Lower-cost GPU production service debuts
KDDI Eyelet launched a GPU procurement and AI inference support service on Akamai Cloud aimed at helping companies move from testing to production without blowing budgets. (itbusinesstoday.com) The offering frames lower-cost managed GPU capacity as a bridge for organisations scaling AI workloads into predictable production. (itbusinesstoday.com)
A graphics processing unit, or GPU, is the chip that handles the heavy math behind modern artificial intelligence, and KDDI Eyelet has started selling a managed way to rent that capacity on Akamai Cloud from April 13. (iret.co.jp) The new service covers GPU procurement, system design, deployment and round-the-clock operations for companies trying to run generative artificial intelligence and large language models in production, not just in tests. KDDI Eyelet said it is offering the service through its cloudpack managed-cloud business. (iret.co.jp) KDDI Eyelet said the entry plan uses one NVIDIA RTX 4000 Ada graphics processing unit for $0.52 an hour, or $350 a month, with 4 virtual central processing units, 16 gigabytes of memory and 500 gigabytes of storage. The company said pricing is the same across regions, including Tokyo and Osaka. (iret.co.jp) Akamai says its cloud GPU lineup includes NVIDIA RTX 4000 Ada, Quadro RTX 6000 and RTX PRO 6000 Blackwell Server Edition options. KDDI Eyelet said its new service also supports custom quotes for RTX PRO 6000 Blackwell systems. (techdocs.akamai.com) (iret.co.jp) The pitch is about inference, the stage where a trained model answers real user requests, not training, which is the longer and more expensive process of teaching the model in the first place. KDDI Eyelet said many companies have struggled to move from pilot projects to live services because GPU cloud costs are hard to predict and domestic options are limited. (iret.co.jp) KDDI Eyelet said Akamai Cloud can run inference from locations closer to end users, and cited response times of 0.24 seconds under 100 concurrent users. The company also said customers can keep data in Japanese regions while using graphics processing units in Tokyo and Osaka. (iret.co.jp) The service also leans on vendor choice as a selling point. KDDI Eyelet said customers can build in an open environment and combine or migrate workloads from existing cloud setups instead of locking themselves into one provider. (iret.co.jp) The launch extends a broader push by Eyelet into managed GPU access. In March 2026, the company introduced a separate Google Cloud-based service focused on securing graphics processing units for artificial intelligence development teams. (itbusinesstoday.com) KDDI Eyelet said it signed a partner agreement with Akamai in November 2025, and it is now using that relationship to sell lower-cost capacity as a packaged production service. The company is betting that predictable hourly pricing and managed operations will matter as much as raw chip performance for businesses putting artificial intelligence into live systems. (iret.co.jp)
