# GSA Claims $1.68bn in OneGov Savings. It Rejected Auditors' Call for Cloud-Cost Reporting.

Published: 2026-11-01

The Trump administration's federal buying agency says its OneGov technology deals have saved roughly $1.68 billion. But the accounting question is more complicated than the headline: $1.4 billion of that figure was attributed by the General Services Administration to AI agreements, while a separate June 2026 Government Accountability Office review found 17 of 24 major agencies struggling to control cloud costs. GAO urged GSA to require cloud financial-management practices and reporting of benefits. GSA Administrator Edward Forst disagreed, arguing that his agency lacked authority to mandate them. GAO pointed to the very executive order that had enlarged GSA's procurement role. These are not findings that OneGov's claimed savings are false; GAO did not audit that total. They expose an accountability gap between negotiating a lower price and proving what government actually pays over time.

For related context, read Factrail's [Donald Trump dossier](https://factrail.ai/people/donald-trump), [Scott Bessent dossier](https://factrail.ai/people/scott-bessent), and [fiscal sustainability analysis](https://factrail.ai/drivers/fiscal-policy-sustainability).

## What the documents actually establish

Three records have to be read together, rather than made to say more than they do. [President Donald Trump](/people/mig-donald-trump)'s [Executive Order 14240, signed March 20, 2025](https://www.whitehouse.gov/presidential-actions/2025/03/eliminating-waste-and-saving-taxpayer-dollars-by-consolidating-procurement/), directed a consolidation of common federal purchasing and a stronger role for GSA in government-wide IT acquisition contracts. GSA's [September 10, 2026 OneGov announcement](https://www.gsa.gov/about-gsa/newsroom/news-releases/gsa-expands-onegov-ai-offerings-with-discounted-openais-chatgpt-09102026) asserted approximately $1.68 billion in cumulative savings, including approximately $1.4 billion from AI agreements. Between them sits the [June 23 GAO cloud-procurement audit, GAO-26-107530](https://www.gao.gov/products/gao-26-107530), which documents uneven federal cloud-cost controls and a disagreement over whether GSA can require agencies to apply a discipline known as FinOps and report its benefits.

That linkage is this investigation's contribution: a map of which institution negotiated discounts, which institution bears operating costs, and which can demand evidence that the savings survived actual usage and renewal. The GAO study is not a forensic examination of the OneGov savings methodology. It would be incorrect to deduct its 17-agency finding from GSA's $1.68 billion, to call the claimed savings fraudulent, or to infer a taxpayer loss of that amount. The narrower, consequential finding is that the government has promoted centralized purchasing as a measurable efficiency reform while its auditors identified weaknesses in the information and management practices needed to compare cloud costs across agencies.

A negotiated discount is evidence of a price concession relative to a defined benchmark. It is not, without purchase quantities, baseline prices, consumption records and renewal terms, proof of equivalent cash savings. An incentive credit can lower a customer's bill but is not itself a Treasury receipt. A contract ceiling is not a payment. These distinctions matter because federal cloud contracts, GAO says, grew from about $2.3 billion to more than $10 billion a year over roughly a decade. Even a modest persistent measurement error can matter at that scale; the audit does not calculate one.

## Documentary timeline: central authority, discounts and resistance

March 20, 2025. Trump signed Executive Order 14240. Agency heads were directed to propose transferring purchasing of common domestic goods and services to GSA where legally permitted. The order also instructed OMB to designate GSA's administrator as executive agent for government-wide IT acquisition contracts, with an exception to preserve continuity where necessary. The order did not abolish individual agencies' statutory responsibilities, nor did it give GSA unlimited control over every cloud workload. The text itself conditions implementation on applicable law and available appropriations.

July 18, 2025. OMB Director [Russell Vought](/people/russell-vought) issued [memorandum M-25-31, Consolidating Federal Procurement Activities](https://www.whitehouse.gov/wp-content/uploads/2025/07/M-25-31-Consolidating-Federal-Procurement-Activities.pdf). It set out two workstreams: greater use of GSA-managed central contracts and centralization of procurement functions where that improves economy and efficiency. That policy supplies an accountability benchmark: centralization is supposed to improve results, not merely move negotiations from one office to another.

August–September 2025. GSA announced an [AWS OneGov arrangement](https://www.gsa.gov/about-gsa/newsroom/news-releases/gsa-announces-onegov-agreement-with-aws-08072025) with up to $1 billion in aggregate incentive credits for federal civilian agencies through December 2028. It also announced a [Microsoft OneGov agreement](https://www.gsa.gov/about-gsa/newsroom/news-releases/multibillion-dollar-gsa-onegov-agreement-with-microsoft-brings-steep-discounts-09022025) with potential first-year savings of $3.1 billion under specified offers, including time-limited Copilot access and Azure-related discounts. These headline maxima are not additive to realized savings without proof of actual orders and the calculation method. AWS credits are conditional benefits; Microsoft's figure was presented as potential, not a payment already made.

January–May 2026. GSA advertised a [Broadcom agreement](https://www.gsa.gov/about-gsa/newsroom/news-releases/gsa-and-broadcom-onegov-agreement-saves-taxpayer-dollars-01212026) with discounts up to 64% for specified VMware-related products. On [April 29](https://www.gsa.gov/about-gsa/newsroom/news-releases/onegov-saves-taxpayers-11-billion-in-first-year-04292026), the agency said OneGov had saved $1.1 billion in its first year through 20 agreements. On [May 21](https://www.gsa.gov/about-gsa/newsroom/news-releases/gsa-announces-onegov-agreement-with-snowflake-to-accelerate-datadriven-technolo-05212026), it announced a Snowflake offer with 20% off compute, 26.67% off storage and potentially higher compute discounts as use rises. The offers differ in product, eligibility, pricing benchmark and duration; treating all percentages as comparable would distort the result.

June 23, 2026. GAO published its 24-agency cloud review. Seventeen agencies reported cloud-cost management challenges; 17 reported conflicting software guidance; 15 reported procurement obstacles caused by outdated regulations; 15 had difficulty obtaining authorized cloud solutions; 11 identified multicloud interoperability issues; and 10 reported cloud-workforce constraints. GSA rejected GAO's proposed requirement for FinOps adoption and benefits reporting. GAO maintained its recommendation, explicitly citing the 2025 executive order and OMB's procurement memorandum.

September–October 2026. GSA's [September 10 announcement](https://www.gsa.gov/about-gsa/newsroom/news-releases/gsa-expands-onegov-ai-offerings-with-discounted-openais-chatgpt-09102026) raised the agency's cumulative savings claim to about $1.68 billion, with $1.4 billion attributed to AI arrangements. The new OpenAI offer was for discounted, consumption-based access with no minimum orders; [Nextgov's reporting](https://www.nextgov.com/acquisition/2026/09/gsa-unveils-new-token-based-onegov-discount-openai/415908/) described the shift from introductory near-zero access fees to metered pricing. GSA [announced a 40% token-based discount with Anthropic on October 27](https://www.gsa.gov/about-gsa/newsroom/news-releases/gsa-anthropic-reach-onegov-agreement-on-discounted-consumptionbased-pricing-10272026). The trajectory matters: a promotional offer and a renewal price are two stages of the same purchasing relationship, not interchangeable measures of lifetime savings.

## The power map: who can say yes, and who can demand proof?

Trump is the political author of the consolidation order. OMB, led by Russell Vought when it issued M-25-31, sets government-wide procurement-management direction and coordinates implementation. GSA Administrator Edward Forst is the named executive who publicly defended the OneGov strategy and rejected GAO's FinOps recommendation. GSA negotiates and manages government-wide purchasing arrangements; the individual agencies still decide which eligible products to order, operate workloads, manage usage and comply with their own missions and appropriations. Contracting officers and agency technology managers therefore remain essential control points.

Congress controls appropriations and can revise the statutory definitions and acquisition framework. The FAR Council and procurement-policy authorities can update regulations. GAO can audit, recommend and report to Congress, but it cannot itself compel GSA to adopt a recommendation. The Federal CIO Council can spread practices across agencies; DHS's CISA can clarify software-component security guidance. Vendors can propose prices and conditions, but cannot lawfully rewrite the government's statutory authority.

The sharpest dispute concerns the boundary between purchasing power and operational control. In its comments reproduced by GAO, GSA argued it lacked authority to require agencies to use FinOps. It characterized FinOps as an agency management framework rather than a procurement obstacle and said GSA contracts could supply FinOps-supporting products and services. GAO disagreed with the jurisdictional conclusion, pointing to GSA's executive-agent role and its mandate to reduce inefficient IT purchasing. GAO wanted GSA to require agencies to use FinOps practices and report the resulting benefits.

This is a genuine institutional disagreement, not evidence that Forst secretly blocked a proven saving. A court has not been identified here as resolving the precise scope of GSA's authority to impose FinOps across agencies. The order's references to law and continuity give GSA a plausible argument that a centralized contract manager is not automatically empowered to dictate every department's internal budget operations. Conversely, GAO has a substantive point: a buyer entrusted with government-wide savings should have a reliable way to test whether negotiated prices translate into agency-level value. The unresolved choice is for executive coordination, Congress or a legally grounded implementation plan.

## What $1.68 billion does—and does not—mean

GSA's September breakdown permits one straightforward calculation: $1.4 billion divided by $1.68 billion is approximately 83.3%. Thus more than four-fifths of the agency's cumulative claimed savings at that date were attributed to AI agreements. This is a calculation from GSA's rounded figures, not a new audited accounting total. It should not be mistaken for the proportion of actual federal cloud spending going to AI, the amount of money paid to any vendor, or an estimate of the profit those vendors earned.

There are at least four separate quantities an oversight committee should request. First, the negotiated price advantage relative to a clearly identified comparator, with a list-price versus prior-paid-price distinction. Second, the quantity actually purchased or consumed under the agreement. Third, the net outlay over time, including integration, security, data transfer, staff training, metered usage and renewal. Fourth, the counterfactual: what the agency would have bought without OneGov, rather than assuming every eligible user would have paid full commercial list price.

A product offered at $1 per agency can generate a very large hypothetical discount against a commercial license price even if the agency would never have purchased full-price seats for all eligible staff. That does not make the offer worthless: cheap experimentation can deliver genuine public value. It means the accounting must label avoided list-price expenditure separately from realized budget reductions. Similarly, a $1 billion AWS incentive-credit ceiling does not establish that every dollar was earned, used or converted into reduced appropriations. Without account-level redemption records, it is a negotiated capacity, not a cash-flow finding.

The separate GAO audit shows why that accounting discipline is difficult. Officials at 22 of the 24 agencies said they primarily relied on historical procurement data when making cloud decisions. GAO found that cloud consumption does not map neatly onto older contract classifications, that 15 agencies encountered outdated FAR definitions, and that costs can vary with data storage, compute use and other metered services. A historical obligation can tell an auditor what was committed under a contract; it may not reveal the cost of a specific workload, the reason a bill rose or the expense of moving it elsewhere.

## Named commercial beneficiaries—and what cannot be inferred

The public OneGov record identifies actual commercial counterparties, not merely hypothetical sectors. Amazon Web Services offered up to $1 billion in incentive credits to encourage cloud modernization and training. Microsoft received a central role in government-wide software and cloud offers, including Microsoft 365, Copilot and Azure. Broadcom offered selected VMware products at large advertised discounts. Snowflake offered reduced compute and storage pricing. Google appears in GSA's [public OneGov inventory](https://itvmo.gsa.gov/onegov/) with cloud and AI offers. These are named firms positioned to compete for orders under GSA-negotiated terms; an offer is not proof of a specific revenue amount.

A distinct but relevant example shows the scale of enterprise bargaining: [Reuters reported on May 27, 2026](https://www.reuters.com/business/pentagon-awards-microsoft-97-billion-deal-bid-cut-costs-end-license-sprawl-2026-05-27/) that the Pentagon announced a five-year, approximately $9.69 billion Microsoft enterprise technology agreement intended to consolidate existing licenses and reduce duplication. Officials said it consolidated existing budgets rather than representing wholly new spending. This agreement should not be silently counted as an OneGov award or added to OneGov savings; it illustrates the broader leverage and concentration trade-off when federal buyers bundle purchasing.

For each firm, the mechanism is a possible combination of access to a very large buyer, lower customer-acquisition friction, wider adoption, recurring consumption and potential renewal business. That is an incentive analysis, not proof of excess profits. Vendors also surrender margin through discounts, face federal security requirements and compete for government business. The AWS credits and time-limited promotions can reduce public costs immediately, even if they also encourage future platform dependence.

There is no evidence in the records examined for this article that Trump, Vought or Forst personally profited from the agreements, that an award was purchased through donations, or that the named companies committed fraud. No such allegation should be inferred from the existence of negotiations. The missing public data are more prosaic and more testable: vendor-level orders, consumption, credit redemption, renewal prices, switching costs and verified net savings.

## Why cloud bills can outrun procurement discounts

Cloud is a variable-usage business. A government office can negotiate a lower unit price for compute and still spend more if it uses more compute, keeps idle resources running, duplicates environments or incurs higher network and storage charges. The opposite can also happen: improved architecture and usage monitoring can save more than a nominal discount. That is why FinOps—shared financial accountability for cloud consumption—matters independently of the legal form of the original purchase.

The GAO report describes cost management, outdated definitions and interoperability as distinct problems. A purchase agreement can improve the first invoice without solving them. Multicloud strategies can reduce dependence on a single provider but also increase complexity; agencies may need to test portability, monitor usage across providers and negotiate clear exit terms. DOD, in its response to GAO, objected to treating vendor lock-in as automatically negative. It argued that provider-specific capabilities can be a rational strategic choice if an agency understands its dependencies and documents an exit strategy. That is a substantive counterpoint: maximal portability is not always the cheapest or most capable solution.

The other challenge is regulatory timing. GAO said the FAR's IT definition was about two decades old and the regulations lacked a cloud-computing definition despite changes made in 2025. Agencies reported difficulty fitting pay-as-you-go services into familiar contract types. The [White House](/organizations/white-house)'s effort to streamline acquisition rules may simplify some processes, but GAO found the cloud-specific gaps still unresolved as of its review. This is not a claim that every contract is invalid. It is a warning that rules designed for different purchasing models can obstruct comparable bids and reliable oversight.

## The strongest counter-case: buying together really can help

The administration has a credible argument for centralized negotiation. When departments separately buy widely used software, they can pay different prices, duplicate procurement work and lose bargaining leverage. GAO itself describes OMB's earlier finding that prices for the same software could vary by up to 20% across agencies. Standardized terms and negotiated discounts can reduce that variance. A common procurement vehicle may also let smaller agencies access security and enterprise features they could not negotiate alone.

Independent reporting supplies additional evidence of use rather than mere publicity. [Nextgov reported in May 2026](https://www.nextgov.com/artificial-intelligence/2026/05/nearly-34m-users-across-government-can-leverage-ai-through-onegov-gsa-official-says/413588/) that GSA officials cited more than 120 orders against AI offerings and identified agencies using them. The report also quoted GSA's defense of the savings as measurable results. Those are meaningful signs of participation, though the reported orders and eligible-user counts do not independently validate a cumulative savings figure. In October, [Nextgov documented an OneGov deal with small firm (US)AI](https://www.nextgov.com/acquisition/2026/10/usai-becomes-gsas-first-direct-onegov-contract-small-business/416789/), complicating any simple claim that the program exclusively benefits established technology giants.

GSA's authority objection deserves serious treatment too. FinOps may require personnel, billing access, technology changes and governance choices that procurement officers alone cannot command. If Congress wants a uniform government-wide mandate, legislation or explicit OMB instructions may be cleaner than relying on an executive-agent designation. GAO's recommendation could be implemented with proportional reporting rather than imposing a costly new compliance system on every agency. The goal is not to make measurement so expensive that it erases the savings.

What weakens the counter-case is not the existence of discounts. It is the absence, in the sources examined here, of a public reconciliation that allows readers to trace GSA's aggregate claims through baseline, order, use, invoice and renewal. This article has not established that such records do not exist internally. It establishes that the public releases and the GAO audit do not supply that reconciliation, and that GAO and GSA disagree about who should require cloud-cost practices.

## Who pays if the numbers are wrong?

The public bears the financial risk through appropriated agency budgets. Taxpayers can benefit when agencies buy the same services for less, but can also finance idle subscriptions, higher metered usage or costly migrations if a discount becomes a commitment that is expensive to unwind. Federal employees and people relying on government services bear operational risk if agencies cut functionality to hit a price target or fail to secure cloud systems. Smaller vendors may gain a route into federal purchasing through shared contracts, or face difficulty competing against heavily discounted introductory offers by incumbents.

None of those possible outcomes is measured by the $1.68 billion figure alone. There is no basis to declare that it represents money lost, money stolen, or verified savings in every agency budget. The public-interest test is whether a federal buyer can show a consistent before-and-after financial and service-quality record, adjusted for usage and workload. An agency should not claim a budget saving merely because a supplier publishes a lower list-price discount; nor should a critic dismiss a real discount because it is negotiated with a large corporation.

## The records that would change this conclusion

A congressional oversight inquiry could ask GSA for the calculation workbook behind the April $1.1 billion and September $1.68 billion announcements, the comparator prices, the distribution between negotiated discounts and realized payments, the number of actual orders, and the share of benefits that depend on promotional terms. Agencies could supply anonymized billing and usage series, credits redeemed, renewal terms and migration costs. The Federal CIO Council could publish a reproducible FinOps template showing both cost and service-quality outcomes. Congress could ask OMB and GSA for a written legal explanation of the FinOps authority dispute.

The tests should be prospective as well as retrospective. If agencies using OneGov report verified lower unit prices and lower net cost for comparable workloads after promotional offers expire, the centralization case strengthens. If net spending rises because of usage growth, that alone is not a failure; the question is whether agencies obtained more public value per dollar. If a vendor raises renewal prices or imposes exit costs, the earlier discount needs to be re-evaluated over the full contract life.

The bounded verdict: Trump moved federal technology purchasing toward a central negotiator; Forst's GSA negotiated real published offers and claims large aggregate savings. GAO documented widespread cloud-cost management difficulties and recommended mandatory FinOps practices with benefits reporting; GSA rejected that recommendation on authority grounds, and GAO disputed the objection. The audit does not disprove the savings figure, but neither the public announcements nor the audit independently reconcile that figure to agency-level net cash savings. The unresolved question is institutional accountability, not an established scandal of personal enrichment.

## Source documents and independent reporting

- [GAO, Cloud Computing: Federal Government Needs to Address Procurement Challenges, GAO-26-107530 (June 23, 2026)](https://www.gao.gov/products/gao-26-107530).
- [White House, Executive Order 14240 (March 20, 2025)](https://www.whitehouse.gov/presidential-actions/2025/03/eliminating-waste-and-saving-taxpayer-dollars-by-consolidating-procurement/).
- [OMB, M-25-31, Consolidating Federal Procurement Activities (July 18, 2025)](https://www.whitehouse.gov/wp-content/uploads/2025/07/M-25-31-Consolidating-Federal-Procurement-Activities.pdf).
- [GSA, OneGov first-year savings announcement (April 29, 2026)](https://www.gsa.gov/about-gsa/newsroom/news-releases/onegov-saves-taxpayers-11-billion-in-first-year-04292026).
- [GSA, OneGov AI renewal and savings breakdown (September 10, 2026)](https://www.gsa.gov/about-gsa/newsroom/news-releases/gsa-expands-onegov-ai-offerings-with-discounted-openais-chatgpt-09102026).
- [GSA, AWS OneGov credit agreement (August 7, 2025)](https://www.gsa.gov/about-gsa/newsroom/news-releases/gsa-announces-onegov-agreement-with-aws-08072025).
- [GSA, Microsoft OneGov agreement (September 2, 2025)](https://www.gsa.gov/about-gsa/newsroom/news-releases/multibillion-dollar-gsa-onegov-agreement-with-microsoft-brings-steep-discounts-09022025).
- [GSA, Snowflake OneGov agreement (May 21, 2026)](https://www.gsa.gov/about-gsa/newsroom/news-releases/gsa-announces-onegov-agreement-with-snowflake-to-accelerate-datadriven-technolo-05212026).
- [GSA, Broadcom OneGov agreement (January 21, 2026)](https://www.gsa.gov/about-gsa/newsroom/news-releases/gsa-and-broadcom-onegov-agreement-saves-taxpayer-dollars-01212026).
- [GSA, Anthropic OneGov renewal (October 27, 2026)](https://www.gsa.gov/about-gsa/newsroom/news-releases/gsa-anthropic-reach-onegov-agreement-on-discounted-consumptionbased-pricing-10272026).
- [FedScoop, K. Sophie Will, Outdated, conflicting guidance causes cloud procurement problems (June 23, 2026)](https://fedscoop.com/outdated-conflicting-guidance-causes-cloud-procurement-problems-watchdog-says/).
- [Nextgov/FCW, Nearly 3.4M users across government can use AI through OneGov (May 15, 2026)](https://www.nextgov.com/artificial-intelligence/2026/05/nearly-34m-users-across-government-can-leverage-ai-through-onegov-gsa-official-says/413588/).
- [Nextgov/FCW, GSA unveils new token-based OneGov discount with OpenAI (September 10, 2026)](https://www.nextgov.com/acquisition/2026/09/gsa-unveils-new-token-based-onegov-discount-openai/415908/).
- [Nextgov/FCW, (US)AI becomes GSA's first direct OneGov contract with a small business (October 23, 2026)](https://www.nextgov.com/acquisition/2026/10/usai-becomes-gsas-first-direct-onegov-contract-small-business/416789/).
- [Reuters, Pentagon awards Microsoft $9.7 billion deal (May 27, 2026)](https://www.reuters.com/business/pentagon-awards-microsoft-97-billion-deal-bid-cut-costs-end-license-sprawl-2026-05-27/).
- [GSA, OneGov public agreement inventory (accessed November 1, 2026)](https://itvmo.gsa.gov/onegov/).

Research limitations: data are incomplete, and the sources do not establish realized savings or any unlawful benefit.

## Methodology

https://factrail.ai/methodology

Canonical: https://factrail.ai/articles/gsa-onegov-savings-finops-gao-cloud-cost-reporting-2026