Alphabet Pivots to AI Hardware Monetization via Expanded Anthropic Deal

Alphabet is aggressively expanding its footprint in the AI infrastructure market, announcing a new agreement with Anthropic to increase the deployment of its custom Tensor Processing Units (TPUs). While Alphabet has historically used TPUs to power its internal search and AI workflows, the company is now leveraging its partnership with Broadcom to offer this hardware to external entities. This transition is supported by the chips' expanded compatibility with industry-standard frameworks such as PyTorch and JAX, moving beyond the original TensorFlow design. The new agreement focuses on delivering 3.5 gigawatts of TPU compute capacity starting in 2027. This follows a previous arrangement where Anthropic agreed to acquire $21 billion in TPUs from Broadcom within the current year. In a notable shift in strategy, Alphabet is now allowing Broadcom to sell TPU racks directly to Anthropic, enabling the LLM developer to manage its own servers. This move is expected to be highly profitable, creating a diversified revenue stream beyond advertising and cloud services. From a market perspective, this vertical integration provides Alphabet with a structural cost advantage over competitors who rely exclusively on Nvidia GPUs for training and inference. According to estimates from Morgan Stanley, every 500,000 TPUs sold could generate approximately $13 billion in additional revenue and add $0.40 to earnings per share. With the stock trading at a forward price-to-earnings ratio of 25 times 2027 estimates, analysts view this hardware expansion as a primary catalyst for growth.