Datacenter operators shift to power wars

Data Center Knowledge reported on May 15 that recent earnings from CoreWeave and Nebius showed investors and customers focusing less on GPU scarcity and more on electricity, cooling, networking and delivery speed. The publication said the market for “neocloud” providers is being judged increasingly on whether operators can secure long-term power and bring dense AI capacity online at industrial scale. (datacenterknowledge.com) CoreWeave and Nebius both paired rapid revenue growth with larger infrastructure commitments, according to company disclosures and the report. The result is that power contracts and grid access are becoming central operating metrics for AI datacenter companies. ### Why are power contracts getting as much attention as GPUs? (marketbeat.com) CoreWeave said on May 7 that first-quarter revenue rose 112% year over year to $2.08 billion, while revenue backlog climbed to $99.4 billion. The company also said it had surpassed 1 gigawatt of active power and expanded contracted power capacity by more than 400 megawatts to more than 3.5 GW. CEO Michael Intrator said the company believed it was “well on our way to more than 8 GW by 2030.” Nebius said on May 13 that its AI business annualized run-rate revenue reached about $1.9 billion at the end of March and that it had contracted more than 3.5 GW of power, up from more than 2 GW disclosed in the prior quarter. CEO Arkady Volozh said the company was targeting at least 4 GW of contracted power this year. (datacenterknowledge.com) Those figures put electricity access alongside revenue and GPU supply as a disclosed measure of expansion. ### What are CoreWeave and Nebius telling investors in earnings? CoreWeave told investors that capital expenditures reached $6.8 billion in the quarter and that it raised the low end of its 2026 capital spending range to as much as $35 billion. Data Center Knowledge said the spending reflected a business that is being built around power, networking, cooling and deployment scale rather than simple GPU procurement. (datacenterknowledge.com) Arkady Volozh described Nebius on its earnings call as building an “AI-native hyperscaler,” according to a summary of the call. MarketBeat said he tied that effort to four areas: capacity and scale, product functionality, customer demand and capital. The company’s first-quarter revenue rose 684% year over year to $399 million, according to earnings summaries. (fool.com) ### Where does operating credibility show up in this market? Data Center Knowledge said the competition now centers on execution: securing power capacity, building high-speed networking fabrics, deploying advanced cooling systems and bringing infrastructure online fast enough for hyperscalers and enterprise customers. That framing puts more weight on whether operators can convert financing and land into energized, usable capacity. (datacenterknowledge.com) CNBC reported on May 7 that CoreWeave ended the quarter with about 3.5 GW of total contracted power and that Intrator told analysts, “We have reached hyperscale.” The company also said 10 clients were each committed to spending at least $1 billion on its products, adding another test of whether infrastructure can be delivered on time. (marketbeat.com) ### How does custom silicon fit into the power fight? Google said in a July 31, 2024 blog post about its Axion processor that compute power in the cloud depends on a range of chips including CPUs, GPUs and its own TPUs, and that shaping hardware and software systems helps the company “shape our own destiny.” The company said its custom silicon program is tied to performance, flexibility and efficiency across cloud services. (datacenterknowledge.com) Cisco says its Silicon One products are designed for hyperscaler AI clusters and datacenter networking, while MediaTek said in an April 23, 2026 post that custom ASIC platforms let designers balance workload density, efficiency and scale across datacenter systems. Those descriptions link chip design to power delivery, network architecture and system efficiency, not only to raw compute. (cnbc.com) ### What comes next for the companies in this race? Nebius said on May 13 that it was targeting at least 4 GW of contracted power in 2026, while CoreWeave said on May 7 that it believed it was positioned for more than 8 GW by 2030. Data Center Knowledge’s May 15 report said future earnings are likely to keep highlighting power, cooling and deployment milestones alongside revenue, giving investors a regular set of checkpoints for whether AI datacenter operators can turn contracts into energized capacity. (cloud.google.com) (marketbeat.com) (cisco.com)