https://www.youtube.com/watch?v=VJsHN7xLeek
In this solo kickoff episode, Glenn Parham explains why the U.S. Department of Defense has plenty of GPUs on paper yet remains functionally "GPU‑poor"—and what senior leaders could do in the next 90 days to reverse the problem.
Kill “GovCloud” for IL‑5 workloads
Authorize the commercial versions of AWS, Azure, and Google Cloud for non‑classified work instead of forcing developers into separate, GPU‑starved “GovCloud” regions.
Wire existing HPC GPUs into everyday dev environments
Federate the DoD’s own high‑performance‑computing clusters with the cloud so that startups and program offices can tap idle government‑owned GPUs on demand.
Use pre‑commit contracts to seed classified clouds (IL‑6+)
Pre‑buy ~$30 million in GPU capacity from each major cloud provider, locking those GPUs in place for two years so developers at higher security levels can plan with confidence.
[00:00] Glenn Parham
Pentagon is GPU‑poor—and it’s a crisis. We do not have enough GPUs in the right places. If we don’t course‑correct, we’ll spend billions and find ourselves in the same spot two or three years from now. Here’s how to fix it in 90 days. Welcome to Model Citizen.
We spend a lot of time debating the downstream impacts of AI in the military—and that matters. I’m literally building a company around making advanced models work for government. When I led generative‑AI efforts inside the Pentagon, my job was to figure out how to make that happen.
Everyone is betting that AI—and eventually AGI—will transform government work. But before any of that, you need GPUs. Sure, there are other accelerators, but GPUs are the industry’s backbone. That’s why NVIDIA just crossed a $4 trillion valuation.
[01:37] If you hang around D.C. AI conferences, vendors plaster “AI‑powered national security” everywhere. Yet when you open their DoD cloud environments and try to build something, the smoke clears: either zero GPUs or just a handful. These are trillion‑dollar companies that spend more on marketing than on GPUs for their gov regions.
For a while I blamed the vendors. Why not just buy more GPUs? But the finger points back at the Pentagon. Money alone won’t fix this. We could drop another billion and still be GPU‑poor because the hardware wouldn’t live where developers can reach it.
Technical policy has to be surgical: not overly prescriptive, not hand‑wavy. And it needs to be written by people who have actually hacked on the problem. The ultimate solution isn’t massively expensive, but it does require political capital and clear mission ownership.
So, if I were the Secretary or Deputy Secretary of Defense, here’s how I’d solve the GPU crisis within 90 days.