Amsterdam, May 2, 2026, 15:05 CEST
- Nebius is picking up Eigen AI in a deal valued near $643 million, aiming to bring in technology that can cut the cost and speed up the deployment of AI models.
- The agreement comes just days ahead of Nebius’ first-quarter earnings release, set for May 13.
- Nebius jumped 11.8% to $154.49 in its latest trade, following word of the acquisition.
Nebius Group N.V. is set to buy Eigen AI, sealing a cash-and-stock deal worth roughly $643 million. With this move, the Amsterdam company aims to bolster its role in the software powering artificial intelligence models post-training. Nebius said the acquisition strengthens its managed inference segment—the slice of AI cloud computing that handles real-time user queries.
Inference — the part where a trained model spits out answers, writes code, or helps run an app — is now where costs are piling up in AI infrastructure. Deloitte projects inference will drive around two-thirds of AI compute demand this year, a jump from nearly half in 2025.
This deal lands just ahead of a key moment for investors. Nebius, for its part, has set May 13 for its first-quarter report, dropping results before the opening bell. An investor call is scheduled for 8 a.m. Eastern, promising updates on demand, spending, and margins after a streak of hefty AI infrastructure wins.
Bloomberg puts the deal at roughly $98 million in cash, plus 3.8 million Nebius shares, all valued using Nebius’s 30-day weighted average price. Nebius expects to wrap up the sale in the next few weeks, pending usual approvals like antitrust review.
Eigen AI’s suite for inference and post-training optimization is set to plug directly into Nebius Token Factory, the managed platform Nebius uses to launch and tune open-source AI models. “Capacity-scarcity world,” is how Roman Chernin, Nebius co-founder and chief business officer, sized up the market. Over at Eigen, co-founder and CEO Ryan Hanrui Wang said both teams plan to “push the boundaries of inference performance.” Nebius
Post-training refers to anything that happens once a model’s initial build is complete—think fine-tuning for accuracy, better speed, or trying to cut costs. Nebius noted that Eigen’s staff, which draws on researchers from MIT’s HAN Lab, are also on deck to build out the company’s engineering and R&D footprint in the Bay Area.
Nebius is pushing deeper into competition with AI-focused cloud firms like CoreWeave, which offer clients access to GPUs—those chips built for intensive AI workloads. Back in March, Reuters noted that U.S. tech companies have started securing limited GPU and power resources from these so-called “neocloud” players, preferring not to depend solely on their own data centers. Reuters
Nebius leans heavily on its biggest clients in its pitch to investors. In March, Meta signed on to purchase $12 billion worth of AI computing power from Nebius through 2027—plus there’s an option to tack on another $15 billion over the next five years. Nvidia, for its part, picked up an 8.3% stake, paying $2 billion, according to Reuters. Nebius had deals in place with Meta and Microsoft before these big moves.
The model comes with a steep price tag. Nebius posted a net loss of $250 million for the fourth quarter, against $228 million in revenue, and had projected its yearly revenue run-rate would climb to somewhere between $7 billion and $9 billion by the end of 2026, according to Reuters. Investors will be watching the May 13 report for signs the pipeline is actually converting contracts into revenue.
Deal risk isn’t off the table. Nebius cautioned that any forward-looking statements tied to the Eigen AI acquisition hinge on actually closing the deal, pulling the team together, keeping current clients onboard, and getting enough growth capital. If there’s a holdup on antitrust approval or if Eigen’s tech doesn’t deliver the expected cost savings, that could eat into the projected upside.
The Nebius narrative takes a turn with this acquisition. Yes, capacity expansion remains central, but the hefty price tag covers software—software that could wring extra output from every chip. That efficiency is what investors will be watching next.
