Tuesday, 26 May 2026

The AI industry is consolidating faster than most people realise, and I think we should be paying attention to why.

Four major labs—Anthropic, Mistral, Google DeepMind, and Meta—each made an acquisition within five days this week. This is not a funding story. It is a signal. When you watch the best-resourced teams in AI simultaneously decide they need to absorb specialised talent and capability rather than build it in-house, that tells you something about where the real bottleneck lies. It suggests the race is no longer about who can publish the largest model, but who can solve the hardest engineering problems fastest. Consolidation at this speed usually means the industry has hit a wall it did not expect.

By the way, Skygen.AI's announcement this week illustrates exactly what those walls look like. Their persistent execution platform completes multi-day workflows in two days where conventional agents take seven. That is not a 3.5x speedup in isolation—it is a rebuke of how we have been thinking about agent architecture. The old approach, polling and restarting and hallucinating its way through long chains of logic, fundamentally does not work at scale. The new approach, stateful and persistent, does. That distinction will matter enormously to anyone building real products on top of these systems.

Separately, something remarkable happened at the Vatican on Monday. The Pope released a 42,000-word encyclical on AI ethics alongside Anthropic co-founder Chris Olah, who explicitly called for external oversight of AI development. I find this significant not because the Vatican is now an AI safety authority—it is not—but because we are seeing the governance conversation move out of conference panels and into spaces where actual institutional power lives. When the Pope is talking about robust regulation and the common good rather than profit, and a leading AI lab founder is standing beside him endorsing external oversight, the industry's window for self-regulation is visibly closing. This is not coming in ten years. It is coming now, and it will reshape how these companies operate.

The robotics side is accelerating in parallel. Hyundai is building out dedicated AI factories and robotics supply chains. Tesla's production lines are switching to humanoid robots. XPeng's robotaxi just went into mass production. These are not pilot projects anymore—these are manufacturing decisions. Meanwhile, Google released Gemini Omni, a multimodal model designed specifically for video creation and editing, which suggests the next frontier is not conversation or code generation but the ability to reason about and manipulate visual media at production scale.

What emerges from this week is a pattern: the industry is moving from capability-building to systems integration, from research to manufacturing, from internal governance to external accountability. The easy part—making smart models—is largely solved. What remains is far harder: building systems that actually work in the world, at scale, under oversight that is both meaningful and not paralyzing. That is where the real competition is now.