OpenAI and NATO Cyber Defense: Up to 300 Billion Tokens a Month, No Public NCIA Price or Contract Terms

By
CTOL Staff Reporter
1 min read

OpenAI has a place in NATO's cyber work. Turning that foothold into a recurring government purchase is a different business.

At Crossed Swords in Tallinn, OpenAI's Daybreak entered the exercise's technical environment for the first time. The exercise also used AI on premises, while OpenAI supported an AI agent training stream. On 17 November 2026, OpenAI described its work with NATO's Communications and Information Agency, NCIA, and said NATO had used up to 300 billion raw input and output tokens a month. The figure measures the scale of reported usage, without showing its price or contract status. The token counter is impressive, although government buyers pay invoices rather than applaud counters.

As of 18 November 2026, we would pass on assigning recurring OpenAI revenue to that NATO usage. We expect frontier AI cyber defence to become a repeat government purchasing category, but we would not bet on NCIA publishing a priced tender or award this soon. The distinction is Meter Before Order: suppliers place models inside government workflows so buyers can test cost, performance and alternatives before committing to repeat purchases. A priced NCIA award naming OpenAI would show a commercial commitment; a tender alone would show buying intent. The vendor earns a foothold while the purchaser controls the future contract. The category attracts us; assumed revenue does not.

The potential bill is large enough to move an investor's spreadsheet. OpenAI lists GPT-6 Astra at $10 per million standard input tokens and $50 per million output tokens. At NATO's stated ceiling of 300 billion monthly tokens, a scenario with 90 percent input and 10 percent output means 300,000 million-token units at a blended $14, or $4.2 million a month and $50.4 million a year. Apply OneGov's hypothetical 50 percent discount and the scenario falls to $2.1 million monthly, $25.2 million annually. The model NATO uses, token mix and commercial terms are unspecified; these are price scenarios, rather than receipts. Investors can value an opportunity; they cannot record this scenario as sales.

OpenAI's subsidy supports adoption across the Daybreak programme. On 3 September 2026, it committed $1 billion in subsidised Daybreak access, training, technical support and partnerships, targeting consumption within six months. The same announcement counted 2,000 approved organisations and workspaces. Daybreak checks applicants and intended uses before approving model access. Permissions come first; widespread trials give OpenAI a foothold. The subsidy covers the programme overall, not a demonstrated NCIA subsidy. OpenAI earns familiarity and the opportunity to bid, while the buyer preserves its choice of supplier.

The detection results make a purchase plausible. In a joint ENISA and CERT-EU assessment, frontier models found 220 potential code issues; owners assessed 88, confirming 95 percent of those as true positives. That is useful evidence of detection quality among reviewed findings. Yet 220 minus 88 leaves 132 findings unassessed at publication. ENISA put the indicative financial cost of its scans at EUR 6,648.90, or EUR 132.27 per 10,000 lines of code, under free model access. A procurement officer must price the review and response, as well as the scan.

Models can generate candidate vulnerabilities cheaply; security specialists must decide which ones matter and who fixes them. ENISA warned that the volume can impose a significant review burden on code owners. We expect recurring spending to gravitate toward triage, remediation and the evidence that a finding was resolved. The contractor responsible for that workflow can compare model suppliers and buy inference as an input. OpenAI's access programme offers a head start in distribution; it does not automatically make OpenAI the company that keeps the service margin. The model bill is only one part of the buyer's eventual expense.

NCIA has almost described Meter Before Order in its own words. Its general manager, Dr Dylan Browne, said: "As we scale adoption, we will build a deeper understanding of token consumption, usage patterns and the underlying economics of AI models". The purpose, he added, was to "optimise costs, evaluate competing solutions and make informed procurement decisions". ENISA's assessment included OpenAI models and Anthropic Mythos 5, a reminder that buyers have frontier alternatives to assess. NCIA can get better at using OpenAI while also getting better at negotiating against it. The same tokens that demonstrate adoption can strengthen the buyer's hand on price.

GSA's $1 trial challenges our position because it led to published metered prices. In August 2025, the US General Services Administration offered OpenAI's enterprise subscription to participating federal agencies for $1 for one year. On 10 September 2026, GSA announced a 27-month consumption-based offer with a 50 percent token discount, expected to take effect on 1 October 2026. A nominal access offer had become a published metered rate without guaranteed usage or spending. Coupled with 2,000 approved Daybreak organisations and NATO's monthly volume ceiling, that is a credible customer acquisition mechanism. Governments can start with subsidised usage and move quickly to commercial pricing.

The buyer kept one valuable right: to spend nothing. GSA's published terms specify "no platform-access fee, no minimum orders, and no spend commitment." At a monthly horizon, that makes GSA's guaranteed minimum $0, against the $2.1 million possible monthly billing in our hypothetical discounted NATO scenario. These concern different buyers: the contrast is between a commitment and an opportunity, not a claim about NCIA's payments. GSA published an offer and a planned start date; NCIA has described evaluating its economics. A priced NCIA tender or award is the observable step that would narrow this distance.

NATO has its own precedent for moving fast. NCIA and Palantir finalised the acquisition of Maven Smart System NATO on 25 March 2025, six months after the requirement was outlined. That was a real acquisition of a named system, making fast conversion entirely plausible for a buyer capable of defining its requirements. Daybreak's work has instead involved model access and support for Crossed Swords, where the exercise organisers confirmed the first use of Daybreak alongside AI running on premises. We would watch for a procurement notice rather than extrapolate revenue from those footholds.

The six-month consumption target attached to the September 2026 Daybreak subsidy runs toward early March 2027. By then, the vendor will have had time to demonstrate utility and buyers to learn the economics. What follows can be a contest between a model provider and a company willing to guarantee the full security workflow, with the agency choosing how to allocate spending. That is why we expect a procurement category to emerge without assuming OpenAI captures its entire value. A price schedule shows possible terms, a tender shows buying intent, an award shows a commitment, and usage alone shows none of these.

We would pass on valuing NATO's up to 300 billion monthly tokens as recurring OpenAI revenue, while expecting frontier AI cyber services to attract government purchasing once buyers price the work. We would be wrong if by 31 March 2027 a priced NCIA contract award naming OpenAI, or NATO disclosure of recurring paid terms for OpenAI's Daybreak models, appears on the NCIA procurement pages or in a NATO press release.

You May Also Like

This article is submitted by our user under the News Submission Rules and Guidelines. The cover photo is computer generated art for illustrative purposes only; not indicative of factual content. If you believe this article infringes upon copyright rights, please do not hesitate to report it by sending an email to us. Your vigilance and cooperation are invaluable in helping us maintain a respectful and legally compliant community.

Subscribe to our Newsletter

Get the latest in enterprise business and tech with exclusive peeks at our new offerings

We use cookies on our website to enable certain functions, to provide more relevant information to you and to optimize your experience on our website. Further information can be found in our Privacy Policy and our Terms of Service . Mandatory information can be found in the legal notice