NTT Inc. is repositioning itself around AI-native infrastructure and photonics after weaker-than-expected mobile profitability in its latest earnings.

The Japanese telecom giant reported approximately $91.7 billion in operating revenue for the fourth quarter (Q4) of its fiscal 2025, which was a 5.1% increase compared to the previous year. Operating profits increased 3.4% to roughly $10.9 billion and net profits increases 3.7% to about $6.6 billion over the same period.

Despite the positive momentum, NTT pushed its previously released consolidated earnings before interest, taxes, depreciation, and amortization (EBITDA) target of approximately $25.5 billion from 2027 to fiscal 2030. NTT CEO Akira Shimada pointed to several reasons, including weaker consumer profitability and NTT's DoCoMo mobile operations having seen its consumer EBITDA drop in half over the past several years.

“Given that the population in Japan will be declining, it's going to be difficult to maintain the number of handsets," Shimada said, adding that the company will have to lean on other operations. This includes IoT and all-photonic network (APN) efforts through the Innovative Optical and Wireless Network (IOWN) Global Forum.

NTT positioned APN as essential to its AI infrastructure pivot.

“In light of these changes in the business environment, we are renaming our growth area as value-added areas and will focus on AI,” Shimada revealed, echoing similar moves by South Korea's SK Telecom. “We will retain our cash generation capabilities by stabilizing profits while renaming it the connectivity area and transitioning it into an IOWN and AI-native infrastructure.”

NTT and IOWN APN

The move hasn’t come out of nowhere; at this year’s Mobile World Congress (MWC) event, DoCoMo revealed an in-network computing capability within its mainline 5G core that connects the mobile network to NTT's IOWN APN service and is designed to help coordinate AI inference processing in tandem with traffic control. This followed an APN service in Hong Kong unveiled by DoCoMo last year, targeting low-latency needs for financial institutions in the region.

In a recent interview, Masahisa Kawashima, NTT’s IOWN technical director and chair of the IOWN Global Forum, highlighted the use case of synchronous database replication as mandatory for many financial service institutions, citing both the Hong Kong case study and another with Japan’s Mitsubishi. Kawashima argued that with traditional technologies, replication can result in significant performance issues, while the IOWN-APN method counteracts by connecting the main data center with a secondary one.

“The performance of databases with IOWN-APN deployment is very high, and [Mitsubishi] were very surprised at this performance,” Kawashima claimed.

The NTT director added that while countries like Japan are interested in developing a sovereign AI infrastructure, many enterprises remain reluctant to use graphic processing unit (GPU) servers shared on the cloud due to concerns over data leakage. This has led to the IOWN Global Forum developinga proof of concept offering a zero-trust security model for GPU sharing, which saw AI training for 3D medical image recognition completed across a 24-mile APN link. According to tests run by NTT DoCoMo, while training using the local GPU took 357 seconds, running the same remote training over APN across the same distance was only 1.9% slower at 364 seconds, with zero data loss due the private GPU infrastructure.

Kawashima also argued that with AI demanding more bandwidth from networks, IOWN-APN helps networks support any data flow.

“With IOWN-APN as the underlying transport layer, packet nodes can be centralized to the cloud … and we can add many value-added services to this packet, such as security functions and so on," Kawashima said. “This way each data center does not have to be so large. Instead, each can be a size that can be operated with locally-available renewable energy.”

Kawashima explained this underpins NTT's AI efforts, with APN better powering both AI training and inference as high‑speed, low‑latency optical links tie together distributed AI data centers.

“What we need is to distribute AI inference models with agility and in accordance with varying workloads like CDNs (content delivery networks)," Kawashima said. This connects high-bandwidth, low-latency between AI model repositories and inference locations.”