The void of funding not only limits computational resources but talent as well. Industry leaders have the revenue to pay hefty stable salaries while those funded through public support don't have enough incentive--positions are unstable and tethered to short projects and budgets for those projects are non-starters. As such, most talented scientists, researchers, and developers don't bother supporting public research and just join industry.
It's one thing to build out lovely infrastructure but then you still need people to use it and innovate on it. The current models rely on student labor (undergrad, graduates, and post docs) to pickup the tabs with low compensation relative to skill and passionate researchers to pick up the rest in unpaid or underpaid time from faculty supported positions. This model isn't sustainable where a BS can start people out in the $70-100k range while research will struggle to maintain half of that and often want a post doc for that rate.
Basic research salaries are never going to be competitive with applied research / practice salaries in a free market.
DARPA's funding and management structure is an explicit admission of that.
The base government problem is always that you want motivated people. You can motivate people by paying them competitive salaries, providing them job security, or providing them access to do work they wouldn't otherwise be able to.
Compensation is out (see above), and job security in government is always subject to weasel-capture (just as it was/is in academia) without a perfect, future predicting oracle.
This seems to (rightly, IMHO) be focused on that last incentive.
Why can't job security (or even compensation) be marginally improved? No one is expecting FAANG-level salaries with lifetime tenure.
On the other hand, it's not hard to see how 1-year contracts at $50k/year chase people out of research. Heck, you can't even stay at this level for many years--a ton of postdoc funding has time-since-PhD limitations. It's either up or out.
Funding some staff scientists at (say) ~75k/year with 4-6 years between renewals might enable more and better science. There's even evidence this improves trainee outcomes too....
They had monopoly markets that printed cash. They had huge science labs because they were hiring to keep talent from forming competitors, and kept them busy with research. The science was more byproduct than ultimate goal
The key difference was that the supply of quality researchers was so small that you and a handful of other monopolies could hire them all and keep them away from your competitors.
"Academia" is such a walled garden it's ironic that they complain about "Imbalance in AI innovation ecosystem between academia, government, and industry".
What about the imbalance between the public who funds research and the institutions who withhold access to it?
It's one thing to build out lovely infrastructure but then you still need people to use it and innovate on it. The current models rely on student labor (undergrad, graduates, and post docs) to pickup the tabs with low compensation relative to skill and passionate researchers to pick up the rest in unpaid or underpaid time from faculty supported positions. This model isn't sustainable where a BS can start people out in the $70-100k range while research will struggle to maintain half of that and often want a post doc for that rate.