Google Cloud delivered 63% revenue growth in Q1 by integrating Gemini AI models and bespoke Tensor Processing Units, enabling the company to lower dependency on Nvidia GPUs and secure a roughly $460 billion backlog of AI-native projects. The shift reflects a broader move by cloud providers to optimize performance and cost efficiency through proprietary AI hardware.

Major cloud operators are accelerating deployment of custom ASIC accelerators, with shipments projected to triple by 2027. Nvidia doubled down on this trend by investing $2 billion in Marvell Technology, positioning the chip designer to capture a growing share of the AI accelerator market as customers seek alternatives to off-the-shelf GPUs.

Cerebras Systems plans to raise $3.5 billion by offering 28 million shares at $115–$125 on Nasdaq, supported by an $850 million credit facility and over $10 billion in investor demand. The company claims its wafer-scale engines outperform standard Nvidia hardware on AI workloads, marking it as a direct competitor in high-performance inference and training solutions.

Anthropic is in advanced talks to secure custom inference chips from Fractile, aiming to reduce reliance on Nvidia’s H100 units and cut operational costs. Fractile’s memory-compute fusion architecture promises up to 100x faster LLM execution and 90% lower expenses, though full-scale deployment is not expected until 2027.