Microsoft Strikes $19B Nebius Deal: Nvidia Tightens Grip on AI GPUs

Microsoft Signs Up to $19B AI Infrastructure Deal With Nebius - Nvidia Scores Again

Microsoft agreed to buy GPU-based compute from Nebius worth $17.4B to $19.4B over five years. Deliveries start in late 2025 with an option to lift capacity during the term.

This is one of the largest AI infrastructure commitments to date. It signals where the next wave of value will accrue: GPU supply, specialized "neoclouds," and the platforms that convert compute into products.

The Deal, In Short

Microsoft secures long-term access to high-performance GPU compute from Nebius to support AI training and inference. The contract runs at least five years, with expansion rights that push the value near $19.4B.

Investors took notice. Nebius shares jumped over 40% on the announcement - a fast repricing of its role in the AI supply chain.

Who Is Nebius?

Amsterdam-based Nebius spun out of Yandex's international operations and brands itself as a "neocloud" - infrastructure built around GPU-heavy AI workloads. This agreement vaults Nebius into the first tier of AI compute suppliers.

To meet demand, Nebius plans to raise roughly $3B using converts and equity. The funds will go to new data centers, network, and reliability so it can fulfill Microsoft's schedule.

Nvidia: The Quiet Winner

Nebius depends heavily on Nvidia GPUs, the default for training and running large models. With Nvidia already holding well over 80% share in AI accelerators, the Nebius buildout likely translates into billions in additional chip demand.

Nvidia also owns a stake in Nebius, creating a double benefit: chip sales plus equity upside. AMD and Intel continue to push alternatives, but the software ecosystem and performance profile keep Nvidia out front.

Why Microsoft Is Partnering Instead of Building Everything

Time and capital. Partnering lets Microsoft add AI capacity faster without taking on all the upfront spend for land, grid connections, and facilities. Nebius carries a big portion of the financing and build risk.

Microsoft locks in the GPUs needed to scale services like Azure OpenAI while keeping balance sheet flexibility for product, software, and go-to-market. For AI demand that is compounding, speed-to-capacity beats owning every asset.

What This Means for You

• IT and developers: Expect better access to GPU instances and shorter queue times as new capacity comes online. Plan for multi-provider strategies that mix hyperscalers with "neoclouds" to hit cost, latency, and data residency targets.
• Finance and investors: This is capex-to-opex substitution at hyperscale. Watch beneficiaries across the stack: accelerators (NVDA), memory, networking, power and cooling vendors, along with new infrastructure financiers.
• Product leaders: Delivery begins late 2025. Roadmaps that depend on large-scale training should sequence around that timeline and consider interim efficiency tactics (smaller models, distillation, inference optimizations).

AI Compute Boom: The Numbers

Global spend on AI chips and cloud capacity could exceed $200B by 2030, with some forecasts pointing near $400B. GPU demand is projected to grow 25-30% annually as generative AI moves into production across industries.

Specialized "neocloud" providers could capture up to 15% of AI infrastructure contracts by 2030. Nvidia is set to remain the dominant accelerator vendor as this demand scales.

The Energy and Carbon Question

Training a single large language model can emit 500+ metric tons of CO₂, depending on model size, grid mix, and training efficiency. Data centers that run AI workloads account for roughly 1.5% of global electricity use today and could reach ~4% by 2030.

• Secure long-term renewable energy contracts and improve measurement of hourly matching.
• Push for higher data center efficiency: advanced cooling, higher rack densities, and better utilization.
• Adopt model efficiency: pruning, quantization, sparsity, and distillation to lower training and inference cost.
• Select regions with cleaner grids and publish lifecycle footprints for major training runs.

For a solid overview of data center energy trends, see the International Energy Agency's report on data centers and networks here. Microsoft's sustainability commitments are outlined here.

Why Everyone Wins Here

Microsoft gets guaranteed GPU capacity to keep pace in AI. Nebius moves into the top tier of infrastructure providers with multi-year revenue visibility. Nvidia adds another multi-billion-dollar demand driver on top of its existing lead.

Expect more deals of this size as enterprises turn pilots into production. Watch three things: delivery timelines, energy availability, and cost-per-token improvements - that trio will decide who compounds fastest.

Upskill for What's Next

If you lead teams in engineering, data, or finance and want structured ways to build AI capability, explore these resources:

• AI courses by job role to align skills with delivery plans.
• Popular AI certifications to standardize training across your org.