Saturday, 20 June 2026

The funding numbers keep getting bigger, and that's starting to tell us something important about where the real money believes the next wave of AI value actually lives.

DeepSeek's $7.4 billion Series A is striking not because of the size—though that's substantial—but because it signals how seriously Chinese capital is moving to build foundational models outside the Western-dominated axis. The round itself is less interesting than what it means: the race to train competitive large language models has become a multi-front competition, and venture capital is betting accordingly. This isn't about one company; it's about the structural shift in how AI development gets funded globally.

But here's what I find more telling: the really significant moves this week aren't all about raw model capability anymore. Microsoft pushing agentic ERP workflows into Copilot Cowork tells you something about where business value is actually concentrating. We're moving past the phase where a better language model automatically wins. Now it's about whether an AI system can reliably integrate with the messy, complicated infrastructure that companies actually run on. Procurement decisions via natural language sounds simple until you realize it means the agent needs to understand your specific vendor contracts, your approval workflows, your budget constraints—all the particular texture of how your organization actually operates. That's harder than generating a better image or writing a better poem, and it's where real money moves.

By the way, Cloudflare's decision to offer temporary accounts for AI agents is a small but revealing piece of this puzzle. The problem it solves sounds mundane—agents need access to deploy things, humans have access controls—but it's the kind of friction that actually slows adoption in production environments. The fact that infrastructure providers are now building specifically around agent workflows tells you this isn't hypothetical anymore; companies are actually shipping agents into their systems.

The robotics angle is worth watching too. Samsung potentially investing in Boston Dynamics fits a pattern: the major tech and manufacturing conglomerates are positioning themselves for the embodied AI phase. This isn't about one robot company succeeding; it's about the hardware-software boundary collapsing as AI agents move from purely digital environments into physical ones.

I'm more curious about what's not getting funding than what is. The sheer capital flowing into agents and workflow automation, the international capital chasing model training, the corporate infrastructure plays—these suggest the market has a clear thesis about what matters next. Whether that thesis is right is a different question entirely, and that's the one I'm watching carefully.