Understanding synthetic intelligence spending by the U.S. federal authorities

In our prior sequence of papers for Brookings, we explored the rise of nationwide synthetic intelligence (AI) strategy documents and sought to make sense of what each country was trying to do and the way successfully they had been doing it. In our concluding paper, we centered on the place the U.S. was lagging behind and proposed choices to treatment the lagging. Particularly, we really helpful three choices: 1) apply classes from the U.S. area race to invigorate expertise improvement, (2) undertake a multi-national consortium method (just like NATO) and (3) create a strong partnership with one different nation.

Following the steering of “Deep Throat” of Watergate fame, on this new sequence of articles, we observe the federal path of cash to grasp the federal marketplace for AI work, the {hardware}, software program, and companies being bought. We additionally observe the important thing gamers who allocate the cash (legislators), spend the cash (program managers), and obtain the cash (distributors). Taken collectively, this sequence gives a complete have a look at federal IT spending, its path, and its key gamers.

Information and Methodology

The entire knowledge for this sequence has been taken straight from federal contracts and was consolidated and supplied to us by Leadership Connect. Management Join has an in depth repository of federal contracts and their knowledge varieties the idea for this sequence of papers.

On this preliminary paper, we analyzed all federal contracts during the last 5 years that had the time period “synthetic intelligence” (or “AI”) within the contract description. As such, our dataset included 663 contracts of which 652 recognized a funding company, 556 recognized a NAICS code and 504 recognized a greenback worth, with contract values ranging as much as $192 million.


Out of the 15 NAICS code classes recognized in contracts, the info present that over 95% of the AI-labeled expenditures are in NAICS 54, which is for skilled, scientific, and technical companies. This class had over $1 billion in expenditures with a complete of 474 distinctive award pairs (agency-vendor) during the last 5 years. This isn’t stunning, as usually the federal authorities is likely one of the largest sources of exterior funding for analysis and improvement, which is the core service of NAICS 54. The second most regularly used code was NAICS 33 (manufacturing), with solely 16 distinctive awards valued underneath $7 million (0.5% of the entire expenditures), so it’s clear that NAICS 54 dominates. Total spending for analysis and improvement (of which NAICS 54 is a part of) has grown at a double digit rate for several years. This discovering suggests an immature market which is concentrated on the event of analysis fairly than a extra mature market which might be centered extra on {hardware} and software program gadgets.

Whereas there are 15 totally different funding companies inside NAICS 54, 53% of the contracts and 87% of contract worth sit inside the Division of Protection (DoD). Well being and Human Companies (HHS) and NASA are distant second to the DoD by way of variety of contracts (with 17% every). Taken collectively, these contracts solely symbolize 1.5% of whole contract spend for this code. Commerce and the Veterans Administration are the subsequent largest by contract spend at roughly 4% every.

There are a complete of 307 totally different distributors with 474 contracts. No vendor offers with greater than three funding companies, which displays a really area of interest method for the seller group. Three distributors (AI Options, AI Sign Analysis and United Options) cope with three companies whereas fourteen totally different distributors cope with two companies and the remaining 290 distributors cope with a single funding company. Moreover, there are solely 62 corporations with a couple of contract however there are 245 with only one. There is just one vendor with greater than 10 contracts (AI Options) and solely two others with greater than 5 (AI Sign Analysis and AI Biosciences).

Whereas there are 15 totally different funding companies inside NAICS 54, 53% of the contracts and 87% of contract worth sit inside the Division of Protection (DoD).

There are 327 funding agency-vendor dyads. Of all of the dyads, AI Options is the one vendor with greater than 5 contracts with a single funding company – 56 with NASA and 10 with DoD. There are 267 distinctive dyads the place just one contract exists. Once more, this displays a really fragmented vendor group that’s working in very slender niches.


Taken collectively, we see a highly-fragmented market which is dominated by smaller distributors who usually have a single contract for AI-related companies. Many of those distributors are small distributors and, for these distributors, the AI-related work represents a considerable proportion of their annual income. Curiously, co-location additionally options prominently. That’s, many of those distributors are positioned in shut proximity to their federal consumer and we suspect that prior relationships – private or skilled – might exist. We see these relationships as wholesome because it displays an ecosystem of distributors which can be rising in response to particular wants.

This doesn’t imply that the bigger conventional know-how corporations equivalent to RAND, Northrop Grumman, Accenture, IBM, Booz Allen Hamilton and Raytheon are absent from these contracts. Nevertheless, like with their smaller counterparts, these bigger corporations are nonetheless establishing beachheads the place AI may probably match. If historical past holds, we’d count on these bigger corporations to comparatively rapidly explode of their total contractual presence given their beforehand present relationships with federal resolution makers and their historic willingness to accumulate smaller corporations to rapidly scale up.

From the consumer aspect, the massive variety of comparatively small contracts reveals that the federal authorities remains to be very a lot in an experimental part of buying AI and is probably going in search of particular use circumstances the place AI is acceptable. This is able to clarify the concentrate on research-based contracts versus {hardware} and software-based contracts. From a threat administration standpoint, this can be a good thought. With numerous small distributors every having a single contracts, we understand that the federal government is adopting a technique of letting a thousand flowers bloom, with the hope that it will result in ultimately determining one of the best method to AI.

Nevertheless, there’s additionally a darker future that might probably end result if this method is taken. As mentioned beforehand, the U.S. is considerably behind in AI improvement and, specifically, trails China by a considerable margin. In our analysis of China’s national AI plan, it’s clear that the Chinese language nationwide authorities is taking the lead on the event of AI methods and tasks. In the meantime, different entities, such because the non-public sector, are very small gamers within the plan. Which means China can rapidly establish one of the best prospects for AI and direct substantial funding on a couple of tasks which would seem to have the best diploma of success.

Against this, the U.S. plan is rather more of a collaborative method, with the federal authorities, non-public industries and universities all anticipated to play an energetic function. Whereas we now have little doubt that the U.S. technique will ultimately yield good outcomes, it should probably be a extra slowly evolving technique since preferential therapy just isn’t being given to probably the most promising concepts. A minimum of at current, the main target seems to be on cultivating all kinds of tasks fairly than specializing in a extra slender set.

The one main focus space for the U.S. federal spending on AI is inside the DoD. It seems that the DoD is getting used as an incubator of kinds for AI tasks, regardless of being solely one among many companies envisioned to take action within the U.S. nationwide plan. However, given the DoD’s willingness to spend on new applied sciences, it’s not stunning that that is the foremost focus space. We might count on that, over time, different federal companies can be rather more prepared to put money into AI.

Whereas the U.S.’s broad method to AI will probably yield advantages sooner or later, it’s prone to be a way more slowly evolving technique than the one embraced by China. Additional, the extra diversified U.S. technique makes it difficult for federal venture/program managers to share insights developed from the person tasks. Which means the federal authorities is basically relearning the identical classes from a number of tasks, which isn’t a time-efficient technique.

The present framework for fixing this coordination challenge is the National Artificial Intelligence Initiative Office (NAIIO), which was established by the Nationwide Synthetic Intelligence Initiative Act of 2020. Underneath the Act, the NAIIO is directed to:

maintain constant help for AI R&D, help AI schooling…help interdisciplinary AI analysis…plan and coordinate Federal interagency AI actions…and help alternatives for worldwide cooperation with strategic AI…for reliable AI techniques.

Whereas the intent of this Act and its formal construction are admirable, the present federal spending doesn’t appear to mirror these lofty objectives. Somewhat, we’re seeing a federal market that seems to be rather more chaotic than fascinating, particularly given the lead China already has on the U.S. in AI actions.


The federal marketplace for AI is in its very early levels however already we see how the market is probably going going to maneuver within the coming years. Whereas we predict that the diversified method taken by the U.S. is prone to ultimately repay, it’s equally prone to trigger the U.S. to fall even additional behind China over the subsequent a number of years. Enhancing the ability and path of NAIIO can be a superb first step to convey construction to this chaotic market.

Constructing on the findings from this paper, additional papers on this sequence will discover the multi-faceted relationships between federal program managers, AI consulting corporations, and elected representatives.

Be the first to comment

Leave a Reply

Your email address will not be published.