The artificial intelligence market is booming, with both AI infrastructure and cloud computing providers experiencing substantial growth. At the forefront of this wave are two companies: Alphabet (GOOG) and Broadcom (AVGO), which jointly developed the well-known Tensor Processing Units (TPUs).
Alphabet has established significant cost and performance advantages in the AI race with its self-developed TPU chips, achieving profit growth through multiple channels such as cloud services, model training, and chip sales to external customers. Broadcom, as the physical deliverer of TPUs and a key partner for custom AI chips, directly benefits from the capital expenditures of Alphabet and other hyperscale customers. Deeply intertwined through TPUs, the two companies have formed a unique and powerful growth driver in the AI infrastructure space, positioning them to continue benefiting from the long-term development of artificial intelligence.
TPUs tightly link Alphabet and Broadcom and have become a crucial part of their success. These are two customized AI-specific integrated circuits, co-developed by the companies over a decade ago, and they now form the foundation of Alphabet’s cloud platform. Alphabet is responsible for chip design, while Broadcom provides proprietary IP centered on high-speed SerDes (Serializer/Deserializer) technology and manages physical design and complex packaging integration.
Alphabet’s TPUs provide a significant edge in the AI race. The company has been using these chips to run internal workflows for over a decade and has optimized its entire hardware and software stack around them. This has given Alphabet a head start in the custom AI chip race, with its chips now in their eighth generation and proven in real-world applications. By owning its own world-class AI chips, Alphabet achieves better economics from its cloud business and can train AI models and run inference—most of which heavily rely on NVIDIA’s graphics processors—at costs far lower than competitors.
Last quarter, Alphabet’s cloud revenue surged 63% to $20 billion, while operating income tripled to $6 billion. As of the end of the first quarter of 2026, the company holds a massive $460 billion in Google Cloud backlog orders. Moreover, the reputation of its chips is so high that Alphabet now allows a select group of customers to order TPUs through Broadcom for use either inside or outside Google Cloud, adding another high-margin revenue stream.
Alphabet also uses its TPUs to train its flagship Gemini frontier models and run inference, which lowers costs and gives the company a strong position in the consumer AI market—leveraging its distribution and advertising network advantages to drive growth cost-effectively. With market-leading Chrome browser, Android smartphone operating system, and a search revenue-sharing agreement with Apple, Google effectively serves as the gateway to the internet. The company has integrated Gemini with search and other products to drive profitable growth.
As the only company that possesses both leading-edge chips and world-class frontier models, Alphabet is well-positioned to be a long-term winner in the AI space.
TPUs are also a major growth driver for Broadcom. Because it is responsible for the physical delivery of the chips, Broadcom records the associated sales revenue. Alphabet is investing heavily in AI infrastructure, with a significant portion flowing into TPUs. Additionally, Anthropic has placed a massive $21 billion order for TPUs with Broadcom for delivery this year, and the large language model maker has also committed to purchasing additional computing capacity equivalent to 3.5 gigawatts from the two companies in the future.
Given the success of TPUs, other hyperscale data center operators have turned to Broadcom for help in developing their own custom AI chips. Broadcom expects to generate over $100 billion in ASIC revenue in fiscal 2027, while Citi recently projected that Broadcom will achieve $180 billion in AI-related revenue in fiscal 2028. This represents massive growth for a company that generated less than $64 billion in total revenue last year (with approximately $20 billion of that being AI-related).
As hyperscale operators seek to diversify their supply chains and reduce dependence on NVIDIA (NVDA), Broadcom is poised to become one of the biggest beneficiaries in the future.
