Tanaka Says He Needs
100,000 NVIDIA GPUs. Microsoft, Lutnick, and a $550B Fund Just Tried to Get Them For Him.
On October 30, 2024, Bloomberg’s Tokyo technology desk published an interview with Kunihiro Tanaka, founder and CEO of Sakura Internet (TYO: 3778), the Osaka-headquartered cloud provider the Japanese government has quietly designated as the domestic champion for sovereign AI compute. Tanaka’s headline number was the kind of figure that, in a different country, would have been read as a complaint. In Japan it was read as a roadmap: Sakura needs 50,000 to 100,000 NVIDIA GPUs to clear the AI-compute demand backlog it sees today.
On April 3, 2026, Microsoft made the corporate answer official. The company announced a $10,000,000,000 four-year Japan AI-infrastructure commitment, partnering specifically with Sakura Internet and SoftBank, with a parallel pledge to train one millionJapanese engineers. It is the largest country-level investment in Microsoft’s history. Sakura’s stock closed +20.2% on the day. On February 10, 2026, Bloomberg reported that U.S. Commerce Secretary Howard Lutnick (R) and Japan econ minister Ryosei Akazawa had begun discussions on first-project allocations for the $550,000,000,000 U.S.-Japan Strategic Investment Fund the Trump administration negotiated as part of its 2025 trade architecture.
That is the financial spine of this story. The editorial spine is one step removed: Japan’s scramble to build sovereign AI compute, with Sakura Internet as the operational tip of the spear, is downstream of U.S. export-control policy choices — the AI Diffusion Rule, the rule’s Trump-Lutnick rollback, and the bilateral deal architecture that replaced it. Allies are now in a position where they must build their own GPU farms, on their own soil, with their own subsidies, partnering with U.S. hyperscalers under U.S. terms. PM Sanae Takaichi (LDP) put ¥1.23 trillion (~$7.9 billion) into Japan’s FY26 AI / semiconductor budget — roughly four times the prior-year envelope — to make that math work.
- 50K–100KNVIDIA GPUsTanaka’s stated demand-backlog need to comfortably handle current Sakura AI-cloud customer requests · Bloomberg interview, Oct 30, 2024
- 10,000GPUs / yearTanaka’s stated annual NVIDIA GPU procurement target — the pace Sakura wants to buy at, supply allowing
- 10,800by March 2028Sakura GPU deployment milestone disclosed in IR materials — the three-year ramp Tanaka is pacing against the backlog estimate
- $10BMicrosoft, 4 yrsApr 3, 2026 Microsoft + Sakura + SoftBank Japan AI infrastructure deal · train 1M Japanese engineers · largest country-investment in Microsoft history
- +20.2%3778:JP closeSakura Internet stock close on Apr 3, 2026 — the market’s read on the Microsoft deal · CNBC
- $550BUS-Japan fundStrategic Investment Fund negotiated under Commerce Sec. Howard Lutnick (R) and Japan’s Ryosei Akazawa — first-project allocations under discussion since Feb 10, 2026
- ¥1.23TTakaichi FY26Japan AI / semiconductor budget commitment by PM Sanae Takaichi (LDP) — roughly 4× the prior-year envelope · ~$7.9 billion
- −93%Sakura net profitFY26 year-over-year net-profit drop at Sakura Internet — revenue grew +12% to a record ¥35.3B but the capex-front-loaded buildout phase pulled the bottom line down sharply · Bloomberg 3778:JP
Bloomberg’s Takashi Mochizuki, Tokyo technology correspondent, sat with Tanaka in late October 2024 at Sakura’s Osaka headquarters. The interview, published on October 30, was the first time a Japanese cloud-provider CEO had put a specific, large GPU-demand figure on the public record under his own name. The number Tanaka gave was not a forecast hedged in ranges and footnotes — it was a backlog estimate stated as fact.
“We need some 50,000 to 100,000 Nvidia GPUs to comfortably handle the demand we see today.”
Kunihiro Tanaka, founder and CEO, Sakura Internet · to Bloomberg · Oct 30, 2024
That is roughly an order of magnitude beyond what Sakura was running at the time. Tanaka’s second number, in the same interview, was the procurement-pace target: Sakura wants to buy 10,000 NVIDIA GPUs per year, supply allowing. The constraint, in Tanaka’s framing, is not capital — it is allocation. Sakura’s position in NVIDIA’s global customer queue is determined by NVIDIA’s own production capacity and by the U.S. export-control regime that determines which countries’ cloud providers can receive which class of chip.
“We would need a million GPUs 10 years from now, if demand rises at the current pace.”
Kunihiro Tanaka · same Bloomberg interview · long-horizon framing
The third number Tanaka has anchored Sakura’s GPU buildout against — this one disclosed in subsequent IR materials — is a March 2028 deployment milestone of 10,800 GPUs. That figure paces the three-year ramp against the publicly-stated demand backlog and against the H100 deployment Sakura completed in late 2024 (the dedicated eight-GPU NVIDIA H100 cloud SKU Telecompaper covered) and the B200 / Blackwell-class plan ITBusinessToday reported on the product-roadmap side.
On April 3, 2026, Microsoft drafted its answer to the question the Tanaka interview had implicitly posed: who pays to close the gap between Japan’s AI compute demand and the domestic GPU footprint that can serve it. Bloomberg broke the $10 billion figure; Microsoft’s own Source Asia press release confirmed it within hours. The four-year commitment is structured around three lines: AI infrastructure (datacenter partnerships with Sakura and SoftBank), cybersecurity (Microsoft Defender / Sentinel deployments), and workforce (the one-million engineer training pledge).
CEO Satya Nadellaframed the deal in the language Microsoft has used elsewhere for sovereign-AI partner geographies — Germany, France, the United Arab Emirates — but with a key difference: the Japan deal is the first to name a specific domestic-champion cloud provider (Sakura) as a named partner. CNBC reported Sakura’s stock closed +20.2% on the announcement. That is not a speculative-froth move; that is the market repricing Sakura from a mid-cap Japanese cloud provider into the local execution layer for a $10 billion U.S. hyperscaler commitment.
Microsoft partners with SoftBank and Sakura Internet to build AI data infrastructure in Japan, investing $10 billion over four years and training 1M engineers (@6d6f636869 / Bloomberg)
Microsoft bets big on Japan AI with SoftBank and Sakura Internet: a four-year $10 billion push to build data infrastructure and train 1 million engineers. Read the Bloomberg take and implications for Asia's AI race.
The Tanaka GPU need on one side, and the financial commitments stacking on the other side, look like this when laid out line-by-line. The two policy layers — U.S. (the Strategic Investment Fund and Lutnick’s rewrite of the export-control regime) and Japanese (Takaichi’s FY26 budget and METI’s subsidy program) — sit on top of the corporate capex layer (Microsoft) and the recipient capex layer (Sakura’s own ¥13 billion three-year plan).
What it is: Japanese legislation passed in 2022 authorizing METI to provide direct subsidies to domestic firms in designated economic-security-critical sectors. Semiconductors and cloud infrastructure are explicit targets.
Why Sakura qualifies: Sakura Internet is the domestic alternative the Japanese government wants to keep functional alongside U.S. hyperscalers (AWS Japan, Azure Japan, GCP Japan), so that Japan retains a sovereign-compute fallback in any scenario where allied access narrows. The METI ¥50.1B subsidy is the operational expression of that policy.
The Takaichi escalation:The ¥1.23 trillion FY26 envelope is the next tier — multiplying the subsidy pool roughly fourfold so that future allocations of this kind can scale with Tanaka’s GPU procurement curve rather than against it.
The U.S. policy layer is where the editorial frame sharpens. Commerce Secretary Howard Lutnick (R), in his first-year-accomplishments release in January 2026, took explicit credit for replacing the Biden-era AI Diffusion Rule — the export framework that restricted advanced-AI chip flows to most non-U.S. countries — with a bilateral-deal architecture. The Strategic Investment Fund is the most visible piece of that architecture for the Japan relationship. The $550 billion headline number is the bilateral commitment envelope Japan agreed to fund; the first-project allocations Akazawa and Lutnick have been negotiating since February 2026 will determine where that money actually lands.
The Bloomberg coverage of the Akazawa-Lutnick channel is the cleanest public read on how the fund is structured. The money is Japanese-sourced but routed into U.S.-aligned strategic priorities: American semiconductors, American steel, American energy, American AI infrastructure built on U.S.-allied soil. The editorial point worth surfacing — without polemic — is that this deal architecture is the operational consequence of the export-control rewrite. Japan’s sovereign AI buildout, the Sakura GPU procurement, the Microsoft $10 billion commitment, and the Takaichi ¥1.23 trillion budget commitment all sit downstream of choices Lutnick and the Trump administration made about how advanced-compute access would be distributed.
“We're not going to run low on power. We're going to bring in semi-conductors, and we're going to make sure America wins.”
Howard Lutnick · U.S. Secretary of Commerce · Commerce Department press posture · February 2026
The $550 billion U.S.-Japan Strategic Investment Fund is the largest economic alliance in history. American compute, American steel, American semiconductors — Japan will pay, build, and source from U.S. companies.
Paraphrased commentary · not a verbatim post
Composite of Trump's stated position on the US-Japan fund post-February 2026. Not a verbatim Truth Social post; reflects the framing his administration has consistently used in public for this deal.
We're not going to run low on power. We're going to bring in semiconductors, and we're going to make sure America wins. The Diffusion Rule was burdensome, overreaching, and disastrous — we replaced it with deals.
Paraphrased commentary · not a verbatim post
Adapted from Lutnick's January 2026 Commerce Department press appearances and the first-year-accomplishments release. Not a verbatim Truth Social post.
On the Japanese side, the political fingerprint is Sanae Takaichi (LDP), who took office as Prime Minister with an explicit economic-security platform. The FY26 AI and semiconductor budget she committed — ¥1.23 trillion, roughly $7.9 billion at current rates — is the first PM-led budget cycle to treat sovereign AI compute as a national-security line item on par with traditional defense procurement.
The line of authority runs Takaichi → METI → Economic Security Promotion Act subsidy programs → Sakura Internet (and parallel recipients including Rapidus, the Tokyo Electron supply chain, and the SoftBank-Microsoft datacenter partnerships). METI’s ¥50.1 billion (~$324M) Sakura subsidy under the Act is the largest single line-item disclosed to Sakura to date. The funding pairs with Sakura’s own ¥13 billion three-year capex plan, and is what makes Tanaka’s 10,000-GPU-per-year procurement target operationally plausible rather than rhetorical.
The pattern: Sakura FY26 revenue rose +12% to a record ¥35.3 billion, and FY26 net profit fell roughly 93% year-over-year. To a casual reader, those two numbers look contradictory. They are not.
What’s happening: The capex bill for the GPU buildout and the Hokkaido datacenter expansion lands before the GPU-cloud revenue ramp clears. Tanaka is paying for the GPU footprint now and earning back the rental revenue from those GPUs over the H100 / B200 product life. Brokerage price targets were cut after the profit print; the equity story remains a long-duration capex bet, not a stable-earnings story.
What the Microsoft deal changes: It de-risks the demand side of the bet. The +20.2% close on April 3 was the market saying: a U.S. hyperscaler that has just named you as its Japan partner is essentially co-signing your GPU procurement curve.
Operationally, Sakura is executing on three parallel tracks. The first is hardware: the dedicated eight-GPU NVIDIA H100 cloud SKU Telecompaper covered when it launched, and the B200 / Blackwell roadmap ITBusinessToday reported. The second is platform: the Generative AI Platform Sakura launched on August 12, 2025 — a corporate-primary release that put a managed-LLM layer on top of the GPU-rental business and made Sakura competitive with the domestic-Japanese AI builds rolling out of NEC, Fujitsu, and the telco-cloud consortiums. The third is physical: the container datacenter announcement on August 14, 2025, which is how Sakura plans to scale rack-density without waiting on conventional datacenter-construction timelines.
NVIDIA’s own newsroom acknowledges the Japan cloud-partner ecosystem — Sakura alongside SoftBank, KDDI, NTT, GMO, and Highreso — as the Japanese leg of its global AI-infrastructure partner program. Of that group, Sakura is the only listed company whose entire enterprise value is currently structured around the AI-cloud transition: SoftBank is a conglomerate, KDDI and NTT are telcos, GMO is an internet holding company. Sakura’s three-letter ticker (3778:JP) is the cleanest publicly-traded read on Japanese sovereign AI compute as a thesis.
Allied sovereign-AI buildouts are not happening in a vacuum. They are the operational consequence of U.S. export-control choices made over the last three years. The Biden-era AI Diffusion Rule pushed allies toward provisioning their own GPU footprints. The Trump-Lutnick rewrite — replacing the rule with bilateral deals — moved the same allies toward U.S.-hyperscaler partnerships on their own soil. Both pathways converge on the same underlying picture: more sovereign datacenters, more domestic GPU procurement, more national budget lines for AI compute.
Sakura Internet is the cleanest single case study of that convergence. The CEO has named the number. The PM has funded the envelope. The Commerce Secretary has cleared the export pathway. The U.S. hyperscaler has signed the partnership and put $10 billion behind it. The stock market has repriced the company. The broker community has flagged the profit collapse as the expected cost of the capex bet, not a thesis-breaking failure. The whole architecture — Microsoft, Lutnick, Trump, Takaichi, METI, Sakura — works only if the GPU supply actually lands and the AI-cloud revenue ramp materializes on the timeline Tanaka has staked the company to.
The first Strategic Investment Fund project allocation. Akazawa and Lutnick have been negotiating first-project allocations since February 2026. The first announced project will tell us whether the fund flows toward semiconductor fabrication (Rapidus, TSMC Kumamoto, Micron Hiroshima), toward cloud infrastructure (Sakura, SoftBank, KDDI, NTT), or toward hardware ecosystem (NVIDIA-partner Japan supply chain).
The H1 FY27 Sakura earnings print. Whether the revenue ramp from H100 and B200 deployments closes the profit gap on the schedule the IR materials projected. If the gap widens, broker price targets compress further; if it closes, the equity-story thesis is validated.
The next METI subsidy round.Whether Takaichi’s ¥1.23 trillion FY26 envelope translates into another large Sakura line-item or whether the funding spreads across Rapidus, NTT, KDDI, and the SoftBank-Microsoft channel. The Japanese government is structurally biased toward multi-recipient distribution; Sakura’s relative share will be the read.
Tanaka’s next number. The 50,000-100,000 figure was October 2024. If Tanaka revises it upward in his next public interview, the demand backlog is widening faster than the GPU procurement curve is closing it — and the architecture has to scale again.