Eye on Policy

As the federal government heads into the crucial fourth quarter buying season, the emergent theme centers on numbers of suppliers – growing in some domains, shrinking in others.

SBA moves to end “abuse” of the 8(a) program.

If you want to understand a sea change in the small business set aside program for federal contracting, get to know the phrase “rebuttable presumption.”

The Trump administration’s Small Business Administration proposed a new rule for eligibility to the 70-year-old 8(a) program. It rebuts the presumption that a business 51% (or more) owned by people of a racial minority are economically disadvantaged. Companies will face a higher bar, to present data proving discrimination or economic disadvantage. Self-certification alone will not suffice under the proposed rule.

You can see both political and practical considerations in how the SBA explained in the rule: “The Trump SBA ended the practice of approving admissions to the 8(a)-program based solely on unsubstantiated claims or Biden-era narratives of racial discrimination immediately upon taking office. Today’s proposed rule will make the change permanent by replacing race-based eligibility with one standard for all applicants and requiring verifiable, fact-based evidence of social disadvantage.

Eligibility rules for other programs – notably Alaska native, Hawaiian native and Community Development Corporations – will not change.

I see this as more a problem for the companies seeking 8(a) admission than for government buyers. The latter can still buy, theoretically, from any company that makes a reasonable bid. Will cutting abuse of the 8(a) program discourage legitimately disadvantaged entities from trying to enter? Too soon to tell.

From the industry standpoint, gaining the advantages of 8(a) status for set-aside contracts will definitely be harder. In fact, the SBA says it has struck 150 8(a) firms based in Washington, D.C. It suspended another 1,000 companies and initiated termination actions against another 650.

SEWP VI charges ahead.

A highly popular government-wide acquisition contract program looks as if a political battle over where it has housed has barely made a dent. NASA’s SEWP crew has awarded nearly 1,500 companies more than 2,100 spots on the latest iteration of the long-running program.

SEWP VI awards include biggies like IBM, Dell Federal Systems and Booz Allen Hamilton. So are hundreds of small and medium-sized companies. The awards fall into three groups – IT solutions, enterprise solutions, and IT mission-based services. A new wrinkle: NASA will now let agencies establish blank purchase agreements with vendors in a strategic sourcing solutions category.

NASA says it expects order for SEWP VI to commence November 1, even though it plans to hold SEWP V open through January 2027. The new GWAC is worth $20 billion over 10 years.

Hanging over the SEWP program: When and how will SEWP depart NASA and migrate to control by the General Services Administration, as the Trump administration has proposed.

NextGov reported that GSA has “embedded” people at the still-NASA program office, in the interest of ensuring continuity. It also reported that GSA alone has executive agent designation to run GWACs, but this is unclear. NASA, after all, is still part of the Federal Acquisition Regulation Council.

NASA has a more efficient record running SEWP than the GSA has in running its various GWACs. It is no trivial task managing thousands of answers to solicitations and fielding multiple protests. Some GWACs never make it to final award and operation, or they take many years in gestation.

SEWP VI had its share of protests, all of which were either tossed out by the Government Accountability Office or resolved by the SEWP program office. SEWP VI solicitations issued in 2023. Yet, as Federal News Network reported, SEWP VI ended up significantly bigger in scope and rosters of vendors than SEWP V.

This is all in stark contrast to the collapse of the CIO-SP series of GWACs operated by the National Institutes of Health’s acquisition shop.

In any case, it does not look like GSA is stumbling over itself to take over the SEWP program, perhaps because there is no evidence it could do a better job than NASA.

True, a March 2025 executive order theoretically consolidated governmentwide procurement authority with the GSA. But the same EO said GSA did not have to take this on in all cases. It is hard to know what the administration really, really wants, given the number (225 last year) and range of EOs it is published, including orders on shower head water pressure and the fallibility of paper straws.

AI: Market research is king.

Among the most critical functions for achieving modernization, digital government is market research. The activity required not only identifying vendors, but also understanding where they fit into a market, how they differentiate from one another, and whether contracting officers and program managers even know the extent of the market.

Artificial intelligence ranks the highest in difficult market research.

The market is quickly fragmenting, with new companies forming daily. It is a full-time job just to keep up.

Many factors drive the AI market, which is still in relative infancy despite the vast investment dollars it draws – or perhaps because of it. A few things going on:

• Chinese-developed models have nearly caught up in capability and quality. Axios reported on Microsoft incorporating the Chinese DeepSeek model into its cloud offerings as a lower-cost alternative to Anthropic. That is not to say Microsoft will use is China data centers or Chinese LLMs for U.S. federal customers, but it is something federal buyers would want to be aware of. Meaning: You have to know the origination of any of the burgeoning number of model suppliers.
• Open-source models have gained capability, and middleware companies with access to brokered clouds give customers affordable computing power for “inference,” the execution of AI processes. Meaning: You have to know where your agency’s inference will occur and who is sharing the facility.
• Startups specializing in open source have drawn billions in funding. Just days ago, a four-year-old company called Together.ai announced $800 million in third round funding. The company offers numerous models, many open source but also the proprietary and Chinese ones. Meaning: It is important to understand the nuances among open source, open weighting, and fully proprietary models and for what types of tasks each is optimized.

Choosing among, say, six 8(a) companies to run facilities will feel like a cakewalk when compared to understanding the market for the full stack of AI products. Some companies will make it, some will not. Others will buy others or they will be acquired. Market research will be crucial for agencies trying to build dependable, durable, safe, and compliant AI services.