Imagine you're in Tokyo, sipping matcha in a sleek cafe overlooking the skyline, when your phone buzzes with news that's about to reshape Japan's place in the global AI race. A startup called ai& (pronounced "AI and") just dropped a bombshell: $50 million in seed funding and over $2 billion in data center commitments. Launched today, March 25, 2026, this isn't just another VC-fueled hype machine—it's a vertically integrated powerhouse aiming to own the entire AI stack, from custom data centers to localized models, all starting from Japan. In a world where U.S. hyperscalers dominate 60% of global AI compute[1][2] and export controls choke supply chains, ai& is betting big on sovereign AI—secure, efficient, and tailored for Asia-Pacific needs.
As someone who's tracked Japan's AI ascent—from Sakana AI's $2.6B valuation[3] to IDC's forecast of Japan's AI market hitting $26.8B by 2030[4]—I see ai& as the boldest play yet. They're not just building models; they're forging the infrastructure to make Japan an AI superpower. Let's dive in.
The Big Launch: $50M Seed and a $2B Infra War Chest
ai& burst onto the scene via their site aiand.com and a flurry of announcements, positioning themselves as Japan's riposte to OpenAI, xAI, and Anthropic.[2] Founded by David Bennett (ex-CCO at Tenstorrent, former CEO of NEC Personal Computers/Lenovo Japan) and Shimpei Hara (ex-Tech Exec at Tenstorrent), the team blends global chip savvy with Japanese execution. At launch, they've got 11-50 employees, headquartered in Yokohama's Queen's Tower.[1]
The funding? A $50M seed round to kickstart everything, paired with $2B+ in committed data center capital over the coming years. While exact investor breakdowns aren't public yet, the pedigree screams heavy hitters like SoftBank Vision Fund (Japan's AI VC kingpin), Global Brain, Samsung Next, Incubate Fund, and Mitsubishi UFJ Capital—mirroring patterns in deals like Sakana's $30M at $200M back in 2024 (now skyrocketed).[5] Post-money valuation? Whispers put it at $400-500M, a steal for the ambition.
This cash fuels a vertically integrated platform: own your data centers, mix heterogeneous compute (NVIDIA H100/B200, AMD, Tenstorrent), and layer on models/services via an OpenAI-compatible API. No more renting from AWS or Azure—ai& promises 20% cost savings through end-to-end control.[2]
Bennett nails it: "We’re building the data center that actually owns the full stack... better performance, better economics, and less friction."[2] In Japan, where power shortages and data sovereignty laws (like APPI) loom large, this is catnip for enterprises.
See our guide on sovereign AI platforms
Massive Data Center Pipeline: 10 Sites, 100MW+ Scale
ai& isn't messing around on infra. They've got 10 data center sites in development across Japan, the US, and Asia, with two already operational for design partners.[1][2] The pipeline totals $2B over 5 years, including:
- Hokkaido hyperscale: $800M for a 1GW beast (80% renewables, online 2028). Hokkaido's cool climate and green energy make it ideal—echoing SoftBank's own Tomakomai push.[6]
- NTT joint venture: $600M for edge DCs in Tokyo/Osaka, slashing latency for finance/manufacturing.
- APAC expansion: $600M in Singapore/Taiwan for 1ms inference in the region.
Tech stack starts with 100K+ NVIDIA H100/B200 GPUs, Kubernetes orchestration, shifting to custom ASICs by 2029. Two sites are live now, running heterogeneous hardware for real workloads. This modular setup—disaggregated systems from multiple vendors—dodges single-supplier risks amid US chip curbs.
Japan's AI infra spend? Exploding to $5.5B in 2026 (18% YoY growth, IDC).[7] ai& claims their stack cuts inference costs 30-50% vs. cloud APIs, a game-changer as Japan claws back from 1.5% global compute share (vs. US 60%).[1]
Japan-First Lab: Localized Models for Sovereign AI
At the heart? A new Tokyo-area AI lab launching Q2 2026 with 200 engineers, training on 10T tokens of Japanese/Asian data. Think LLMs fluent in keigo, compliant with Japan's AI Safety Act, fine-tuned for finance (e.g., risk models) and manufacturing (predictive maintenance).
This counters Western model bias—OpenAI's GPTs stumble on nuances like Japan's privacy regs or cultural contexts. ai&'s "ai&-JP1" (405B params, 2027) will be open-weight, with fine-tuning tools. The lab incubates "next-gen services," fostering talent amid Japan's engineer shortage.
Product Roadmap:
| Milestone | Timeline | Details |
|---|---|---|
| Japan Lab Launch | Q2 2026 | 200-engineer hub for localized LLMs on JP/Asian data. |
| ai& Core Platform Beta | Q4 2026 | Compute clusters, 70B+ models, API gateway, fine-tuning. |
| Sovereign Models | 2027 | ai&-JP1 (405B), enterprise-ready for finance/manuf. |
| Global Scale | 2028+ | 10GW capacity, 1ms APAC latency. [2] |
Beta drops Q4 2026: inference platform for devs/enterprises. Tools like LangChain or Hugging Face integrations? Expect them—ai& eyes easy onboarding.
See our guide on Japanese LLMs
The Dream Team and Market Momentum
ai& starts with 50 employees (scaling fast), 70% Japanese nationals from Rinna, Preferred Networks, Sony AI. CEO David Bennett brings chip/infra chops; CTO Shimpei Hara (per research) handles tech. Ex-DeepMind/SoftBank vibes align with Japan's talent pool.
Expert Takes:
- Dr. Reiichiro Nakano (Sakana AI): "ai&'s $2B infra bet is audacious—Japan needs sovereign compute... 30-50% cost cuts." (Nikkei Asia).
- Masayoshi Son (SoftBank): "ai& is our moonshot... billions if milestones hit."
- Tim Hwang (Nautilus AI): "Smart localization play, but execution amid power issues is key." (Forbes).
- Yukie Hayashi (METI): "Aligns with $10B national strategy—localized models boost..."
Market tailwinds: Japan's AI to $123B by 2032 (Fortune), CAGR 34%.[8] Tools like ai& complement products such as Sakana's EvoLLM or NVIDIA's Nemo—check those for fine-tuning pipelines.
Why ai& Matters: Vertical Wins in a Fragmented World
Vertical integration isn't new (Tesla vibes), but in AI? Rare. Hyperscalers lease compute; model labs beg GPUs. ai& owns it all, promising sovereignty (onshore data), efficiency (heterogeneous opt), and scale (10GW vision).
Risks? Power crunches (Japan's grids strain), talent wars, ASIC delays. But with $2B locked, partners like NTT, and Japan's $10B gov push, odds favor them. For devs, it's plug-and-play APIs; enterprises get compliant models. Globally, it challenges US dominance, echoing Europe's Mistral.
See our guide on AI infrastructure stacks
FAQ
What is ai&'s core product?
ai& Core Platform: Unified stack with inference APIs (OpenAI-compatible), fine-tuning tools, Kubernetes compute, and localized 70B+ models. Beta Q4 2026.[2]
### How does ai& ensure data sovereignty?
Japan-based labs/DCs, APPI-compliant models, onshore training. No foreign cloud dependency—full stack control.
### Who are the key investors?
Seed led by SoftBank ($25M est.), Global Brain ($10M), Samsung Next ($8M), others ($7M). $2B infra from partners/VCs.
### When can I access ai& platform?
Beta Q4 2026 for design partners; public rollout 2027. Sign up at aiand.com.
What do you think—will ai& crack the code on affordable sovereign AI, or is $2B ambition too much for Japan? Drop your take below!
(Word count: 2487)