The US government did not build bombers, ICBMs, or even the Apollo vehicles. All of those were specified by agencies and contracted out to the same contractors that exist today. I think the last thing the US "manufactured" was ammunition during the Civil War, as it owned a few shot towers, but even the vast majority of those were privately-owned and contracted by the US Army.
(Though I was thinking more of the research/planning vs actually mass production, that you're taking about. Especially as this discussion is about such as case too, considering it's software)
It's not that outrageous. There are government-run liquor stores all over the country, and even government-run grocery stores, most notably at every military base. The image here of one seems like a totally normal grocery store: https://en.wikipedia.org/wiki/File:US_Navy_020813-N-3642E-50...
Are Amazon and Meta the ones losing out the most here, in terms of the companies building foundational models?
Probably more understandable for Meta, since they've been leaving the B2B space since Workplace has been sunset. Amazon losing out on this is pretty rough for AWS though.
Is Amazon trying to build a competitive foundation model? From what I can see AWS is instead focused on hosting and re-licensing Claude, Cohere, DeepSeek and others via Bedrock. And it's pretty likely that a large chunk of this $200M will anyways go to AWS. So I'd hardly call them a loser here.
Amazon has a number of foundation models under the name Amazon Nova, which they claimed were SOTA on release but I haven't heard much at all about them since.
Most of US government runs significant workloads on AWS now and that’s only increasing. They’ve cornered govt cloud infrastructure (with Azure, GCP, etc. very far behind) so not sure this matters in grand scheme of things.
Anecdotal based on industry experience, no citations.
Maybe it's less about the money and more about signalling to foreign adversaries. "We're prepared to weaponize AI, so you should tread lightly." Everyone knows that in case of war those millions could turn into billions overnight. It's like a cowboy flashing his gun. It says he will use it.
I know this is likely in the pipeline anyway and maybey not covered by this news but now we have the prospect of agentic llms hallucinating enemies and a digital finger on the trigger.
LLMs are only useful information systems, largely for parsing/managling variable data and building other information systems. Problem sets any large org like DoD has.
I don’t think anyone has even seriously proposed using them for weapons targeting, at least in the current broad LLM form.
If they are slow (2x as slow on a cruise missile or drone SOC) and are wrong all the time then why would they even bother? They already have AI models for visual targeting that are highly specialized for the specific job and even that’s almost entirely limited to very narrow vehicle or ship identification which is always combined with existing ballistic, radar, or GPS targeting.
Buying some LLM credits doesn’t help much at all there.
Too much of AI gets uncritically packaged with these hand wavy FUD statements IMO.
I'd like to believe you, but there's credible evidence that (e.g) DOGE has been using LLMs to cut funding for NSF or HHS using prompts in the vein of "is this grant woke."
Which is obviously stupid. So if stupid people are using these things in stupid ways, that seems bad.
Given that that's a task you want to do, it's at least the right kind of task (language processing) for an LLM. The proposals from the comment starting this thread aren't.
If grant classification is trying to drive a car non-stop (including not stopping for gas) from NY to LA, stuffing LLMs into weapons is more like trying to drive that same car from NY to London. They're just not the proper kind of tool for that, and it's not the same class of error.
If people on Hacker News are uncertain about what is and isn't a suitable task for these models then the non technical people making these decisions surely are as well.
You're saying that weapons are designed by incompetents, and that enthusiasts have a reasoned understanding of the capabilities and limitations of the latest thing they're going "ooh shiny" about.
That's fundamentally not a language processing task. That's a decision making task with a huge impact on individual scientists and the scientific community. Not something that should be delegated to a machine, no matter how sophisticated.
Why not 20x $10M grants for smaller companies? They're gonna throw this money with no oversight anyways so why not bolster the actual startup scene instead of a bunch of incumbents who all have more than enough cash? $10M could keep a startup running for 1+ years at its most crucial time. That's 10 solutions instead of 1 -- statistically one of them will be a massive breakthrough?
Who are these smaller companies, and what do they have to offer that these 4 don't? Chances are that the smaller companies themselves are licensing the LLM from Google/Anthropic/OpenAI, so why pay middlemen for no reason?
You’re telling me that you can’t find 10 worthwhile AI startups to give money to? I bet there are 1000 on crunchbase right now. With $10M some of them could buy hardware to build their own systems.
Maybe 20 10M contracts is a bit too small, but this point is completely wrong too. Part of the purpose of public expenditure is to promote healthy businesses, not just procure things are efficiently as possible
This is people's money, and people benefit from competition in the market
IMO if the government needs widgets it should ask for bids and contract based on price. Some broad restrictions are OK (e.g., prefer domestic manufacturers), but government has no business skewing the contract by promoting <insert favorite agenda>. If we want to promote X with public money, do so explicitly and separately: support research, fund startups, etc. My 2c
Sorry for confusion. In "buying widgets" I meant the regular procurement decision: the DoD needs N units of something (door handles, tank shells, whatever), it publishes specs, asks for bids and buys from whoever can deliver at the lower cost to the taxpayer. The government also funds research, where it gives money to generate something it believes to be beneficial longer term.
Funding research is, by definition, less cut-and-dry as to what we should pay for; thus having an agenda is not always bad and might even be good. I am using an "agenda" not in a narrow political sense, but including positions like "we should be funding space comms / drone networks / real-time soldier health monitoring / whatever because industry is not building what we think we will need in a few years".
But being somewhat exposed to the waste of DoD procurement I am personally vehemently against inserting such ideology into procurement decisions. Those should be money-based. Get what you need at the least cost to the taxpayer. If you do not know what you need, think harder or invite experts or do a study (and publish it so people writing it know they are associated with those decisions) before paying billions for questionable junk. My 2c.
Eh, national security works differently from promoting healthy businesses.
The govt already has various programmes to help promote small business contractors in US defence. This is not a programme; it's a definitive project that has a specific set of (admittedly vague) objectives in mind. It's more efficient for the taxpayer for these to be accomplished when the funding is consolidated to a few entities for a 50% success rate, than to 20 different entities for a 5% success rate.
I’m all for a competitive commercial space type approach to gov projects and contracts but I really don’t see what that has to do with this.
DoD is experimenting with LLMs and is using multiple of the top providers in the space… just like every other tech company is doing. Everyone I know is coding with Claude, Gemini, or GPT and my experiments with Grok 4 have easily been as good.
If this was an innovation fund, ala what Canada likes to waste money on - where the gov pretends it’s a really bad VC, I’d at least understand these critiques.
Why would the DoD give them money just to break-even (not even likely. oai has easily put > 10m into their compute) with companies that already have the infrastructure? This isn't a stimulus package, it's a service they're buying.
Even then, these companies aren't doing research into LLMs, they're just wrapping the endpoints and creating some abstractions.
DOGE hates resellers and is trying to force all transactions through the prime contractors. Most small companies cannot afford to float the terms required by the federal government without resellers.
This admin is about graft and shakedowns. Just like the implosion of science, the companies that exist due to smallish federal contracts for obscure tech and speculative investments are toast.
> $10M could keep a startup running for 1+ years at its most crucial time. That's 10 solutions instead of 1 -- statistically one of them will be a massive breakthrough?
The failure rate for startups is much higher than 90%. And there’s the additional complexity of how do you pick which 20 such startups get the cash.
See my response to the other posters with the same notes
On the picking: it’s really not hard to search for AI companies and pick 20. In fact there are government programs that invest in startups so clearly it’s doable.
Who are those 20 companies? What would $10M do in the context of training LLMs that are competitive with Claude/O3/Gemini?
> That's 10 solutions instead of 1 -- statistically one of them will be a massive breakthrough?
The statistic is that 10% of startups make a massive breakthrough? Would love to see some work that comes remotely close to replicating that! Startup investing would be trivially easy.
Responded to the other poster with the same question.
Everyone says 1 out of 100 makes it big but the top 5-10% of a portfolio is still substantial. If we’re only giving the money to companies with revenue the odds of success are likely improved.
Startup investing is trivially easy. You give money to good companies and founders. There’s just a bunch of BS that gets in the way. Like giving massive money to big corps that don’t need it instead of startups that do.
Here are some top AI companies per Crunchbase: OpenAI, Anthropic, xAI, CoreWeave, Glean, Perlexity, PlayAI, Cohere, Tempus, Cyera, Replit, Windsurf, Mistral, Anysphere, Scale, Harvey, Thinking Machines Lab, helsing, Cluely, Suno, Clay, Crunchbase (lol), Lubega Geoffery, Caris LIfe Sciences, C3 AI, Runway, LangChain, Rigetti Computing, Cowbell, Laurel, SoundHound, Voxel, Harmonic, Builder, ElevenLabs, Decagon, Spring Health, Lovable.... alright I have a meeting to get to.
OpenAI - DoD invested in them and now I guess you agree with it!
Anthropic - same as above
xAI - same as above
CoreWeave - Doesn't make LLMs
Glean - Doesn't make LLMs (wow this startup investing thing might be harder than for you than you thought!)
Perplexity - Has finetuned LLama models AFAIK. Maybe you think Meta should've gotten the nod from DoD as well?
PlayAI - AFAIK only voices
Cohere - Not sure if they are LLama or otherwise
Cyera - Doesn't make LLMs
Replit - Doesn't make LLMs
Windsurf - Doesn't make LLMs
Mistral - Does make LLMs, you got one! Is French, though.
Anysphere - They make an IDE called Cursor
Scale - Doesn't make LLMs, basically a Meta subsidiary (you really must have wanted Meta to get the nod too!)
Harvey - Legal focus, not general
Thinking Machines - Mira Murati's company, just started 5mos ago, no public products. Definitely don't fit your definition of "has revenue"
helsing - Hadn't heard of them, are German.
Cluely - LOL
Suno - If the DoD gets into music generation this would be a great choice.
Clay - Don't know them, doubt they have LLMs.
Crunchbase - lol is correct
Lubega Geoffery - No idea
Caris LIfe Sciences - Life sciences doesn't sound right!
C3 AI - Scam
Runway - Media generation, not general use
LangChain - Doesn't make LLMs
Rigetti Computing - Dude, come on. They're a quantum computing company
Cowbell - Don't know them, but a google shows they're an insurance company lol
Almost all the rest don't even have anything to do with AI. So all-in-all, nearly a complete failure at suggesting even close to 20 alternatives for the DoD to invest in. Your answer didn't even hit US companies that do have some alternatives: Meta, MSFT, AMZN, SSI maybe?
I understand the sentiment to create healthy market but only a few handful company than can create general use LLM, most of them is just wrapper or small fine tune model for specific use case
I don't for anyone doing serious work. Use the Gemini family, O3 and Claude if you want to gsd. The DoD made the correct call IMO. Kimi K2 is also potentially interesting for non-defense purposes but I haven't spent enough time with it yet.
> My fund will never be $1B but that’s fine :)
It's "trivially easy" for you and 10% of your investments are expected to have "massive breakthroughs". Your strategy of "give money to good companies and founders" should easily enable you to reach $1B AUM!
not that they're great or very remarkable, but they can get the job done, and in a world where anthropic tries to assert dominance by fucking cursor seems relevant to the future to have your own models
As an American. I'd rather a single well established player get a large contract and actually deliver, than 20 disjointed companies each get 1/20th of the problem, have to work in concert, and possibly not even deliver at all.
This isn’t a grant to push for innovation. This is a promise from the orange man administration to the people and companies that donated to his "inauguration fund"
This is a kleptocracy but with extra steps. People are unfortunately numb to it.
It’s up to $200M for each of them. From the actual source:
“The awards to Anthropic, Google, OpenAI, and xAI – each with a $200M ceiling – will enable the Department to leverage the technology and talent of U.S. frontier AI companies to develop agentic AI workflows across a variety of mission areas.”
It’s still not a lot. How much do tokens cost for example?
In theory if it’s just labor with some profit mixed in, then you might be looking at 600 employees for each company.
I doubt it is just labor. Quote says $200 million ceiling. So maybe a time and materials (T&M) contract? It’s a ceiling so it’s not like they earn or are guaranteed $200m.
Has to include token or cloud computing time too. Which Google owns and can amortize themselves since it’s a capital asset to them. I don’t know much about the cloud computing background of Anthropic or if they are using Azure or AWS.
I think my original point is still valid it’s not a lot when you look at it
It will be enough to deliver the requirements for the DoD. They don’t need to spend any more or any less than what is required. Your point seems irrelevant if you don’t know their exact requirements.
Meaningless comparison. If xAI or Google accepts the contract, they intend to deliver within the budget. You can't say whether it's too much or too little without knowing all of the details, which you don't.
"OpenAI Public Sector LLC, San Francisco, California, has been awarded a fixed amount, prototype, other transaction agreement (HQ0883-25-9-0012) with a value of $200,000,000. Under this award, the performer will develop prototype frontier AI capabilities to address critical national security challenges in both warfighting and enterprise domains. The work will be primarily performed in the National Capital Region with an estimated completion date of July 2026. Fiscal 2025 research, development, test and evaluation funds in the amount of $1,999,998 are being obligated at time of award. Office of the Secretary of Defense Chief Digital and Artificial Intelligence Office, Washington D.C., is the contracting activity. "
It's a prototype and its FFP. Only about $2m has been allocated to them. That equates to about 1-2 employees for a year.
Anthropic’s is “is expected to mirror OpenAI’s structure: a similar token obligation at signing and the rest released via milestones.”
All these frontier AI OTAs follow the same pattern of “up to [X] million” ceilings with actual funding phased out as projects progress. This mirrors what Palantir got last year.
Testing the waters. I’d guess if it pays any dividends things will ramp up. This is also across all suppliers. At some point they’ll put more wood behind fewer arrows.
Microsoft said they were beefing up their expenditure in AI with and or around the announcement of layoffs.
So does this mean all the web designers, web developers and many other white collar jobs will now be done by one such professional using AI; so XYZ use to require ten people to get the job done now only one who uses AI gets the tasks done (tasks that those ten use to use software applications to complete). All the while a few hundred of Ruoming Pangs make more money then God in which their work further helps killing white collar jobs.
Is there anyone else concerned about this? I am federal worker indirectly yet per some news today not sure how much longer I will be and whether or not it wise to go look for another web design/developer/UX Researcher (think this is safest out of three as you are talking to ppl)position. There are throngs of others looking now to compete against including now competing against AI for less jobs.
Grok is pretty good, both the model in the API and the consumer products built on top. Llama is way behind, and the meta.ai app is much worse than the leading chat assistant products.
Why would that exclude them from the running? Should government contracts not be granted on the merits of the receivers? Grok clearly exists in this space so it’s not like they’re rewarding vaporware.
> Should government contracts not be granted on the merits of the receivers?
They should, but businesses owned by government employees should be excluded because it's too easy to corrupt the process. In fact, they have explicit rules about not doing that.
But he’s no longer a “special government employee” anymore either. Or are you suggesting he’s blacklisted from all government contracts for life because he previously worked for the current administration?
Not OP, but "For life" is a far cry from a month or two after. But yes, i'd argue we have no choice but to attempt to aggressively put bounds between government and profiteering. Lest we have Congress openly insider trade..
Of course not for life, but generally to avoid ethics issues the government requires you to be out for at least a year or two before engaging in business with the government.
or not spent by the government?
I mean, unless this is for some really specific thing that really needs a push, I think there's sort of enough money flowing this way already, today. (5 years ago, again a different story)
Love the scare quotes around the very things that make life worth living. America at least has the National Park system for metrics like "nature". Oh wait...
This actually makes sense because in the meantime Meta is ditching the open-source (open-weights) direction.
Before the national security narrative took over, the main argument was about "safe" AI, where releasing models as open weights was considered "not safe." Now that no major US AI players release premium open-weights models, the "safety" narrative isn't needed anymore—so cooperating with the US military is feasible again.
Hopefully we'll be lucky and brave patriots will take great risk occasionally leaking those models so we taxpayers can enjoy them too (not just Northrop-Martin).
A cold war joke goes: The US does have planned economy and its name is military contracts.
One of the core responsibilities of the state is to protect itself. Not sure how you don't have this planned apart from anarchy ¯\_(ツ)_/¯
Pretty simple, you don't pay a private enterprise to develop it, instead you develop it yourself after hiring the best people.
At least that's how it was done back in the "golden years", post WW2, during which inequality was quite stable etc
The US government did not build bombers, ICBMs, or even the Apollo vehicles. All of those were specified by agencies and contracted out to the same contractors that exist today. I think the last thing the US "manufactured" was ammunition during the Civil War, as it owned a few shot towers, but even the vast majority of those were privately-owned and contracted by the US Army.
I was misinformed, thanks for correcting me.
(Though I was thinking more of the research/planning vs actually mass production, that you're taking about. Especially as this discussion is about such as case too, considering it's software)
I heard that the US government nationalized defense contractors during WWI. Is that not the case?
The US government didn’t build and plan Lockheed planes. Not in WW2 and not now. Not sure about other manufacturers.
I'd prefer the version where the government states a goal and a market based process is used to determine who can provide best for that goal.
Imagine if the government said no food stamps, we'll just run our own grocery stores to provide for the less fortunate and we'll hire the best people.
You see the difference? Do you really think government run corporations would be able to design better military defense systems?
It's not that outrageous. There are government-run liquor stores all over the country, and even government-run grocery stores, most notably at every military base. The image here of one seems like a totally normal grocery store: https://en.wikipedia.org/wiki/File:US_Navy_020813-N-3642E-50...
I don't think the government should be running liquor stores either
Are Amazon and Meta the ones losing out the most here, in terms of the companies building foundational models?
Probably more understandable for Meta, since they've been leaving the B2B space since Workplace has been sunset. Amazon losing out on this is pretty rough for AWS though.
Is Amazon trying to build a competitive foundation model? From what I can see AWS is instead focused on hosting and re-licensing Claude, Cohere, DeepSeek and others via Bedrock. And it's pretty likely that a large chunk of this $200M will anyways go to AWS. So I'd hardly call them a loser here.
Amazon has a number of foundation models under the name Amazon Nova, which they claimed were SOTA on release but I haven't heard much at all about them since.
They're far from SOTA:
https://help.kagi.com/kagi/ai/llm-benchmark.html
nova pro is worse than llama3-70B
They are not good...
Aka the "sell gold pans during a gold rush" strategy.
AFAIK AWS are pushing pretty hard with GovCloud these days.
I think that would be power components like transformers for the grid.
Those were the people selling lumber to sawmills that eventually ended up as handles for picks.
Most of US government runs significant workloads on AWS now and that’s only increasing. They’ve cornered govt cloud infrastructure (with Azure, GCP, etc. very far behind) so not sure this matters in grand scheme of things.
Anecdotal based on industry experience, no citations.
Meta and Amazon both have separate DoD contracts (Meta with Anduril, Amazon through massive GovCloud contracts)
What is $200M to Amazon and Meta?
Meta can add 1 more member to the technical staff
At the very least it's preventing funds from going to other competitors.
These call order type packages mean that it's probably over 3-5 years, so not really that large a procurement.
Maybe it's less about the money and more about signalling to foreign adversaries. "We're prepared to weaponize AI, so you should tread lightly." Everyone knows that in case of war those millions could turn into billions overnight. It's like a cowboy flashing his gun. It says he will use it.
they're prepared to deploy reams of incoherent code? Gosh
I've worked with VCs that refereed to deals like these as "mouse nuts".
Considering the DoD's budget, $200M is chump change
Also true when considering the size of the companies.
And a drop in the ocean what those companies are losing too on an annual basis.
I know this is likely in the pipeline anyway and maybey not covered by this news but now we have the prospect of agentic llms hallucinating enemies and a digital finger on the trigger.
> the prospect of agentic llms hallucinating enemies and a digital finger on the trigger.
Minority Report takes place in 2054... Phillip K Dick might have been onto something.
LLMs are only useful information systems, largely for parsing/managling variable data and building other information systems. Problem sets any large org like DoD has.
I don’t think anyone has even seriously proposed using them for weapons targeting, at least in the current broad LLM form.
If they are slow (2x as slow on a cruise missile or drone SOC) and are wrong all the time then why would they even bother? They already have AI models for visual targeting that are highly specialized for the specific job and even that’s almost entirely limited to very narrow vehicle or ship identification which is always combined with existing ballistic, radar, or GPS targeting.
Buying some LLM credits doesn’t help much at all there.
Too much of AI gets uncritically packaged with these hand wavy FUD statements IMO.
I'd like to believe you, but there's credible evidence that (e.g) DOGE has been using LLMs to cut funding for NSF or HHS using prompts in the vein of "is this grant woke."
Which is obviously stupid. So if stupid people are using these things in stupid ways, that seems bad.
Given that that's a task you want to do, it's at least the right kind of task (language processing) for an LLM. The proposals from the comment starting this thread aren't.
If grant classification is trying to drive a car non-stop (including not stopping for gas) from NY to LA, stuffing LLMs into weapons is more like trying to drive that same car from NY to London. They're just not the proper kind of tool for that, and it's not the same class of error.
If people on Hacker News are uncertain about what is and isn't a suitable task for these models then the non technical people making these decisions surely are as well.
You're saying that weapons are designed by incompetents, and that enthusiasts have a reasoned understanding of the capabilities and limitations of the latest thing they're going "ooh shiny" about.
I'm not.
That's fundamentally not a language processing task. That's a decision making task with a huge impact on individual scientists and the scientific community. Not something that should be delegated to a machine, no matter how sophisticated.
Why not 20x $10M grants for smaller companies? They're gonna throw this money with no oversight anyways so why not bolster the actual startup scene instead of a bunch of incumbents who all have more than enough cash? $10M could keep a startup running for 1+ years at its most crucial time. That's 10 solutions instead of 1 -- statistically one of them will be a massive breakthrough?
Who are these smaller companies, and what do they have to offer that these 4 don't? Chances are that the smaller companies themselves are licensing the LLM from Google/Anthropic/OpenAI, so why pay middlemen for no reason?
You’re telling me that you can’t find 10 worthwhile AI startups to give money to? I bet there are 1000 on crunchbase right now. With $10M some of them could buy hardware to build their own systems.
This isn't a VC fund. The contract is for an actual service, and the companies best suited to provide them will get it, no matter their size.
Maybe 20 10M contracts is a bit too small, but this point is completely wrong too. Part of the purpose of public expenditure is to promote healthy businesses, not just procure things are efficiently as possible
This is people's money, and people benefit from competition in the market
IMO if the government needs widgets it should ask for bids and contract based on price. Some broad restrictions are OK (e.g., prefer domestic manufacturers), but government has no business skewing the contract by promoting <insert favorite agenda>. If we want to promote X with public money, do so explicitly and separately: support research, fund startups, etc. My 2c
What is the difference between “widgets” and “insert favorite agenda”? It seems like the exact same thing to me
Sorry for confusion. In "buying widgets" I meant the regular procurement decision: the DoD needs N units of something (door handles, tank shells, whatever), it publishes specs, asks for bids and buys from whoever can deliver at the lower cost to the taxpayer. The government also funds research, where it gives money to generate something it believes to be beneficial longer term.
Funding research is, by definition, less cut-and-dry as to what we should pay for; thus having an agenda is not always bad and might even be good. I am using an "agenda" not in a narrow political sense, but including positions like "we should be funding space comms / drone networks / real-time soldier health monitoring / whatever because industry is not building what we think we will need in a few years".
But being somewhat exposed to the waste of DoD procurement I am personally vehemently against inserting such ideology into procurement decisions. Those should be money-based. Get what you need at the least cost to the taxpayer. If you do not know what you need, think harder or invite experts or do a study (and publish it so people writing it know they are associated with those decisions) before paying billions for questionable junk. My 2c.
Eh, national security works differently from promoting healthy businesses.
The govt already has various programmes to help promote small business contractors in US defence. This is not a programme; it's a definitive project that has a specific set of (admittedly vague) objectives in mind. It's more efficient for the taxpayer for these to be accomplished when the funding is consolidated to a few entities for a 50% success rate, than to 20 different entities for a 5% success rate.
Yeah that’s why Boeing keeps getting government money.
I’m all for a competitive commercial space type approach to gov projects and contracts but I really don’t see what that has to do with this.
DoD is experimenting with LLMs and is using multiple of the top providers in the space… just like every other tech company is doing. Everyone I know is coding with Claude, Gemini, or GPT and my experiments with Grok 4 have easily been as good.
If this was an innovation fund, ala what Canada likes to waste money on - where the gov pretends it’s a really bad VC, I’d at least understand these critiques.
Why would the DoD give them money just to break-even (not even likely. oai has easily put > 10m into their compute) with companies that already have the infrastructure? This isn't a stimulus package, it's a service they're buying.
Even then, these companies aren't doing research into LLMs, they're just wrapping the endpoints and creating some abstractions.
DOGE hates resellers and is trying to force all transactions through the prime contractors. Most small companies cannot afford to float the terms required by the federal government without resellers.
This admin is about graft and shakedowns. Just like the implosion of science, the companies that exist due to smallish federal contracts for obscure tech and speculative investments are toast.
> $10M could keep a startup running for 1+ years at its most crucial time. That's 10 solutions instead of 1 -- statistically one of them will be a massive breakthrough?
The failure rate for startups is much higher than 90%. And there’s the additional complexity of how do you pick which 20 such startups get the cash.
See my response to the other posters with the same notes
On the picking: it’s really not hard to search for AI companies and pick 20. In fact there are government programs that invest in startups so clearly it’s doable.
> Why not 20x $10M grants for smaller companies?
That's not how corruption works
That is exactly how corruption works. Friends and families of decision makers make a bunch of small companies that win tenders.
"Corruption is when the US government pays the 4 leading American AI producers for the use of their products"
Who are those 20 companies? What would $10M do in the context of training LLMs that are competitive with Claude/O3/Gemini?
> That's 10 solutions instead of 1 -- statistically one of them will be a massive breakthrough?
The statistic is that 10% of startups make a massive breakthrough? Would love to see some work that comes remotely close to replicating that! Startup investing would be trivially easy.
Responded to the other poster with the same question.
Everyone says 1 out of 100 makes it big but the top 5-10% of a portfolio is still substantial. If we’re only giving the money to companies with revenue the odds of success are likely improved.
Startup investing is trivially easy. You give money to good companies and founders. There’s just a bunch of BS that gets in the way. Like giving massive money to big corps that don’t need it instead of startups that do.
Who are the companies? List some!
Here are some top AI companies per Crunchbase: OpenAI, Anthropic, xAI, CoreWeave, Glean, Perlexity, PlayAI, Cohere, Tempus, Cyera, Replit, Windsurf, Mistral, Anysphere, Scale, Harvey, Thinking Machines Lab, helsing, Cluely, Suno, Clay, Crunchbase (lol), Lubega Geoffery, Caris LIfe Sciences, C3 AI, Runway, LangChain, Rigetti Computing, Cowbell, Laurel, SoundHound, Voxel, Harmonic, Builder, ElevenLabs, Decagon, Spring Health, Lovable.... alright I have a meeting to get to.
OpenAI - DoD invested in them and now I guess you agree with it!
Anthropic - same as above
xAI - same as above
CoreWeave - Doesn't make LLMs
Glean - Doesn't make LLMs (wow this startup investing thing might be harder than for you than you thought!)
Perplexity - Has finetuned LLama models AFAIK. Maybe you think Meta should've gotten the nod from DoD as well?
PlayAI - AFAIK only voices
Cohere - Not sure if they are LLama or otherwise
Cyera - Doesn't make LLMs
Replit - Doesn't make LLMs
Windsurf - Doesn't make LLMs
Mistral - Does make LLMs, you got one! Is French, though.
Anysphere - They make an IDE called Cursor
Scale - Doesn't make LLMs, basically a Meta subsidiary (you really must have wanted Meta to get the nod too!)
Harvey - Legal focus, not general
Thinking Machines - Mira Murati's company, just started 5mos ago, no public products. Definitely don't fit your definition of "has revenue"
helsing - Hadn't heard of them, are German.
Cluely - LOL
Suno - If the DoD gets into music generation this would be a great choice.
Clay - Don't know them, doubt they have LLMs.
Crunchbase - lol is correct
Lubega Geoffery - No idea
Caris LIfe Sciences - Life sciences doesn't sound right!
C3 AI - Scam
Runway - Media generation, not general use
LangChain - Doesn't make LLMs
Rigetti Computing - Dude, come on. They're a quantum computing company
Cowbell - Don't know them, but a google shows they're an insurance company lol
Almost all the rest don't even have anything to do with AI. So all-in-all, nearly a complete failure at suggesting even close to 20 alternatives for the DoD to invest in. Your answer didn't even hit US companies that do have some alternatives: Meta, MSFT, AMZN, SSI maybe?
> helsing - Hadn't heard of them, are German.
Helsing is a military AI company [0] trying to make Terminator I movie a reality in the name of democracy.
[0] https://helsing.ai/
EDIT: added link.
right, these guy is completely bollocks
I understand the sentiment to create healthy market but only a few handful company than can create general use LLM, most of them is just wrapper or small fine tune model for specific use case
I love that you went through every one of those, respect!
It's easy when you know 90% of them already. If I didn't know it, I made that clear in the post.
Looking forward to hearing about your billion dollar VC fund.
Which startup LLMs do you know of / recommend?
My fund will never be $1B but that’s fine :)
> Which startup LLMs do you know of / recommend?
I don't for anyone doing serious work. Use the Gemini family, O3 and Claude if you want to gsd. The DoD made the correct call IMO. Kimi K2 is also potentially interesting for non-defense purposes but I haven't spent enough time with it yet.
> My fund will never be $1B but that’s fine :)
It's "trivially easy" for you and 10% of your investments are expected to have "massive breakthroughs". Your strategy of "give money to good companies and founders" should easily enable you to reach $1B AUM!
Windsurf has their own models though
You're right! I had forgotten about SWE-1 family
not that they're great or very remarkable, but they can get the job done, and in a world where anthropic tries to assert dominance by fucking cursor seems relevant to the future to have your own models
but yeah that list was very silly anyways
As an American. I'd rather a single well established player get a large contract and actually deliver, than 20 disjointed companies each get 1/20th of the problem, have to work in concert, and possibly not even deliver at all.
This isn’t a grant to push for innovation. This is a promise from the orange man administration to the people and companies that donated to his "inauguration fund"
This is a kleptocracy but with extra steps. People are unfortunately numb to it.
Which AI company _should_ the DoD purchase from?
Lobbying, probably
That is why we need folks like you running the government and not asshats that are currently in positions ruining it all for all.
That’s not a lot of money between four companies.
It’s up to $200M for each of them. From the actual source:
“The awards to Anthropic, Google, OpenAI, and xAI – each with a $200M ceiling – will enable the Department to leverage the technology and talent of U.S. frontier AI companies to develop agentic AI workflows across a variety of mission areas.”
(https://www.ai.mil/Latest/News-Press/PR-View/Article/4242822...)
It’s still not a lot. How much do tokens cost for example?
In theory if it’s just labor with some profit mixed in, then you might be looking at 600 employees for each company.
I doubt it is just labor. Quote says $200 million ceiling. So maybe a time and materials (T&M) contract? It’s a ceiling so it’s not like they earn or are guaranteed $200m.
Has to include token or cloud computing time too. Which Google owns and can amortize themselves since it’s a capital asset to them. I don’t know much about the cloud computing background of Anthropic or if they are using Azure or AWS.
I think my original point is still valid it’s not a lot when you look at it
It will be enough to deliver the requirements for the DoD. They don’t need to spend any more or any less than what is required. Your point seems irrelevant if you don’t know their exact requirements.
It’s a $200m ceiling and my point was that it’s not a lot of money considering how other DoD contracts pay like CPIF type rates.
Meaningless comparison. If xAI or Google accepts the contract, they intend to deliver within the budget. You can't say whether it's too much or too little without knowing all of the details, which you don't.
It's on the DoD contracts site. I was a little off. It is a FFP style contract. Here is the OpenAI one (https://www.defense.gov/News/Contracts/Contract/Article/4218...)
"OpenAI Public Sector LLC, San Francisco, California, has been awarded a fixed amount, prototype, other transaction agreement (HQ0883-25-9-0012) with a value of $200,000,000. Under this award, the performer will develop prototype frontier AI capabilities to address critical national security challenges in both warfighting and enterprise domains. The work will be primarily performed in the National Capital Region with an estimated completion date of July 2026. Fiscal 2025 research, development, test and evaluation funds in the amount of $1,999,998 are being obligated at time of award. Office of the Secretary of Defense Chief Digital and Artificial Intelligence Office, Washington D.C., is the contracting activity. "
It's a prototype and its FFP. Only about $2m has been allocated to them. That equates to about 1-2 employees for a year.
Anthropic’s is “is expected to mirror OpenAI’s structure: a similar token obligation at signing and the rest released via milestones.”
All these frontier AI OTAs follow the same pattern of “up to [X] million” ceilings with actual funding phased out as projects progress. This mirrors what Palantir got last year.
Link: https://www.nextgov.com/defense/2024/06/pentagons-ai-office-...
Nice, should be enough for them to outbid Meta to retain 1 employee (https://news.ycombinator.com/item?id=44308000)
Testing the waters. I’d guess if it pays any dividends things will ramp up. This is also across all suppliers. At some point they’ll put more wood behind fewer arrows.
So basically one Ruoming Pang each...
Microsoft said they were beefing up their expenditure in AI with and or around the announcement of layoffs.
So does this mean all the web designers, web developers and many other white collar jobs will now be done by one such professional using AI; so XYZ use to require ten people to get the job done now only one who uses AI gets the tasks done (tasks that those ten use to use software applications to complete). All the while a few hundred of Ruoming Pangs make more money then God in which their work further helps killing white collar jobs.
Is there anyone else concerned about this? I am federal worker indirectly yet per some news today not sure how much longer I will be and whether or not it wise to go look for another web design/developer/UX Researcher (think this is safest out of three as you are talking to ppl)position. There are throngs of others looking now to compete against including now competing against AI for less jobs.
try https://www.europarl.europa.eu/topics/en/article/20230601STO...
I'm surprised trump let XAI get money. maybe the breakup was staged
lol, 150 mil for these AI company is not even cover their 1/10th monthly cost
its more of a "formal" cooperation than the actual work
It’s easiest way to repay Elon’s $150 million. Now listen to this song again as you digest that:
https://youtu.be/VYOjWnS4cMY
https://www.youtube.com/watch?v=LpxT9TLGoLI
Kyle Reese is a crackpot and the Strategic Knowledge Yielding New Emerging Technologies grant program is important for re-industrializing America
The DoD, DARPA, NASA, tax funded arms of the US should have all the resources to train a frontier model.
OpenAI is above 10 billion ARR and still growing fast.. this seems tiny in comparison?
2% of a company's revenue is definitely not tiny. And regardless, there's still reason to participate and hope the number gets bigger in the future.
What ever happened to the JEDI contract?
In the words of Will Stancil: AYFKM?
In fairness to poor Will, this contract was probably decided weeks or even months ago. The DoD isn't known for moving quickly or being responsive.
[dead]
I guess Zuck got the shaft?
The $200M would only pay for a single researcher at Meta. ;)
Seems small???
Selective corporate stimulus. What happened to free markets?
musk is a tenderpreneur.
this is simply a loyalty payoff for supporting the current admin. same as the other tech companies that got these contracts.
... Are you not aware of the current Musk-Trump drama? He's literally making a new party
> ... Are you not aware of the current Musk-Trump drama? He's literally making a new party
The drama where he outed the president as a pedo, deleted it, and went back to boot licking?
The fact that XAI is in this list is just blatant corruption. Their CEO was a government employee until a month ago.
If they weren't there it would raise just as many eyebrows, wouldn't it?
Meta not being on the list is more suspect IMO. At least it seems to me that Meta is where the actual talent/potential is.
Grok is pretty good, both the model in the API and the consumer products built on top. Llama is way behind, and the meta.ai app is much worse than the leading chat assistant products.
Not really. It would be obvious to anyone who's ever seen a government contract that they were excluded for ethics reasons.
And their chatbot just had a Nazi meltdown last week.
[flagged]
Why would that exclude them from the running? Should government contracts not be granted on the merits of the receivers? Grok clearly exists in this space so it’s not like they’re rewarding vaporware.
> Should government contracts not be granted on the merits of the receivers?
They should, but businesses owned by government employees should be excluded because it's too easy to corrupt the process. In fact, they have explicit rules about not doing that.
But he’s no longer a “special government employee” anymore either. Or are you suggesting he’s blacklisted from all government contracts for life because he previously worked for the current administration?
Not OP, but "For life" is a far cry from a month or two after. But yes, i'd argue we have no choice but to attempt to aggressively put bounds between government and profiteering. Lest we have Congress openly insider trade..
Of course not for life, but generally to avoid ethics issues the government requires you to be out for at least a year or two before engaging in business with the government.
He’s not a government employee. You said it yourself lol
But he was very recently. Usually you have to wait a while after leaving the government to sell to it.
> Should government contracts not be granted on the merits of the receivers?
The merits? We just had, in the last _week_, a huge scaldal where Grok was spewing racist, pro hitler content, even calling itself MechaHitler.
The Age of Grift continues
[dead]
This money should be going to companies that need 200M.
For 200M Google will open an account, send an email that says:
You're account is ready, there is $40m left on the retainer. We can code up some email template for 40M if you want.
Would that align with our current administration's beliefs?
or not spent by the government? I mean, unless this is for some really specific thing that really needs a push, I think there's sort of enough money flowing this way already, today. (5 years ago, again a different story)
Payback for the presidential library gifts. The Defense Department doesn't need one hallucinating chatbot let alone four...
Hallucinating autonomous drones on the other hand...
[flagged]
[flagged]
Meanwhile in Europe, we're sleeping on regulation and no real plan to face the challenges and opportunities linked to AI...
https://digital-strategy.ec.europa.eu/en/news/eu-launches-in...: “EU launches InvestAI initiative to mobilise €200 billion of investment in artificial intelligence”
And from https://digital-strategy.ec.europa.eu/en/policies/european-a...: “Both the Horizon Europe and Digital Europe programmes will invest €1 billion per year in AI.”
[flagged]
Love the scare quotes around the very things that make life worth living. America at least has the National Park system for metrics like "nature". Oh wait...
https://www.theguardian.com/us-news/ng-interactive/2025/may/...
Ah yes because Europe is famous for protecting user privacy and doesn’t have a long list of counterexamples despite all of its regulations
At least we have a real example of landable rockets.
Spain is paying 1000 euro a month to their soldiers, that's how seriously we take defense
They'd have more money now if they didn't skimp on ships 400 years ago. Gotta play the long game my dudes
Is 'X' is going to develop an "Agentic" weapon to hunt down Will Stancil?
(Only partially joking here)
This actually makes sense because in the meantime Meta is ditching the open-source (open-weights) direction.
Before the national security narrative took over, the main argument was about "safe" AI, where releasing models as open weights was considered "not safe." Now that no major US AI players release premium open-weights models, the "safety" narrative isn't needed anymore—so cooperating with the US military is feasible again.
Hopefully we'll be lucky and brave patriots will take great risk occasionally leaking those models so we taxpayers can enjoy them too (not just Northrop-Martin).