Key Points:
- Alphabet will sell its custom Tensor Processing Units directly to select companies for their own private data centers.
- The technology giant recently revealed two new artificial intelligence chips called the TPU 8t and the TPU 8i.
- Amazon also competes in the chip market and generates an estimated $50 billion in real value from its custom processors.
- Major startups like Anthropic and OpenAI signed massive gigawatt capacity agreements with both Google and Amazon.
Alphabet plans to change how it distributes its most powerful artificial intelligence hardware. Google’s parent company announced on Wednesday that it will begin selling its custom Tensor Processing Units directly to a select group of clients. These customers will take the physical chips and install them inside their own private data centers. This strategic shift represents a direct attack on Nvidia, the undisputed king of the artificial intelligence chip market.
For years, Google kept its custom hardware locked away from the outside world. The company previously forced customers to rent access to these processors exclusively through the Google Cloud platform. If you wanted to use a Google chip, you had to play by Google’s rules in its own digital backyard. Now, executives want to open the doors to expand their business reach. Chief Executive Officer Sundar Pichai explained this new strategy during the first-quarter earnings call. He noted that demand for massive processing power continues to explode among research labs, financial firms, and high-performance computing centers. Giving these clients physical chips allows them to keep their most sensitive data strictly on their own property.
This major business pivot arrives just one week after Alphabet introduced its newest generation of hardware. The company recently revealed two brand new chips designed to handle different artificial intelligence tasks. Engineers built the TPU 8t specifically to train massive new models from scratch. Training requires running billions of calculations simultaneously for months at a time. Meanwhile, the new TPU 8i focuses entirely on inferencing. Inferencing occurs when a finished artificial intelligence system processes new data and responds to user prompts in real time.
Alphabet refused to name all the specific clients receiving these direct hardware shipments, but some major partnerships have already leaked to the public. Earlier this month, the company signed a massive power agreement with the artificial intelligence startup Anthropic. The deal secures multiple gigawatts of power using next-generation hardware, and these systems should come online around 2027. Technology reporters at The Information also revealed that Alphabet secured a multibillion-dollar chip contract with social media giant Meta.
By putting physical chips directly into the hands of outside companies, Alphabet forces a tougher fight with Nvidia. Right now, Nvidia holds a massive monopoly over the hardware that powers the modern artificial intelligence boom. Almost every major technology company relies heavily on its products. Nvidia executives largely brush off the growing threat from Google. They argue that their graphics processing units offer software developers far more flexibility than the highly specialized hardware Google builds.
Google does not stand alone in its quest to challenge the dominant chipmaker. Amazon also aggressively pushes its own custom silicon to technology clients. The e-commerce and cloud computing giant offers a deep lineup of processors, including the Graviton, Trainium, and Nitro chips. Amazon desperately wants its cloud computing customers to use these in-house processors instead of paying premium prices for expensive Nvidia products.
Amazon Chief Executive Officer Andy Jassy highlighted the massive success of this hardware strategy in his recent annual shareholder letter. He revealed that the custom chip business currently operates with an annual revenue run rate of more than $20 billion. However, Jassy explained that this large number actually understates the hardware’s true value. Because Amazon only makes money from these chips indirectly through its Elastic Compute Cloud service, the CEO estimates the real financial value sits closer to $50 billion.
Amazon uses its custom hardware to attract the biggest names in the artificial intelligence industry. Similar to Google, Amazon struck a massive infrastructure deal with Anthropic. The cloud provider agreed to supply Anthropic with 5 gigawatts of processing capacity to train future models. Amazon did not stop there. The company also signed a separate 2-gigawatt capacity agreement with OpenAI, the creator of ChatGPe.
The battle over hardware extends beyond specialized artificial intelligence accelerators. Amazon also competes aggressively in the traditional central processing unit market. The company recently announced a major partnership involving its custom Graviton processors. Amazon will deploy these specific chips to help Meta run its complex agentic artificial intelligence workloads. These workloads require systems that can make independent decisions and complete multi-step tasks without constant human input.
The entire artificial intelligence industry is currently undergoing a massive hardware transformation. The wealthiest technology companies in the world realize they cannot rely on a single supplier to provide the tools for the next computing revolution. By designing, building, and directly selling their own custom silicon, companies like Alphabet and Amazon hope to control the physical foundation of the internet. Customers ultimately win, as this fierce competition drives down prices and sparks brand-new hardware innovations across the entire technology sector.
