Nebius Group's AI Cloud Earnings Reveal a Rising Ad Cost Floor

Nebius Group’s FY 2025 results, released February 12, 2026, belong in a media buying cost tracker for one simple reason: the company did not just report fast AI cloud growth; it reported a market where capacity was already spoken for. FY 2025 revenue reached $529.8 million, up 479% year over year, while core AI cloud revenue reached $480.3 million, up 603%. Annualized run-rate revenue ended the year at $1.25 billion, roughly 14 times the level at the end of 2024.[1]

That growth rate is not the most useful part for Advantage+ or Performance Max planning. The useful part is the constraint signal. Nebius management reported that the company sold out every quarter in 2025, that Q1 2026 was also sold out, and that more than four customers were competing for every GPU, according to Futurum Group’s analysis of the Q1 2026 earnings call.[2] A cloud vendor can grow quickly in an immature category and still have slack. This is not that setup, at least based on the company-reported operating picture.

The earnings signal is scarcity, not just growth

For ad buyers, the mistake would be to read Nebius like another AI stock chart. The better read is more operational: when platforms keep adding automated bidding, creative generation, retrieval, ranking, prediction, and measurement layers, those systems need compute before they become dashboard defaults. Nebius is one receipt from that supply chain.

The distinction matters because ordinary software scale and AI infrastructure scale behave differently. If a platform ships a pure software feature to twice as many advertisers, the marginal cost can fall sharply. If a platform ships a feature that leans harder on model inference, training, personalization, synthetic creative testing, or ranking pipelines, more usage can mean more demand for GPUs, power, cooling, networking, and reserved capacity.

Nebius’s FY 2025 release does not prove that Meta or Google will raise auction prices. It does support a narrower and more useful conclusion: the compute layer underneath AI advertising is not clearing like a cheap, abundant software input. In Nebius’s reported market, customers were still taking capacity while utilization was full and pricing power was present.

Meta’s Nebius commitment is demand pressure, not an ad pricing formula

The Meta relationship is the part that makes this more than a niche cloud earnings story. Nebius announced on March 16, 2026, that it had signed a new AI infrastructure agreement with Meta, expanding the relationship to a total potential value of about $27 billion.[3] The structure matters: the figure includes a $12 billion dedicated capacity commitment using Nvidia Vera Rubin systems expected to start in early 2027, plus $15 billion of backstop capacity rather than guaranteed dedicated spend.[3][4]

That split is important for forecast discipline. Treating the full $27 billion as firm revenue would overstate what the disclosed agreement says. Treating the deal as irrelevant because some capacity is backstop would understate what it says about Meta’s willingness to reserve external AI infrastructure ahead of need.

Meta’s broader AI spending plan points in the same direction. CNBC reported that Meta’s AI capital expenditures were expected to be $60 billion to $65 billion in 2025 and $115 billion to $135 billion in 2026.[4] That spending includes data centers, custom chips, networking, and other infrastructure, so it should not be mapped one-for-one onto ad delivery costs. Still, it is hard to defend a 2026 or 2027 plan that assumes AI ad automation gets cheaper automatically while one of the largest ad platforms is expanding AI infrastructure commitments at that scale.

The operating takeaway is not “Meta capex goes up, CPMs go up.” Auction prices are not set by GPU invoices. Meta can absorb some infrastructure cost through margins, product mix, ad load decisions, efficiency gains, or pricing power elsewhere. Google has its own infrastructure stack and may see different unit economics. But when the platforms selling automated ad performance are also buying or reserving scarce AI capacity, the default forecast should stop assuming that scale alone pushes the cost base down.

Capacity expansion is large, but so is the backlog

Nebius is not standing still. On its Q4 2025 earnings call, the company described active power of roughly 170 megawatts and a target of 800 megawatts to 1 gigawatt by the end of 2026, a five- to sixfold expansion. It also guided to $3.0 billion to $3.4 billion of 2026 revenue against a $7 billion to $9 billion ARR target.[5]

Those numbers cut both ways. On one hand, supply is being built aggressively. On the other, the company is already talking about future run-rate demand far above current recognized revenue. Investing.com reported that Nebius had a $49 billion contracted backlog, including the Meta deal.[6] Backlog is not the same thing as revenue recognized tomorrow, and the Meta total includes optional backstop capacity. But it does show why “more data centers are coming” is not, by itself, a clean argument for falling AI compute costs.

Q1 2026 did not look like a pause after the FY 2025 surge. Futurum Group reported Q1 2026 revenue of $399 million, up 684% year over year, alongside the same sold-out capacity framing.[2] The company-reported demand signal therefore extended beyond the fiscal year being reviewed.

Why this belongs in an ad platform tracker

Most media buyers do not need a long definition of a neocloud. The relevant point is that companies like Nebius sell access to the GPU-heavy infrastructure that large AI workloads need. When that capacity is sold out, when customers compete for allocation, and when a major ad platform reserves multiyear capacity, the economics behind AI ad products look less like a free software upgrade and more like a commodity input under pressure.

This affects how forecasts should be challenged inside a marketing plan. If a 2026 budget assumes that Advantage+ or Performance Max efficiency will keep improving while platform costs fade into the background, Nebius gives finance and marketing a dated counter-signal. The counter-signal is not that CPAs must rise. It is that the infrastructure floor supporting AI ad automation is still tightening, and that the platforms may have less room to turn every efficiency gain into lower advertiser cost.

The cleanest way to use the signal is to separate three questions that often get blurred:

• Adoption: Are platforms putting more AI into campaign setup, bidding, creative, and measurement? For Meta and Google, yes, that is already the operating environment for many buyers.
• Effectiveness: Are those features improving outcomes in a specific account? That still has to be measured campaign by campaign.
• Cost floor: Is the infrastructure behind those features becoming cheap and abundant fast enough to assume lower platform economics? Nebius’s 2025 and early 2026 signals argue against making that assumption automatically.

That third question is the one Nebius helps answer. It does not tell a buyer whether a specific Advantage+ campaign will beat a manual structure next month. It does tell the buyer that the upstream compute market behind AI automation is being allocated under shortage conditions, at least in this visible slice of the market.

What not to infer

There are several ways to overread this earnings record. Nebius is not the entire AI cloud market. Its customer competition figures are management-reported, not independently audited market share data. Meta’s capex includes far more than ad serving. Google’s Performance Max economics depend partly on Google’s own chips, data centers, and auction design. A campaign’s CPM, CPC, or CPA can move because of demand, creative quality, conversion rate, auction density, tracking changes, or seasonality before infrastructure cost is visible at all.

So the useful forecast adjustment is modest but important. Do not model AI ad platform costs as if they will fall automatically with scale. Treat compute scarcity as a live constraint next to platform defaults, benchmark results, auction competition, and account-level efficiency. If infrastructure pressure later eases, the tracker can change. As of the FY 2025 Nebius report and the Q1 2026 continuation, the better assumption is that the AI ad cost floor has support underneath it.

References

1. Nebius reports fourth quarter and full year 2025 financial results, Nebius, February 12, 2026.
2. Nebius Q1 FY 2026 Earnings Show AI Cloud Capacity Scaling, Futurum Group.
3. Nebius signs new AI infrastructure agreement with Meta, Nebius, March 16, 2026.
4. Meta expands AI infrastructure deal with Nebius to $27 billion, CNBC, March 16, 2026.
5. Nebius Group N.V. (NBIS) Q4 2025 Earnings Call Transcript, Yahoo Finance.
6. Nebius Just Locked Up $49B in AI Contracts: 5 Ways to Play the Neocloud Boom, Investing.com.