SK Telecom, NTT and Chunghwa Telecom launched the IOWN AI Fund on June 10, 2026, a roughly $500 million vehicle backing startups in optical networking, power-efficient cooling and AI chips. Catalight Capital, a new firm with offices in Silicon Valley and Tokyo, will manage it. The Development Bank of Japan and former Samsung strategy chief Young Sohn join as founding partners, with the fund set to close by the end of June 2026.
The bet is that AI’s real constraint is power and bandwidth, not algorithms. Over 20 companies, including SK Hynix, have signalled interest in joining.
Telecom operators have spent two decades as the pipes other companies got rich on top of. The three behind the IOWN AI Fund are now trying to own the architecture of the AI boom itself, not just carry its traffic.
On June 10, 2026, SK Telecom of South Korea, Japan’s NTT and Taiwan’s Chunghwa Telecom signed off on a roughly $500 million fund in Tokyo’s Otemachi district. The money targets startups working on the unglamorous parts of AI: cooling, power efficiency, chips, and the optical links that move data between servers. The pitch is simple. The thing slowing AI down is no longer how clever the models are. It is whether the grid can power them and the network can feed them fast enough.
That reframing is the whole story. A telco that funds the standard for low-power optical networks does not just sell capacity. It decides where data centres get built, which chips connect to its fibre, and which cloud platforms must play by its rules.
The fund is a bet on physics, not software
The fund is named after NTT’s Innovative Optical and Wireless Network, and its centrepiece is the All-Photonics Network. NTT claims the design can cut network power use by up to 99% and lift capacity more than 100-fold by keeping data as light from end to end. Those are the company’s own figures, published on its All-Photonics Network technology page. They describe a target, not a deployed average.
There is one real-world data point worth more than the marketing. In August 2024, NTT and Chunghwa lit the first international IOWN link between Taiwan and NTT’s Musashino centre near Tokyo. It carried a one-way latency of 17 milliseconds over roughly 3,000 kilometres. That is a demonstration, not a commercial service. But it proves the physics works across an ocean, which is the part most optics pitches never get to.
Catalight Capital will hunt across the AI value chain — accelerators, GPUs, liquid cooling, inference software, and optics — in North America, Asia and Europe. Young Sohn, the fund’s operating partner and former Samsung chief strategy officer, frames the thesis as backing startups that fuse optics, chips and AI to break the energy and bandwidth ceiling. The partners are offering more than cash: technology validation, customer access, and a live network to test on. The forward bet is that within twelve to eighteen months, whoever sets the low-power optical standard locks in the vendor ecosystem around it. That is the prize, and it is bigger than any single startup return.
AI’s new bottleneck is the power bill
The reason telcos can make this move is that AI hit a wall everyone saw coming. The International Energy Agency projects global electricity demand from data centres, AI and crypto could top 2,000 TWh in 2026 — roughly Japan’s entire 2022 consumption, according to the agency’s Electricity 2024 report. IEA executive director Fatih Birol has warned that this demand risks straining power systems unless efficiency investment speeds up.
This is now a race, not a single project. NTT, Ciena and Cisco are pushing coherent optics for AI backbones; Nokia leads in Europe; Huawei and China Mobile are building domestic photonic networks. US hyperscalers like Google and Microsoft design custom interconnects for their own data centres. Whoever standardises a cheap, low-power optical fabric first shapes who everyone else has to buy from.
So the pipes are trying to become the platform. A telco that funds the optics, runs the fibre and validates the startups stops being a utility and starts being a gatekeeper. The fund closes by the end of June 2026. If the money actually commits, it means financial and industrial backers believe optical-first AI infrastructure is real enough to bet on now — not in some later cycle.
Beyond the headline
The power behind it
Control over AI’s plumbing is shifting from cloud giants alone to the network owners who decide where low-latency, high-bandwidth links get built. By backing a standard they designed, NTT and its partners can steer traffic patterns, vendor choices and even data-centre geography — leverage over both chip suppliers and the cloud platforms that must connect to their fibre.
The bigger picture
AI’s growth is now limited by electricity grids, cooling and long-haul bandwidth more than by algorithms. This fund reflects infrastructure operators moving upstream into AI architecture itself, turning energy efficiency and photonics from back-office engineering into the battleground that decides which regions can afford to host the next wave of compute.
The reach
For European cloud and enterprise buyers, a telco-driven optics ecosystem in East Asia could quietly shape equipment choices and compatibility standards baked into undersea cables and cross-border links. If IOWN-style networks prove cheaper per AI transaction, European data-centre operators may face pressure to adopt compatible photonics or risk higher costs and weaker appeal as AI-hosting sites.
The questions to ask before the June close
With the fund set to formally establish by the end of June 2026, three groups have specific reasons to track what commits and what does not.
- Infrastructure investors and advisers
Monitor NTT’s official IOWN updates at rd.ntt/e/iown for real deployments of All-Photonics links, not demonstrations. Live commercial rollouts over the next few quarters could shift demand toward Western optical vendors like Ciena and Arista — or away from them if Asian telcos standardise their own gear.
- Equity holders in AI-networking firms
If you hold NVIDIA, Broadcom, Marvell or Cisco exposure, watch whether the fund’s portfolio companies compete with or complement these vendors. A telco-led standard pushing into Asia and Europe new builds is both a demand tailwind and a pricing risk within six months.
- Energy and ESG-focused readers
Read the IEA’s Electricity 2024 report at iea.org to benchmark any AI-related investment against power-market limits. The same constraint driving this fund — data centres approaching Japan’s national electricity use — will define which regions can host AI compute at all.
Explainer
- IOWN AI Fund
- A roughly $500 million investment fund launched by SK Telecom, NTT and Chunghwa Telecom on June 10, 2026, to back AI infrastructure startups. Founding partners include the Development Bank of Japan and former Samsung strategy chief Young Sohn, with management run by the new firm Catalight Capital. SK Hynix is preparing to join, and Japanese reports name Sony, Toshiba, Fujitsu and the country’s three megabanks as potential backers.
- All-Photonics Network
- NTT’s network design that keeps data as light from end to end, avoiding the repeated optical-to-electrical conversions that waste power and add delay. NTT claims it can cut network power use by up to 99% and raise capacity more than 100-fold, though these are design targets rather than measured field results. It forms the core technology of NTT’s broader IOWN initiative, first proposed in 2019.
- Energy Conservation Act
- Japan’s framework, revised in 2024, requiring large energy users including data centres to submit periodic energy-saving plans and report power consumption to METI. It governs AI facilities through general efficiency rules rather than AI-specific law. This lighter approach lets operators deploy experimental optics like IOWN faster than the EU’s stricter data-centre reporting regime under its recast Energy Efficiency Directive.
RELATED STORIES
