NTT’s Pitch to Hong Kong Banks: Prove Where Your AI Lives

Every bank has some version of the same story. A fraud model trains beautifully. It clears every benchmark. Then it hits production, and a compliance officer asks the question nobody prepped for: where does the data actually sit? Not which cloud. Which building, which jurisdiction, whose law applies if something goes wrong.

Often there’s no clean answer, and the project stalls in a folder marked "pending legal review."

NTT Com Asia used the recent LEAP East 2026 to pitch a fix for exactly that folder. It calls the offering the Financial AI Fabric, a blueprint for what it describes as “sovereign private AI” infrastructure for Hong Kong’s financial sector. The underlying claim: financial institutions in Hong Kong aren’t short on AI ambition. They’re short on infrastructure they can point to and prove, to a regulator, exactly where it lives.

Three pieces, one argument

The blueprint links three things NTT already operates in Hong Kong.

“To successfully move AI from experimentation and sandbox environments into full-scale production, financial institutions need trusted, high-performance infrastructure,” said Steven So, NTT Com Asia’s chief operating officer, in the company’s announcement. Below is what NTT says it built to back that up.

The photonic highway. APN InterLink, NTT’s fiber network, launched commercially in Hong Kong on November 1, 2025, using IOWN technology, a photonics standard that moves data as light instead of electrical pulses. NTT describes the result as delivering “ultra-low latency” connectivity “with zero electronic jitter,” a characterization from the company rather than an independently measured figure. For a CAIO, the underlying stakes are real regardless: the gap between a fraud model scoring a transaction in real time and one that lags by a few milliseconds can be the difference between catching fraud and missing it.

The financial-grade data centers. NTT owns and operates two data centers in Hong Kong, at Tseung Kwan O and Tai Po, which it has built out with direct liquid cooling. LLM training and inference generate a density of heat that standard air-cooled facilities weren't engineered to absorb, a constraint any CDO running GPU clusters in an older server room has likely run into firsthand. Both sites also run on OCEAN Intelligence, NTT's own AI platform for predictive maintenance and energy optimization, according to the press release.

The sovereign GPU layer. This piece doesn’t exist yet. NTT says it’s building toward a GPU-as-a-service launch later this year, a date the company set for itself. The pitch is dedicated compute rather than a shared pool, which the blueprint describes as a “secure compute environment” built for private AI adoption. For a CDO facing a data residency requirement, that distinction is the difference between telling a regulator “our vendor says it’s isolated” and telling them “it’s contractually and physically ours alone.”

Fiber, data centers and GPUs are the plumbing every cloud provider already sells. NTT’s argument is that bundling and auditing all three as one system solves a problem that has been stalling bank AI projects: proving where the thing actually ran, not just whether it worked.

Follow the money to Hong Kong

There’s a geopolitical backdrop to this launch, and NTT names it directly.

Stephen Tsang, the company’s chief revenue officer, pointed to capital moving from the Middle East into Hong Kong, alongside the city’s push into crypto and stablecoins. Infrastructure spending, he said, now sits at the center of “boardroom discussions around ROI optimization and risk mitigation,” which is another way of saying boards are asking CDOs the kind of hard questions once reserved for the trading desk.

In NTT's telling, “sovereign private AI” is a sentence a CDO can say to a board, a regulator, and an auditor and have it mean the same thing to all three. Data stays inside the border. Compute isn’t shared with another tenant. The audit trail sits in a building rather than a clause buried in a vendor’s terms of service. Whether NTT’s infrastructure delivers that in practice is something only a bank running a live production workload on it will be able to say.

It’s time to prove the blueprint

The press announcement focuses on infrastructure: where data and compute physically sit. A photonic network and a liquid-cooled data center don't decide what a model trains on, and they don’t decide who queries its outputs or what a customer-facing chatbot is allowed to say. Model governance, access controls, and output monitoring sit in a different layer of a regulated AI deployment, one the announcement doesn't directly address.

The GPUaaS launch hasn’t happened yet, so the sovereign compute layer this blueprint leans on most heavily remains, for now, a plan rather than a product. Yet, it offers a strong roadmap for FSIs as they come under increasing regulatory scrutiny in and around Hong Kong.

“By establishing this service roadmap, we are reinforcing Hong Kong’s position as a leading financial AI hub and providing the entire FSI ecosystem with a clear, compliant path toward secure, sovereign private AI,” added Daisuke Kuroda, NTT Com Asia’s chief executive officer. 

Image credit: iStockphoto/bingfengwu