AI infrastructure boom continues as Mistral lines up backers, CoreWeave extends chip outlook

Mistral, the French AI company that initially focused on building models to compete with OpenAI and Anthropic, is expanding into cloud services and plans to build one gigawatt of AI compute capacity across Europe by 2030. Sifted reported that ASML, CMA-CGM, Caisse des Dépôts, Capgemini and Amadeus have made multi-year financial commitments that will convert into what Mistral calls European Compute Units, giving customers access to future compute and related services. Mistral CEO Arthur Mensch has estimated that one gigawatt of compute would cost $50 billion to build. The company has already secured an $830 million loan for its first data center south of Paris, which will add 44 megawatts of capacity. Less than a month ago, Microsoft made a multi-billion-dollar commitment to use Mistral's compute infrastructure in Europe.

Mistral also said it is opening its platform to third-party open models, starting with GLM-5.2 from Chinese AI lab Z.ai. The company said third-party models will run on the same infrastructure, regional controls and service commitments as its own models. Customers can also choose whether inference runs in the U.S. or Europe.

CoreWeave's commentary this week offered a counterpoint to the bear case that AI hardware will become obsolete too quickly, according to CNBC. Chief executive Mike Intrator said older generations of Nvidia GPUs will have a longer useful life than anyone anticipated and will be contracted for longer terms and at higher prices. Chief financial officer Nitin Agrawal said on the earnings call that the company recently signed an A100 contract extending into 2029 at an attractive price; the chip was introduced in 2020 and used by OpenAI to train the first version of ChatGPT.

CoreWeave's results and commentary also supported the investment case for data-center builders. On Wednesday, Nvidia shares rose 3%, Corning gained 5.2%, GE Vernova rose 2.7% and Micron jumped 4.9%, while CoreWeave's stock surged almost 20%. Amazon CEO Andy Jassy has said servers have a useful life of at least five to six years, most AI capacity is contracted for at least five-year terms, and average break-even is under three years.

Cerebras, one of Nvidia's most prominent challengers, reported second-quarter core revenue of $210 million, up 50% from a year earlier and above analysts' expectation of $201 million, according to SiliconANGLE. The company raised its full-year core revenue forecast to $880 million to $890 million, but its shares fell more than 17% in extended trading. The quarterly net loss was $450.5 million, mostly from stock-compensation costs, and order backlog remained at roughly $25 billion. Chief executive Andrew Feldman said demand for the company's inference chips is 'going through the roof' and that fast inference, which serves low-latency AI tasks, is priced at a premium.

South Korea's internet company Naver said it has invested in Panthalassa, an Oregon-based startup developing wave-powered floating data centers for AI workloads, Yonhap reported. Financial details were not disclosed. Panthalassa plans a demonstration deployment this year and aims to commercialize by 2027, sending computing results to land via low-earth-orbit satellite links such as SpaceX's Starlink.

In January, cloud platform Railway said it raised $100 million in Series B funding led by TQ Ventures, according to VentureBeat. The San Francisco-based company says it has two million developers, processes over 10 million deployments a month and handles more than one trillion requests through its edge network, with only 30 employees. It abandoned Google Cloud in 2024 to build its own data centers and says its platform delivers deployments in under one second. A customer, G2X, reported seven times faster deployments and an 87% cost reduction after moving to Railway, with its monthly infrastructure bill falling from $15,000 to about $1,000.

Qianhe Yibang, a Chinese company specializing in 3D DRAM in-memory computing, said it closed a Series B round of more than 2 billion yuan, according to Chinese technology outlet Leiphone. Investors included a state-backed national fund, China Mobile's chain fund and about a dozen other institutions. Founded in early 2024 and incubated by NetEase, the company says its three-dimensional chip architecture overcomes the memory-wall bottleneck by stacking memory and logic in 3D space.

Naïve, which offers infrastructure for running businesses with AI agents, raised $28.5 million in Series A funding led by Nexus Venture Partners, according to TechCrunch. The startup says it has more than 30,000 developer customers and has grown annualized revenue to the low double-digit millions. It is building a serverless runtime and other tools to cut the cost of operating large numbers of agents.