Preferred Networks Seeks IPO Path to Mass-Produce MN-Core AI Chips
Tokyo-based Preferred Networks outlined an aggressive path to commercialize its custom MN-Core accelerators, with CEO Daisuke Okanohara telling Japan Times and Bloomberg interviewers around September 7–8, 2026 that sample chips are targeted for the first half of 2027 and commercial supply by December 2027. The company, long known for deep-learning research and industrial partnerships, now frames public listing as a tool to fund foundry capacity and keep pace with global AI silicon races rather than as an immediate exit.
Filed under Funding and dated September 13, 2026, this AI4Japan briefing situates the IPO talk as Japanese semiconductor-industrial news—not a government endorsement. Okanohara’s timeline ties profitability to roughly a three-year window after commercial launch before a listing, which implies investors should expect multi-year cash burn for process qualification, packaging, and software stacks that make MN-Core usable outside Preferred’s own labs.
Why it matters: Japan wants domestic AI compute options as hyperscalers bid up HBM and advanced nodes. A credible MN-Core product could give Japanese manufacturers and cloud operators a second source—but only if compilers, power envelopes, and supply contracts clear the same bar as Nvidia-centric stacks already embedded in Tokyo data centres.
What it means in practice
Japanese technology leaders should treat MN-Core sample calendars as planning inputs, not purchase orders; confirm lawful IP and export-control handling for custom silicon; assign a hardware architect as owner; run time-boxed evaluations against existing GPU baselines on Japanese workloads; and demand published FLOPS-per-watt and software roadmap commitments. Prefer contracts that preserve local logging and fallback paths if accelerator supply slips.
Caveats come first. Interview timelines can slip; foundry allocation is scarce; and profitability-before-IPO messaging can stretch. AI4Japan therefore treats December 2027 commercial targets and the three-year profitability horizon as directional until tape-outs and third-party benchmarks appear.
What to watch next: first customer sample reports in 2027; independent performance audits; and whether Preferred discloses foundry partners and packaging plans. Readers can continue on the AI4Japan homepage for related stories, or browse the Newsroom for additional briefings.
Bottom line: treat this update as orientation, not instruction. Japanese custom AI silicon is accelerating and uneven. Organizations that benefit most will measure pilots honestly, keep engineers accountable for production readiness, and avoid betting critical capacity on a single unproven node.