CoreWeave, Nebius, and Cerebras closed the second quarter of 2026 with a common denominator: demand for AI infrastructure continues to grow strongly, but financing that expansion costs billions. All three companies increased their revenues, especially in inference services, while still allocating significant capital to data centers, energy, and computing capacity.
The key points of the neocloud in 20 seconds
- CoreWeave reported $2.575 billion in revenue and increased its active capacity to 1.5 GW.
- Nebius reached $582.3 million, a 454% year-over-year increase, with $575 million from its AI cloud.
- Cerebras earned $180.1 million, and its cloud business grew by 281%.
- All three companies continue expanding capacity despite losses.
- AI inference emerges as one of the main growth vectors.
The results also offer an interesting snapshot of a relatively new segment of the cloud market. Compared to the giants like AWS, Microsoft Azure, and Google Cloud, so-called neocloud providers have built much of their offerings around AI accelerators and large specialized clusters.
Their challenge is no longer just acquiring GPUs. They now need to secure electricity, build or lease data centers, finance equipment over several years, and ensure that all this capacity remains sufficiently utilized to justify the investments.
CoreWeave reaches 1.5 GW and plans up to $39 billion in investments
CoreWeave remains one of the most extreme examples of this expansion.
The company reported $2.575 billion in revenue during Q2 2026, up from $1.212 billion in the same period last year. That’s more than double in twelve months.
This growth still coexists with significant losses. CoreWeave reported operating losses of $49 million, compared to an operating profit of $19 million in Q2 2025.
Net loss reached $626 million, up from $290 million a year earlier. Adjusted EBITDA nearly doubled from $752 million to $1.51 billion.
Behind these figures is a physical infrastructure growing at an unusually fast pace—even for the data center sector.
CoreWeave allocated $9.4 billion in CapEx during the quarter and anticipates ending 2026 with capital investments between $35 billion and $39 billion.
In three months, it added nearly 500 MW of active capacity, with 300 MW added just in June. The company finished the quarter with 1.5 GW across 51 active data centers and expects to reach approximately 1.85 GW by year’s end.
Its portfolio is even larger: it has 4.2 GW of contracted power and aims to reach 8 GW by 2030.
Commercial visibility is also considerable. As of June 30, CoreWeave estimated its contracted revenue pipeline at approximately $104 billion, not including over $25 billion in new customer commitments added early in Q3.
One of the most interesting data points is inference. CoreWeave claims its managed inference platform has gone from about one million to more than $100 million in annual recurring revenue (ARR). The company expects to end 2026 with at least $250 million.
There’s also a noteworthy sign for those expecting rapid depreciation of GPUs.
CoreWeave asserts that there is still demand for older generations of NVIDIA accelerators and has recently signed a contract to use A100 GPUs until 2029. These accelerators were launched in 2020.
The immediate availability of installed and energized infrastructure may matter more than always having the latest hardware generation.
Nebius multiplies its AI cloud business sixfold
Nebius experienced the highest percentage growth among the three companies.
The group reported $582.3 million in revenue, a 454% increase compared to Q2 2025.
Almost all revenue now comes from its AI infrastructure business. Nebius AI Cloud contributed $575 million, with a year-over-year growth of 514%.
Adjusted EBITDA also changed significantly. The company shifted from a loss of $21 million in Q2 2025 to a positive $236.2 million.
Adjusted net losses decreased from $91.5 million to $33.2 million. In accounting terms, Nebius posted net losses of $190.4 million compared to a profit of $584.4 million last year, the latter influenced by gains related to revaluation of equity investments.
Expansion comes with visible costs. Operating expenses increased from $216.3 million to $758.2 million year-over-year.
Nebius is contracting and building capacity in different markets. Recent moves include deployments in Wales and Estonia, along with plans to establish a second facility at its Mäntsälä campus in Finland.
Founder and CEO Arkady Volozh told investors that Nebius closed during the quarter four deals valued at over $1 billion each, with returns of $20 to $25 million per megawatt.
The company contracted another gigawatt during the quarter and expects to end 2026 with 5 GW of contracted capacity. From 2027 onward, it plans to deploy over 1 GW of computing capacity annually.
Again, inference stands out: in-production inference loads at Nebius more than tripled during the quarter.
Cerebras takes its cloud beyond its own AI systems
Cerebras is a different case because it is not limited to providing infrastructure based on third-party accelerators. The company develops its own wafer-scale processors—chips built at wafer scale, then offered via local systems and cloud services.
GAAP revenue reached $180.1 million, a 74% year-over-year increase. Of that, cloud and other services generated $126 million, growth of 281% compared to the previous year.
Cerebras also reports about $210 million in “core revenue,” a different metric from GAAP revenue, so both figures should not be directly compared.
Growth again comes with losses.
Gross profit declined from $32.1 million to $25.56 million, while Cerebras recorded net losses of $450.4 million. Adjusted EBITDA was negative at $53.1 million, compared to a loss of $38.3 million a year earlier.
Its physical expansion also begins to match the scale of the big neoclouds.
Operational or contracted data center capacity slated to come online before the end of 2027 exceeds 600 MW. Cerebras claims to have a potential project pipeline reaching several gigawatts.
During the quarter, it also secured contracts with AI-driven software companies like Cognition and Lovable.
The new cloud race is also measured in gigawatts
The results of these three companies show how specialized AI infrastructure is shifting away from simply being a GPU rental market.
CoreWeave talks about 4.2 GW contracted. Nebius expects to finish 2026 with 5 GW. Cerebras has over 600 MW operational or contracted for delivery before the end of 2027.
These figures bring their energy needs close to those of some of the largest data center operators.
They also explain a large part of their costs. Buying accelerators is just part of the investment—the new clusters require electrical capacity, cooling, high-speed networking, storage, and buildings to house increasingly dense hardware.
Additionally, a parallel trend is emerging in the results of all three: inference is gaining weight relative to the initial phase of the industry dominated by training large models.
CoreWeave expects to surpass $250 million in managed inference ARR by year-end. Nebius reports having more than tripled its production inference loads. Cerebras has nearly quadrupled its cloud business year-over-year, leveraging inference speed as a key commercial advantage.
Q2 2026 paints a peculiar picture of the market: revenues grow rapidly, future contracts are valued in tens of billions, yet the infrastructure needed to support this growth requires similarly extraordinary investments.
For the neoclouds, the next test is demonstrating that those gigawatts can transform into utilized capacity over time. CoreWeave’s contract to keep A100s until 2029 offers a first hint: in this market, the economic lifespan of a GPU could be much longer than its stay at the cutting edge.
Frequently Asked Questions
What is an AI neocloud?
The term refers to cloud providers specializing mainly in AI infrastructure, accelerators, and large compute clusters. CoreWeave and Nebius are clear examples, while Cerebras combines its own hardware with cloud services.
How much did CoreWeave revenue generate in Q2 2026?
CoreWeave reported $2.575 billion in revenue, compared to $1.212 billion in the same quarter last year. The company posted net losses of $626 million.
How much is Nebius AI Cloud growing?
Nebius AI Cloud reached $575 million in revenue, a 514% year-over-year growth. Total group revenue was $582.3 million.
Why is a 2020 NVIDIA A100 still important?
CoreWeave says there is demand for older GPU generations because they offer ready, available capacity. The company recently signed a contract to keep A100s in service until 2029, nine years after their launch.
via: datacenterdynamics
