WhiteLine Daily gathers the thinking of the Wu Shuo team to provide readers with the most valuable information and analysis of the day, and seize the changing trends in the AI era.
1. Applied Digital adds 75 MW of AI data center capacity online, increasing operating capacity to 250 MW
American AI data center developer Applied Digital announced that the second phase of Building 2 of its Polaris Forge 1 campus in North Dakota has reached Ready for Service, adding three 25 MW data halls, adding a total of 75 MW of operational IT load.
So far, all 150 MW of Building 2 have been completed, and the actual operating capacity of Polaris Forge 1 has increased from 175 MW to 250 MW. The park is fully leased and has a contracted capacity of 400 MW when fully completed, meaning that 62.5% of the contracted capacity has now entered the actual operation stage.
One-sentence analysis: Compared with continuing to sign new leases, whether the remaining 150 MW can be delivered on time will more directly determine the speed of realization of rental income.
2. Supabase raised $150 million and acquired Turso, with 70% of new databases created by AI Agents
The open source Postgres development platform Supabase announced that it has received $150 million in new financing, led by GIC, with participation from Alphabet's independent growth funds CapitalG, IronArc and SquarePeg; the company also announced the acquisition of the database platform Turso to further expand the database infrastructure for AI Agents.
Supabase is currently adding more than 1 million new users and 4 million databases per month, with approximately 70% of new databases already being created by Agent or AI-driven tools. The company said the number of databases had grown 600% year-over-year in June this year. Supabase currently has more than 13 million developer users. After acquiring Turso, it will focus on supporting a large number of agents to create independent databases on demand.
One-sentence analysis: When 70% of new databases have been created by Agent, the increment brought by AI Coding no longer stops at code generation, but begins to directly change the creation frequency, cost and infrastructure structure of the database.
3. JERA, Dell, and RHAELM plan to invest more than US$15 billion to build a 400 MW AI infrastructure project in Japan
Japan’s largest power company JERA and Dell Technologies signed a memorandum of understanding with British AI infrastructure developer RHAELM to establish a nationwide AI infrastructure development framework in Japan. The first project, located near the JERA Chiba Thermal Power Station, is planned to have a total capital investment of more than $15 billion and provide up to 400 MW of power capacity.
The project will use a behind-the-meter model to directly connect JERA's existing power generation assets to shorten traditional grid connection waiting times; RHAELM is responsible for project development, operations and financing, Dell provides standardized rack-level AI infrastructure, and Apollo Global Management plans to participate as a strategic investment and financing partner. The project is expected to operate in phases starting in 2028 and reach full capacity in 2029.
One-sentence analysis: The 400 MW project is built directly next to existing power generation assets. The core value is not to simply increase the data center area, but to reduce the construction bottleneck of network connection time by co-locating power supply and computing power.
4. Volantis raised US$88 million and plans to use optical interconnects to connect up to 220 memory chips
San Francisco AI chip startup Volantis has raised $88 million in financing to develop optical interconnect technology that connects computing chips and memory chips through lasers. Currently, high-end GPUs such as Nvidia and AMD mainly rely on HBM to provide high-speed storage, but the distance of traditional electrical interconnections limits the number of memory chips that can be connected around a single GPU.
Volantis plans to use VCSEL lasers to transmit data, which the company says will allow up to 220 memory chips to be connected around a single GPU, compared with currently around eight for top-end GPUs. The company plans to launch its first chip next year and hopes to make development and manufacturing easier by leveraging its already mature VCSEL supply chain.
One-sentence analysis: increasing from about 8 to a maximum of 220 is equivalent to an expansion of about 27 times. Volantis is trying to shift the memory expansion bottleneck of AI chips from electrical interconnection distance to optical interconnection, but the real verification point is still next year’s product performance and mass production.
Today’s main line:
Supabase is facing the problem of infrastructure expansion after Agent creates databases in batches. Applied Digital has put 250 MW of capacity into operation, JERA is planning a 400 MW direct-connect power supply project, and Volantis is trying to break through the connection distance between GPU and memory. As AI applications increase, database scaling, available power, and memory interconnect each become next-level issues that need to be addressed.