Government AI Procurement Guide: Contract Vehicles & Set-Asides

Government agencies buy AI through pre-competed contract vehicles — the GSA Multiple Award Schedule, OASIS+, and IT GWACs like 8(a) STARS III — structuring the purchase as a task order or Blanket Purchase Agreement, scoping the work by labor category, and often routing it through a small-business set-aside. Understanding which vehicle fits an AI requirement is the difference between a program that ships in a quarter and one that dies in the acquisition queue.

Procurement, not technology, is where most public-sector AI stalls. The model works; the buy does not. An agency cannot sign up for an AI platform the way an enterprise swipes a card — it must satisfy competition rules, socioeconomic set-aside goals, and new AI-specific contract terms, all through an approved vehicle. Gain America exists to move work through that machinery: we staff and subcontract the forward-deployed engineers, data-center teams, and MLOps talent that actually deliver AI under these vehicles, plugging into the primes and integrators who already hold the paper.

The federal AI contract vehicles that matter

A contract vehicle is a pre-competed master agreement that lets an agency issue task orders without running a full open-market acquisition each time. For AI, four families cover the overwhelming majority of federal spend.

GSA Multiple Award Schedule (MAS). The MAS — and specifically its Information Technology category, which absorbed the old Schedule 70 — is the default path for AI software, cloud services, and IT-services labor. It is the largest and most flexible vehicle, holding hundreds of thousands of commercial products and services under negotiated ceiling prices. In 2026 GSA is proposing sweeping MAS changes, including standardized AI terms and broader transaction-data reporting that adds 112 previously exempt special item numbers. For most agencies, an AI proof-of-concept that needs to reach authorized production will run through a MAS IT order.

OASIS+. OASIS+ is GSA's family of government-wide, best-in-class contracts for complex professional services — the vehicle of choice when an AI requirement is really an integration, transformation, or advisory engagement rather than a software license. As of 2026 OASIS+ has moved into Phase II with 13 service domains, no contract ceiling, and a continuous, rolling-enrollment model that keeps its solicitations open indefinitely. More than 1,300 small businesses hold small-business-pool awards alongside several hundred unrestricted contractors, giving agencies deep bench options for AI systems-engineering work.

8(a) STARS III and agency GWACs. Governmentwide Acquisition Contracts (GWACs) are IT-specific vehicles usable across agencies. The 8(a) STARS III GWAC is a $50 billion small-business set-aside reserved for SBA-certified 8(a) firms, with an ordering period running through 2029 and task-order performance extending into the early 2030s. Other GWACs — NASA SEWP for hardware-heavy buys, NIH CIO-SP families for health IT, and agency-specific IDIQs — carry AI work as well. GWACs are attractive for AI because they are built for IT scope and pre-clear the socioeconomic credit an agency needs.

The vehicle you pick is not a formality. It determines who can bid, how fast you can award, which labor categories price your engineers, and whether the buy counts toward an agency's small-business goals. Choosing wrong can cost a program six to twelve months.

Small-business set-asides: 8(a), HUBZone, SDVOSB, and WOSB

Federal buyers operate under statutory small-business goals, and set-asides are the primary tool for meeting them. For AI programs, set-asides shape who can prime the work — and where a firm like Gain America subcontracts specialized delivery talent behind a small-business prime.

• 8(a) Business Development. The SBA's flagship program for socially and economically disadvantaged firms. It uniquely permits sole-source awards below defined thresholds, making it the fastest path to put an AI capability on contract without a full competition. STARS III is the dedicated 8(a) IT GWAC.
• HUBZone. Reserved for firms in Historically Underutilized Business Zones. HUBZone set-asides and price-evaluation preferences steer AI work toward firms rooted in economically distressed communities.
• SDVOSB. Service-Disabled Veteran-Owned Small Business set-asides — often the preferred path at the VA and DoD, both heavy AI buyers for records intelligence and constituent service.
• WOSB / EDWOSB. Women-Owned and Economically Disadvantaged Women-Owned Small Business set-asides apply in industries where women are underrepresented, which includes many technical NAICS codes relevant to AI services.

The practical pattern: a small disadvantaged prime holds the vehicle and the set-aside credit, and staffs the hard AI roles through subcontractors and staffing partners. That is precisely where Gain America operates — supplying forward-deployed engineers for government and cleared MLOps talent to the small-business primes who win the set-aside but cannot carry a full frontier-AI bench in-house. Working out of a hub like government AI consulting in Virginia puts that talent inside the federal contracting corridor.

GSA AI acquisition guidance and 2026 regulatory changes

The biggest 2026 shift is the arrival of standardized AI contract terms. Building on OMB memo M-25-22, "Driving Efficient Acquisition of Artificial Intelligence in Government," GSA is incorporating AI-specific provisions into the MAS so that terms are consistent across the government for the first time. Two provisions matter most for buyers and vendors alike.

First, data-protection terms: vendors must get agency permission before using non-public government data to train publicly available AI models. Second, testing and monitoring rights: contracts increasingly build in ongoing performance testing and monitoring, so an agency can verify a model keeps working after award — a contractual echo of the continuous evaluation that production AI demands.

These terms interlock with the compliance stack. An AI system almost always still needs FedRAMP authorization before it touches real data, and NIST's AI Risk Management Framework increasingly shapes the security and governance language written into task orders. Procurement, authorization, and governance are three gates, not one — and a buy structured to satisfy only the first will stall at the second. The reason so many government AI projects fail is that teams treat the acquisition as the finish line rather than the starting gun.

Sole-source, BPAs, and task orders: how AI services get scoped

Owning or accessing a vehicle is only the setup. The actual buy takes one of a few shapes, and how an AI requirement is scoped inside that shape determines how the work — and the staffing — gets priced.

Task orders are the workhorse. Against a multiple-award vehicle like MAS, OASIS+, or a GWAC, an agency competes a task order among holders and awards the work. Competition rules are satisfied by the underlying vehicle, so an AI task order can move in weeks rather than the year an open-market acquisition consumes.

Blanket Purchase Agreements (BPAs) are established against a Schedule to lock in terms, ceiling rates, and labor categories for recurring needs — ideal for an AI program that will issue repeated calls for model tuning, evaluation, and MLOps support over a multi-year deployment.

Sole-source awards skip competition where justified: an 8(a) sole-source below threshold, a brand-name justification, or an only-responsible-source finding. Sole-source is the fastest path onto contract but carries the heaviest documentation burden.

Scoping the AI work itself is where programs succeed or fail. A well-scoped AI task order separates the software/cloud buy (the platform, GPUs, or inference capacity) from the services buy (the engineers who integrate, deploy, and operate it). Conflating the two — buying a platform and assuming it deploys itself — is a recurring cause of enterprise and public-sector pilots that never reach production.

SLED procurement: state IT contracts and cooperative purchasing

State, local, and education (SLED) buyers reach AI through a different, and often faster, set of channels. The SLED market exceeds $1.5 trillion in annual procurement across more than 90,000 entities, with state and local enterprise IT spending projected around $125 billion by 2026 — a vast AI opportunity that does not touch a single federal vehicle.

The dominant SLED channel is cooperative purchasing. A lead state competitively awards a master contract, and then tens of thousands of other agencies can buy off it without running their own procurement. The largest is NASPO ValuePoint, the cooperative arm of the National Association of State Procurement Officials, which aggregates demand across all 50 states, D.C., territories, and their subdivisions. Cooperatives like OMNIA Partners work the same way. A vendor that wins one cooperative award can, in effect, sell AI services to thousands of jurisdictions off a single contract.

SLED buyers also use statewide IT contracts and IDIQs run by each state's central procurement office, plus their own small-business and diverse-supplier preferences that mirror the federal set-aside model. On the compliance side, many states now require StateRAMP/GovRAMP authorization — the state analogue to FedRAMP — before an AI cloud service can process resident data, and justice-adjacent AI must satisfy CJIS requirements. The vehicle gets you in the door; authorization lets you stay.

Staffing AI talent under a vehicle: labor categories and delivery

Every services buy — federal or SLED — ultimately prices against labor categories (LCATs). A vehicle defines roles such as data scientist, machine-learning engineer, cloud/DevOps engineer, and enterprise architect, each with a negotiated ceiling rate and minimum qualifications. An AI task order is built by mapping the required work to those categories and staffing bodies against them.

This is the choke point. The LCATs on a vehicle assume you can find the people — and most agencies and even many primes cannot hire frontier-caliber AI engineers at the rates the market commands. That is the public-sector AI talent gap in operational form: the vehicle is in place, the task order is funded, and there is no bench to staff it. Rather than absorb a $500K-plus salary load per engineer, buyers increasingly augment their teams rather than hire, contracting the talent through a vehicle holder.

A contract vehicle is a permission slip, not a workforce. The agencies that ship AI are the ones that pair the right vehicle with a partner who can actually staff the labor categories on it.

That is Gain America's role across the procurement landscape. We staff and subcontract forward-deployed engineers, MLOps, data-center delivery teams, and public-sector-ready talent to the primes, integrators, and small-business set-aside firms who hold the vehicles — MAS, OASIS+, STARS III, and the state cooperatives — but need the delivery bench to execute. When an agency needs a government AI staffing firm that understands both the acquisition and the engineering, mapping the requirement to the right vehicle and the right LCATs is the first deliverable, and putting qualified engineers behind it is the second.