TinyFish and NESIC target AI agent data bank for Japan's enterprises

A US web-native AI startup and Japan's NESIC are co-building a verified data layer to underpin autonomous AI agents across Japanese enterprises.

By Disrupts Editorial Team Fri, 7 August 2026

A brightly lit, modern control room features a curved wall of multiple screens displaying abstract blue data graphics and text, with a sleek white counter in the foreground.

TinyFish, a Palo Alto-based AI startup founded in 2024, and NEC Networks and System Integration Corporation (NESIC), a Tokyo-headquartered systems integrator, have announced a strategic partnership to co-develop what they are calling a "data bank" for AI agents. The initiative sits inside NESIC's AI Agent Ready Project, which aims to accelerate enterprise and local-government adoption of autonomous AI across Japan. The partnership represents a direct collision between two distinct capability sets: TinyFish's web-native model architecture and NESIC's deep penetration into Japan's enterprise and public-sector infrastructure.

At the centre of the collaboration is Mako, TinyFish's purpose-built web model. Unlike general-purpose frontier models, Mako is trained specifically on production enterprise web-task data and is designed to operate the live web rather than merely reason about it. The company says the model runs at 35 billion total parameters but activates only three billion at inference, making it materially cheaper to run continuously than frontier-scale alternatives. The design logic is deliberate: web execution, TinyFish argues, does not demand frontier-scale reasoning, and right-sizing the model opens the door to always-on deployment across an enterprise, rather than rationed, high-cost usage.

Verified data as the missing layer

The core problem the partnership is trying to solve is not AI reasoning but AI reliability. As generative AI spreads through enterprise workflows, the bottleneck has shifted from answering questions to acting on accurate, real-time information. Web-sourced data is notoriously patchy: pages are dynamic, APIs are absent, and AI hallucination remains a structural risk. TinyFish says Mako correlates and cross-checks the information it gathers, identifies gaps in its own knowledge, and returns only data it has verified, with the aim of feeding downstream AI agents that make consequential operational decisions.

The first live use case under the partnership is Business Continuity Planning (BCP). In a disaster scenario, accurate, real-time information from fragmented social media and official sources is both critical and scarce. NESIC and TinyFish say the system will extract only verified facts in real time to support rapid decision-making. From that starting point, the two companies plan to extend the data bank across supply chain, marketing, and finance domains, pairing TinyFish's web-gathered data with the proprietary enterprise data NESIC's customers already hold.

Cross-sector and geopolitical read-across

The strategic logic here reaches beyond a bilateral technology deal. Japan has a well-documented structural dependency on foreign AI infrastructure, and NESIC's involvement signals that large domestic systems integrators are now actively seeking to build sovereign-adjacent AI data layers rather than simply reselling foreign cloud AI services. For cross-sector investors watching the geography of AI infrastructure build-out, Japan sits in an interesting position: a mature enterprise base, significant public-sector digitalisation pressure, and a regulatory environment that is cautious but not hostile to AI adoption.

The "data bank" framing also carries implications for the broader agentic AI market. The current wave of enterprise AI spend has concentrated on reasoning models and orchestration platforms. What is less funded and less solved is the reliable data-acquisition layer that agents need to act rather than merely plan. If TinyFish and NESIC can establish a replicable architecture for verified, structured web data at enterprise scale, the model is portable to other geographies and verticals, including defence procurement monitoring, financial intelligence, and pharmaceutical supply chain surveillance, all sectors where data freshness and accuracy carry direct cost or risk consequences.

Capital interest in agentic AI infrastructure has been intensifying across 2025 and into 2026, with VC and corporate balance-sheet spending shifting from foundation-model training to the operational stack around it. TinyFish is an early-stage company, founded only in 2024, and this partnership with a large incumbent integrator follows a well-worn internationalisation playbook for US AI startups seeking enterprise distribution in markets where local trust and compliance relationships are decisive. The terms and any associated capital commitment are not disclosed.