• Google is launching two new AI chips, the TPU 8tx and 8i Max, targeting secure computing and advanced AI workloads as it challenges Nvidia (NASDAQ:NVDA).
  • The TPUs reportedly cost around 30% less than Nvidia GPUs, offering competitive performance and greater power efficiency for certain tasks.
  • Bloomberg Intelligence estimates Google's TPU-related sales could surge from $12 billion in 2026 to over $135 billion by 2028, though these figures require qualification.

New Specialized Chips

Alphabet (NASDAQ:GOOGL)’s Google is expanding its artificial intelligence chip lineup with two new specialized processors, the TPU 8tx and 8i Max, according to people familiar with the matter. The chips are designed for secure computing and advanced AI workloads, respectively, as the company intensifies its challenge to Nvidia’s dominance in AI accelerators.

The TPU 8tx is aimed at confidential computing, allowing customers to process sensitive data in a secure enclave, while the 8i Max is optimized for handling large language models and other information-intensive AI tasks, particularly as the industry moves toward AI agents. Google’s TPUs are reportedly around 30% cheaper than Nvidia’s GPUs, offering competitive performance and greater power efficiency for certain workloads, though the exact discount varies by application.

The move marks a significant escalation in the AI chip war, as Google transitions from an internal chip developer to a broader supplier of AI computing systems. The company began recognizing direct TPU system sales in the second quarter of 2026, and these sales—which include associated hardware and software—became a meaningful contributor to Google Cloud revenue in the third quarter, according to Reuters (NASDAQ:TRI).

Commercial Traction and Financials

Google’s TPU business is gaining momentum. The company’s latest quarterly results showed Google Cloud revenue surging 103% year over year to $30.7 billion, with operating income jumping to $11.1 billion from $3.6 billion a year earlier. Alphabet’s overall revenue rose 25% to $127.6 billion, though net income was weighed down by investment losses.

Major customers for Google’s TPUs include Anthropic, which uses a mix of Google TPUs, AWS Trainium, and Nvidia hardware, as well as Meta (NASDAQ:META), which has deployed TPUv7 in production using the SGLang-TorchTPU software stack, according to SemiAnalysis. Oracle (NYSE:ORCL) is also a customer, and OpenAI is reportedly planning purchases. These relationships often involve cloud capacity or deployed systems rather than direct chip purchases, highlighting the complexity of comparing Google’s TPU revenue with Nvidia’s chip sales.

Bloomberg Intelligence estimates that Google’s TPU-related sales could climb from $12 billion in 2026 to over $135 billion by 2028, but these figures are projections and have not been independently verified. Alphabet does not disclose standalone TPU revenue or product-level margins.

The new chips build on Google’s existing eighth-generation TPU lineup, which includes the TPU 8t for training and TPU 8i for inference. Management has cited improvements of up to 2.7 times in training performance and 80% in inference price-performance versus the previous generation, though these gains are not directly comparable to Nvidia’s products.

Competitive Landscape and Challenges

Google’s push into external TPU sales puts it in more direct competition with Nvidia, which has long dominated the AI accelerator market. While TPUs offer cost advantages for certain workloads, Nvidia’s GPUs remain more versatile and benefit from a mature software ecosystem. SemiAnalysis’s testing found that Nvidia’s FP4 GPUs delivered more tokens per dollar at most tested speeds in certain comparisons, underscoring that workload, numerical precision, and software optimization matter.

The broader economic constraints are advanced-chip availability and data-center power. Google must also manage hefty capital expenditures—$51.4 billion in the latest quarter—and negative free cash flow of $9.1 billion as it builds inventory ahead of system sales. Inventory rose to $13.3 billion at September 30 from $2.4 billion at year-end 2025, tying up cash even as demand increased.

Regulatory and geopolitical factors also loom. U.S. export controls on advanced computing to China and scrutiny of third-country intermediaries influence where accelerators can be sold. The available reporting does not specify export-license requirements for the new TPU variants, so it would be premature to assume they face the same restrictions as Nvidia’s products.

Google’s expansion is part of a larger infrastructure race. Combined 2026 capital-expenditure plans for Amazon (NASDAQ:AMZN), Microsoft (NASDAQ:MSFT), Alphabet, and Meta are at least $720 billion, according to recent reporting. Against this backdrop, Google’s TPU strategy aims to provide customers with another large-scale computing platform and reduce dependence on a single accelerator supplier.

A Google spokesperson did not immediately respond to a request for comment on the new chip variants. An Nvidia spokesperson declined to comment.

Correction: An earlier version of this article misstated the timing of Google’s first recognition of TPU system sales. It was the second quarter of 2026, not the third.