Enterprise AI is entering a new phase — one where the central question is no longer what can be built, but how to make the most of our AI investment. Inside large organizations, Misryoum newsroom reported, the operational reality is messy: AI sprawl, rising inference costs, and limited visibility into what those investments are actually returning.
It’s that “Day 2” moment when pilots give way to production — and suddenly cost, governance, and sustainability feel like bigger problems than the original engineering work. And yeah, in the room you can almost feel everyone doing mental math. GPUs don’t care about your optimism.
Misryoum newsroom reported that Brian Gracely, director of portfolio strategy at Red Hat, framed the issue with a scenario many enterprises will recognize. In his account, customers say, “I have 50,000 licenses of Copilot. I don’t really know what people are getting out of that. But I do know that I’m paying for the most expensive computing in the world, because it’s GPUs,” then ask, “How am I going to get that under control?” In other words: the bill is real, but the receipts—usable ones—are not.
For a while, cost wasn’t the primary concern for organizations evaluating generative AI. The experimental phase gave teams cover to spend, and the promise of productivity gains was enough to justify aggressive investment. But Misryoum editorial team noted the dynamic is shifting as enterprises move through second and third budget cycles with AI. The question is sliding from “can we build something?” to “are we getting what we paid for?” And it’s not just about expense—it’s about instrumentation. Without it, connecting spending to outcomes becomes guesswork, which makes renewals and responsible scaling brutally hard.
Then there’s procurement. The dominant model has been pretty straightforward: pay per token, per seat, or per API call, and let someone else manage the infrastructure. Misryoum analysis indicates that model is increasingly being questioned by organizations with enough experience to compare alternatives. Gracely’s framing was blunt: “Instead of being purely a token consumer, how can I start being a token generator?” Misryoum newsroom reported he suggested some enterprises may want use cases where they generate value by owning more of the workload—operating GPUs themselves, or renting GPUs—and then asking whether the workload truly needs the greatest state-of-the-art model.
The tricky part is that the alternatives are multiplying. The “math” is getting more complicated as capable open models grow in number—he name-checked DeepSeek—and as those models become available through cloud marketplaces. Misryoum editorial desk noted this isn’t a binary decision; it depends on workload, organizational needs, and risk tolerance. Still, the budgeting headache keeps returning to one paradox: even if inference gets cheaper, usage can rise faster. Some enterprise leaders point to the claim that inference costs are declining roughly 60% per year, attributed to a statement from Anthropic CEO Dario Amodei. But Misryoum newsroom reported that falling unit costs don’t necessarily mean declining total spend, because broader adoption can overwhelm the efficiency gains—Jevons Paradox in practice.
So what’s the move? The prescription Misryoum editorial team stated isn’t to slow down AI investment. It’s to build with flexibility. Gracely argued the organizations that win aren’t always the ones moving fastest or spending most; they’re building abstractions and operating models that can absorb whatever shows up next—without turning every change into a budget crisis. And actually, that’s the part that hits home when you’ve been doing this “for three years,” he said, but it still feels like the timeline is measured in years, not decades. Late in the session, someone in the back shifted in their chair and the air conditioner made this steady hum—like the room itself was waiting for the next pivot. “It’s early and it’s moving really fast,” Gracely added, and you could tell he meant it.
For enterprise leaders still calibrating their AI strategies, the clearest takeaway from Misryoum reporting may be simple but not easy: don’t optimize only for today’s cost structure. Build organizational and technical flexibility so you can adapt when—yes, when—things change again. Which is good. Because nobody really knows exactly what’s next, even if everyone’s pretending they do.
I traded my Sonos Era 300 for Denon’s Home 400—and didn’t miss a beat
Sequoia raises about $7B as it expands its AI push
Netflix leans into vertical video with a redesigned mobile app