Datasection: The Neocloud Risk That Starts Before the GPUs Go Live
Japan's Datasection cut its FY2027 revenue forecast by 89.3%, from ¥162.19B to ¥17.29B.
The disclosures reveal two separate failures.
First, an AI infrastructure project that had been expected to contribute ¥79.86B in revenue was terminated. Only ¥5.58B in previously recognized arrangement fees remains in the revised forecast.
The company did not identify the customer or explain who initiated the termination.
Second, an overseas data center project was delayed after a customer requested a switch from NVIDIA B200 to B300 servers. Prolonged negotiations over specifications and contract terms led the facility owner to terminate the site agreement.
Datasection recognized ¥3.07B in lease termination losses and network-contract provisions—17.7% of its revised annual revenue forecast.
These are distinct events. Neither establishes that Tencent canceled an order or that new US export controls caused the forecast cut.
The deeper issue is the mismatch between GPU upgrade cycles and infrastructure commitments.
Customers can demand newer hardware before deployment. But operators may already be locked into property leases, network contracts and construction schedules.
For neocloud investors, contracted revenue alone is not enough.
The critical questions are how much revenue is contingent on deployment, who bears the cost of specification changes, and what liabilities survive when a project fails to launch.
A data center can incur real losses before its first GPU generates revenue.
#stocks