Saturday, 23 May 2026

The U.S. intelligence establishment is finally waking up to a problem that's been obvious to anyone paying attention: they can't run modern AI on classified networks because they don't have the chips. Nine billion dollars now flows toward fixing that gap, which tells you something important about where real power in AI deployment actually sits. It's not in the labs. It's in the foundries and the semiconductor supply chains, and the government is spending tens of thousands per agency employee just to catch up to what industry has already built.

What strikes me about this is the lag time. Intelligence agencies work slowly by necessity, but AI moves at venture-capital speed. By the time a $9 billion appropriations bill clears Congress and the money reaches classified systems, the models and the hardware they run on will have shifted again. The CIA and NSA aren't buying to lead—they're buying to keep from falling too far behind. That's a different problem than most people realize.

Meanwhile, Walmart is doing something more interesting from a business perspective. Its Sparky AI agent is moving the needle on actual revenue: higher average order values, better unit sales, smarter fulfillment decisions. This is the AI story that matters right now—not the models themselves, but what companies do with them once they're in production. Walmart treats Sparky as infrastructure, not a novelty, and that distinction explains why some organizations are pulling real value from AI while others are still experimenting.

By the way, there's a political subplot here worth noting. The same tech figures who successfully lobbied Trump to scrap an AI safety executive order at the last minute are also the ones who'll benefit most from lighter regulatory touch. David Sacks, Elon Musk, and Mark Zuckerberg got what they wanted. But I find it curious that substantial portions of Trump's own political coalition—the MAGA contingent—actually support AI regulation. The war, as one headline put it, may be far from over. A tactical win in a phone call doesn't settle the longer argument about how AI gets governed.

On the technical side, Google's partnership with FANUC to scale physical AI in industrial robotics is more consequential than it might sound at first. A thousand robots shipped and deployed isn't a headline number until you realize what it means: embodied AI is leaving the lab. Google and FANUC aren't building one perfect robot. They're building a platform that learns, scales, and improves. That's how real automation actually transforms manufacturing.

GitHub, by contrast, is learning that user growth doesn't equal competitive durability. Microsoft's coding platform has stumbled on outages while newer rivals picked up momentum. It's a useful reminder that in fast-moving spaces like AI tooling, yesterday's market leader has to keep earning its position every single day.