Broadcom announced during its Q4 2025 earnings call that it secured a substantial $10 billion order in the previous quarter to supply Google’s latest Tensor Processing Units (TPUs) to Anthropic. CEO Hock Tan disclosed that, in Q4, the company received an additional $11 billion order from the same customer, intended for delivery in late 2026. This brings Anthropic’s total TPU orders to a remarkable $21 billion.
In addition, Broadcom revealed a significant $73 billion backlog of AI product orders, which are anticipated to be shipped over the next six quarters. The growing demand reflects the escalating interest in AI technologies across various sectors, as companies seek to enhance their computational capabilities.
TPUs are specialized accelerators developed by Google to handle AI workloads. Currently in their seventh generation, TPUs are available to clients through Google Cloud and are critical for many of Google’s internal operations, including the training and deployment of the Gemini family of models. While Google designs the TPU architecture, Broadcom is responsible for transforming these designs into manufacturable silicon, thus managing volume production. This partnership exemplifies Google’s strategy of maintaining control over AI hardware design while leveraging semiconductor partners for fabrication expertise.
Anthropic, a long-time user of TPUs, recently announced ambitious plans to scale its infrastructure significantly, aiming to deploy one million TPUs, supported by over one gigawatt of new compute capacity set to come online in 2026. This expansion represents one of the most substantial dedicated AI compute buildouts in the industry, underscoring the rising demand for efficient AI processing capabilities.
Other tech giants such as Meta, Cohere, and Apple have also confirmed their use of TPUs, with reports indicating that Meta is assessing the deployment of TPUs in its data centers starting in 2027. The increasing adoption of TPUs can be attributed to their power efficiency and optimization for AI training and inference, which poses a growing competitive threat to the dominance of NVIDIA in the GPU market.
Broadcom has stated that it now has five TPU/XPU (custom AI accelerator) customers, with Google and Anthropic confirmed during the call. Industry analysts suggest that companies like Meta and ByteDance are also among its custom AI chip customers, although Broadcom has yet to publicly verify the complete list.
The rise of TPUs, which are fine-tuned to handle AI workloads, is challenging NVIDIA’s long-held dominance in the GPU arena. Recent analysis from SemiAnalysis indicates that while the TPU v7 has roughly 10% lower peak floating-point operations per second (FLOPs) and memory bandwidth compared to NVIDIA’s GB200 platform, it still offers superior performance when considering the total cost of ownership (TCO).
SemiAnalysis estimates that Google’s internal cost to deploy its TPU solution, dubbed Ironwood, is approximately 44% lower than deploying an equivalent NVIDIA system. Even when priced for external customers, the TPU v7 presents an estimated 30% lower TCO than NVIDIA’s GB200, and about 41% lower TCO than the upcoming GB300. This cost efficiency is increasingly appealing to companies looking to optimize their AI infrastructure.
If Anthropic can achieve around 40% machine-fraction utilization (MFU) on its TPUs—a figure deemed realistic given the company’s compiler and systems expertise—the effective cost of training per FLOP could be 50-60% lower than what is expected from GB300-class GPU clusters. This potential for cost savings positions TPUs as a compelling alternative for companies seeking to enhance their AI capabilities.
The escalating demand for TPUs and their proven effectiveness in AI workloads suggest a transformative shift in the landscape of AI hardware, with significant implications for the competitive dynamics between major players in the semiconductor and AI sectors. As companies like Anthropic continue to expand their computational resources, the adoption of TPUs may play a pivotal role in shaping the future of AI technology.
See also
Oracle’s $50B AI Spending Plan Backfires, Sparks 15% Stock Drop and Market Reassessment
OpenAI Launches GPT-5.2 to Compete with Google, Enhances Productivity Tasks by 70.9%
LG and Qualcomm Unveil AI Cabin Platform for Next-Gen In-Car Infotainment at CES 2026
Disney Commits $1 Billion to OpenAI for Innovative Sora Video Creation Platform
