready to crack everything open again.
