Sakura Internet is leaning further into AI infrastructure as Japan’s demand for domestic GPU capacity keeps rising. The hard question is whether regional cloud providers can turn expensive chips and data centers into durable profits.
Sakura Internet is becoming one of Japan’s clearest tests of whether domestic AI infrastructure can compete in a market still dominated by American hyperscalers, scarce chips and rising power demands.
According to a Bloomberg Technology report published Monday, Sakura is considering additional capital spending to meet demand for AI data centers. The timing matters because the company has already spent heavily. Its latest filings show net cash used in investing activities rose to 24.6 billion yen in the fiscal year ended March 2026, up 196.1% from the previous year, mainly because of equipment tied to generative AI services.
This is the new shape of the AI race. It is not only Microsoft, Amazon, Google and Meta building huge compute estates in the United States. Regional providers are trying to win enterprise and public-sector customers that want powerful AI systems without sending every sensitive workload through a foreign cloud platform.
Sakura has a real opening in that market. The company was selected as a provider for Japan’s Government Cloud services for fiscal years 2023 and 2026, and it has been building GPU infrastructure into a core business line. In its April earnings release, Sakura said revenue from GPU infrastructure services reached 8.14 billion yen in the year ended March 2026, up 20.3% from a year earlier.
The company is also gaining customers, not just investor attention. In April, Sakura announced a roughly 3.8 billion yen order from a national institution for generative AI, covering its SAKURAONE Managed HPC Cluster H100 and H200 systems. The service period is planned to run through March 2027. That is important because AI infrastructure only becomes a business when expensive GPUs are committed to workloads that can pay for them.
Sakura’s pitch is practical. Japanese companies building language models, robotics systems, medical AI tools or government applications often need local control, predictable compliance and support in their own market. A domestic provider cannot match the global breadth of AWS or Azure, but it can offer something different: proximity, trust and a national infrastructure story that procurement teams understand.
That explains the Microsoft connection as well. Microsoft said in April that it would work with Sakura and SoftBank so Azure customers in Japan can access GPU-based AI compute from domestic providers while keeping data in the country. In plain terms, Microsoft gets a stronger local compute option, while Sakura gets a route into customers that already live inside the Azure ecosystem.
The spending is still the risk
The opportunity is easy to see. The economics are harder. Sakura’s revenue rose 12.4% to 35.3 billion yen in the fiscal year ended March 2026, but the company posted an operating loss of 403.7 million yen after reporting operating profit of 4.15 billion yen the year before. Management pointed to human resource investment, higher GPU-related depreciation, server maintenance costs, data center rent and other expenses.
That is the trade every AI infrastructure company is making. Buy early and you can capture demand. Buy too much, or buy the wrong generation of hardware, and depreciation starts doing the explaining. GPUs are not like office buildings. They age quickly, power requirements change, and customers will not pay premium rates forever for yesterday’s compute.
Sakura’s own forecast shows both confidence and caution. For the year ending March 2027, it expects net sales of 45 billion yen, up 27.5%, and operating profit of 1.5 billion yen. But the company also said its top priority this year is stable operation of existing GPU resources, with additional investments to be considered based on market conditions. That is a careful sentence, and it should be. The AI buildout rewards speed, but it punishes waste.
Nvidia’s recent Japan update shows how broad the domestic buildout has become. SoftBank, GMO Internet Group, Highreso, KDDI, Rutilea and Sakura are all adding Nvidia-based AI infrastructure, supported in part by Japan’s Ministry of Economy, Trade and Industry program to supply critical compute resources. Nvidia said Sakura plans to expand its Koukaryoku cloud services from 2,000 to nearly 4,000 Hopper GPUs, install Blackwell infrastructure at its Ishikari data center and eventually provide about 10,800 GPUs as part of its high-performance AI computing offering.
That puts Sakura in a stronger position, but not an easy one. The company is competing with Japanese telecom groups, global cloud providers and specialist GPU clouds while navigating power capacity, cooling demands and chip availability. AI customers want capacity immediately. Infrastructure companies have to finance assets that may take years to earn back.
The most interesting part of Sakura’s move is not simply that it may spend more. It is that Japan’s AI market is giving domestic infrastructure providers a reason to matter again. If enterprises decide that sovereignty, language specialization and local partnerships are worth paying for, Sakura can become more than a niche cloud operator. If price and scale dominate, the hyperscalers will keep pulling the center of gravity back toward themselves.
For investors and founders watching the AI infrastructure cycle, Sakura is a useful signal. The next phase of AI will not be won only by whoever owns the largest global data center fleet. It will also be shaped by who can place compute where customers, regulators and national strategies actually need it.
Also read: Japan's AI rally is changing how retail traders reach the market • Microsoft's Claude pullback shows AI coding still has a budget problem • Japan’s Nikkei 225 breaks 64,000 as the AI trade spreads
