Amazon Web Services and Microsoft Deploy $1 B and $2.5 B AI Engineer Units as Chip Startups Scale Up

Amazon Web Services announced a $1 billion investment to create a Forward‑Deployed Engineering group that places AI engineers inside customer organisations for short‑term projects. Teams of five‑to‑six engineers will work on‑site for about 45 days to build production‑ready AI agents that remain on the customer’s own AWS cloud. Initial pilots include the National Basketball Association and electronics maker Ricoh. Microsoft formed a new Frontier Company with $2.5 billion of funding to help large enterprises choose and integrate a mix of AI models and tools, keeping the results in‑house. Early customers such as Unilever and Novo Nordisk will receive advisory services and integration support across Microsoft‑provided and third‑party models. Nvidia introduced a partnership framework with specialised AI‑cloud providers, offering revenue‑sharing and credit support to build “AI factories”. Early collaborators include Sharon AI, which will deploy up to 40 000 Grace Blackwell GPUs, and Firmus, which is expanding a 360 MW campus in Batam, Indonesia for up to 170 000 GPUs. AI‑chip startup Etched, valued at $5 billion after an $800 million funding round, reported $1 billion of contract orders for its “frontier inference clusters”, integrated systems of its custom‑designed chips, racks and software. The company claims its low‑voltage inference technology delivers high throughput without thermal throttling and its hybrid HBM‑SRAM memory architecture reduces latency. Oxmiq raised $35 million to develop a single‑block AI‑chip IP that merges CPU, GPU and tensor‑engine functions and a chiplet‑based fabric for future custom‑chip products. Taiwan Semiconductor Manufacturing Company (TSMC) is advancing a new CoPoS (Chip‑on‑Panel‑on‑Substrate) packaging technology that replaces circular wafers with rectangular glass panels, aiming to raise AI‑chip output by up to four‑fold and lower costs. Socionext will use TSMC’s A14 process to design an AI‑data‑centre SoC, with tape‑out planned for September 2026.