AWS ジャパン wraps フィジカル AI program with 13 projects, 14 firms

AWS Japan's program ran from January through the following months, with development beginning in March. According to AWS Japan's Yoshitaka Haribara, who designed and ran the program, the company had felt since early 2023 that a support program for Japan's foundation model development was needed, and it began supporting domestic LLM work in July 2023 before extending that support to physical AI. Participating companies applied, and AWS Japan chose to structure the aid around technology, cost optimization, community and go-to-market support.

The results presented at the August 31, 2026 briefing were wildly different in scale and ambition. ZEALS reported moving from importing Chinese robots to developing its own D1 hardware and its Omakase Zen AI model in four to five months, and said it had already deployed D1 at a hospital in Nagoya for 35 hours. Mercari, an app marketplace rather than a robot maker, wants to automate 20% of its own warehouses by 2028 with about 200 million yen in annual cost savings. Takenaka, whose workforce numbers about 8,000, reported a 10–15% success rate for a wall-rolling task and flagged that construction sites restrict data collection. Highlanders reported that it is preparing a domestic humanoid for mass production around April 2027. These differences suggest that physical AI is still being tested against real-world constraints rather than being a single finished technology.

Haribara's summary was that physical AI requires speed, on-site knowledge and cloud. The fact that companies reported half a year of progress on model and pipeline work, but with varying success rates and often against simulated or limited real-world data, suggests the next test is whether these results hold up in daily operations. Mercari's cost-savings target and ZEALS's production and operating-hour goals are the clearest benchmarks that readers can watch to see whether this support translates into sustained deployment.