NTT's IOWN controls factories 300km away, tsuzumi 2 runs on single GPU

Photonics demonstration and lightweight AI address manufacturing labor shortages and energy consumption simultaneously

While NTT's optical quantum computing partnership captured headlines at the company's R&D Forum 2025, two more immediately significant announcements demonstrated technologies working today rather than promising breakthroughs by 2030.

NTT and Toshiba Corporation successfully controlled manufacturing equipment from 300 kilometers away—the distance between central Japan's industrial heartland and Tokyo—maintaining the 20-millisecond response times required for real-time factory operations while running AI visual inspection simultaneously at four frames per second (4fps).

Separately, the company launched tsuzumi 2, a lightweight large language model that runs on a single GPU rather than the dozens or hundreds required by competing AI systems. Neither announcement carried the futuristic appeal of million-qubit quantum computers. But both address urgent problems: Japan's manufacturing labor shortage and AI's accelerating energy consumption crisis.

The IOWN factory demonstration wasn't a laboratory proof-of-concept. Production equipment in the Tokai region was operated by systems in the Kanto region, with AI visual inspection running in real-time. Expensive GPU resources and inspection systems can now be centralized in data centers and shared across multiple facilities, standardizing quality control while reducing capital expenditure.

Meanwhile, tsuzumi 2's single-GPU operation directly challenges the trajectory NTT DATA's own October whitepaper warned about: AI workloads consuming 50% of data center power by 2028. As companies race to deploy AI, NTT is positioning lightweight models as the sustainable alternative to power-hungry frontier systems.

The dual announcements illustrate NTT's parallel strategy: pursue moonshot quantum technologies for long-term transformation while deploying practical photonics and efficient AI solutions delivering immediate business value.

IOWN proves commercial readiness: Remote factory control at scale

The 300-kilometer remote control demonstration represents IOWN's transition from research curiosity to commercial infrastructure. The achievement is technically significant: maintaining 20-millisecond control cycles while simultaneously performing AI visual inspection at 4fps meets industry standard requirements for manufacturing operations.

But the business implications matter more for Japanese manufacturing. The demonstration shows factories in the Tokai region (central Japan's industrial heartland including Nagoya) utilizing visual inspection AI and GPU resources located in the Kanto region (greater Tokyo area).

This geographic span—roughly equivalent to controlling manufacturing in Singapore from Kuala Lumpur—proves IOWN can handle real-time industrial control across distances impractical with conventional networks.

For manufacturers, this enables two critical capabilities:

Labor shortage mitigation: A single team can monitor and manage multiple factories across different regions, addressing Japan's demographic crisis and manufacturing labor shortages through remote control and centralized operations.

Resource optimization: Expensive AI processing power and GPU resources can be centralized in data centers rather than duplicated at each factory site. Multiple facilities can share visual inspection AI, standardizing quality control criteria while reducing capital expenditure.

The demonstration moves IOWN from NTT's 2019 vision—which seemed like distant science fiction promising ultra-high capacity, ultra-low latency, and ultra-low power consumption through photonics—to 2025 deployment with measurable return on investment.

tsuzumi 2: Japan's answer to AI's energy crisis

On the AI front, NTT announced general availability of tsuzumi 2, the second generation of its proprietary large language model designed explicitly to address unsustainable AI energy consumption trajectories. Unlike Western LLMs adapted for Japanese markets, tsuzumi 2 was built from scratch with superior Japanese-language processing ability and built-in multimodal support, handling text, images, and voice within enterprise applications.

The lightweight model operates on a single GPU—contrasting sharply with dozens or hundreds required by frontier models from OpenAI, Anthropic, or Google—while NTT claims performance "on par with or exceeding that of larger models" in business-case deployments.

The positioning is strategic. NTT DATA's October 2025 whitepaper "Sustainable AI for a Greener Tomorrow" warned that AI workloads will drive more than 50% of data center power consumption by 2028. The tsuzumi 2 launch offers NTT's solution to a problem the company itself is highlighting.

The business case centers on three pillars:

Operational efficiency: Single-GPU operation dramatically reduces infrastructure costs compared to deploying massive models requiring specialized GPU clusters.

Energy sustainability: Critical for Asia-Pacific markets where power infrastructure quality and carbon commitments vary significantly. Reducing per-inference power consumption by 10-100× makes AI deployment viable in markets where hyperscaler approaches would strain local grids.

Data sovereignty: As a "purely domestic LLM" built from scratch in Japan, tsuzumi 2 appeals to organizations with data residency requirements or concerns about sending sensitive information to Western or Chinese AI providers.

According to Frost & Sullivan's 2025 recognition of NTT DATA for Asia-Pacific Competitive Strategy Leadership in generative AI solutions, implementations have achieved productivity improvements up to 40%, automation efficiencies exceeding 70%, and significant reductions in development timelines and costs.

Beyond IOWN and tsuzumi: Wider innovation portfolio

NTT's R&D Forum showcased additional technologies moving toward practical application:

Mind captioning technology that generates text descriptions of what humans see by analyzing brain activity—potential applications in assistive technologies for speech impairments or brain-computer interfaces. Other technologies unveiled includes:

Satellite AI data analysis: Collaborating with US provider Loft Orbital, NTT demonstrated AI analysis of hyperspectral images directly in orbit, reducing data transfer by 90% to enable near-real-time disaster monitoring, environmental tracking, and agricultural assessment.

Large Action Model (LAM): Developed with NTT DOCOMO, this system predicts customer behavior and optimizes marketing interventions by pre-learning behavioral patterns—shifting from conversational AI to predictive action AI.

Mobility and security moves

NTT also announced the formation of NTT Mobility, Inc., targeting Level 4 autonomous vehicle services across Japan by fiscal year 2027. Supporting technologies include automatic video quality evaluation for remote monitoring and high-precision position estimation that generates 3D city models from in-vehicle cameras—serving autonomous driving, urban planning, and disaster prevention.

On cybersecurity, NTT Research announced a quantum-secure zero trust suite powered by attribute-based encryption (ABE), addressing the "harvest now, decrypt later" threat as quantum computers approach commercial viability.

Infrastructure play, execution challenge

The R&D Forum reveals NTT's core strategy: position as infrastructure provider for the quantum-AI era rather than competing in end-user applications. IOWN photonics, optical quantum computing, and autonomous mobility networking represent infrastructure bets, while tsuzumi 2 and specialized AI capabilities target specific applications.

This makes strategic sense for a telecommunications company with decades of optical technology R&D but limited ability to out-innovate Silicon Valley or Chinese tech giants in consumer applications. For Asia-Pacific enterprises, it offers alternatives to Western hyperscalers or Chinese cloud providers with regional data sovereignty and energy-efficient solutions.

The challenge is execution. With US$3 billion in annual R&D investment—30% of total profit—NTT pursues multiple moonshots simultaneously: quantum computing (10,000 qubits by 2027), IOWN scaling, tsuzumi 2 market traction, and NTT Mobility operations by 2027.

Unlike quarterly-focused software companies, infrastructure transformation operates on longer timescales. NTT's systematic approach—announcing IOWN in 2019, demonstrating applications in 2025, targeting widespread deployment through the late 2020s—reflects this reality. The smart factory demonstration proves IOWN works; whether quantum computing and autonomous mobility achieve similar commercial success will become clear over the next five years.