Google is quietly dismantling what many call the "NVIDIA tax" by selling its custom-built TPU chips to select external customers, marking a fundamental shift in how AI infrastructure gets deployed. While NVIDIA continues to dominate the merchant GPU market with record-breaking revenue, Google's vertical integration of silicon, AI models, and cloud services is positioning the company as a credible alternative for enterprises looking to reduce their dependence on NVIDIA hardware.

What Are TPUs and Why Do They Matter?

TPUs, or Tensor Processing Units, are custom chips that Google designed specifically for running AI workloads. Unlike NVIDIA's GPUs, which are general-purpose processors that work across many applications, TPUs are optimized for the exact type of math that powers large language models and other AI systems. Google's latest generation, called Ironwood, enables native FP8 training and inference, which is technical shorthand for running AI models more efficiently without sacrificing accuracy.

The practical implication is significant: companies using Google's TPUs can sidestep what industry observers describe as the NVIDIA tax, a premium that enterprises pay for NVIDIA's dominant position in the AI chip market. Google confirmed that it is now offering TPU sales "to a select group of customers in their own data centers," a direct challenge to NVIDIA's installed base.

Google

How Do the Two Companies' Latest Financial Results Compare?

Both companies posted AI-heavy quarters, but from opposite ends of the silicon stack. NVIDIA reported Q1 FY2027 revenue of $81.61 billion, up 85.2% year-over-year, with its Data Center segment alone generating $75.25 billion in revenue, a 92% increase. The company's gross margin hit 75.0% on a non-GAAP basis, reflecting the extraordinary pricing power it holds in the AI infrastructure market.

Google's results tell a different story. Alphabet posted Q1 FY2026 revenue of $109.90 billion, up 21.79% year-over-year, with Cloud revenue jumping 63% to $20 billion. More importantly, Google's Cloud backlog nearly doubled sequentially to $462 billion, suggesting that customers are committing to long-term relationships with Google's full-stack offering.

"The fact that we own frontier models and own the silicon really helps us stay ahead of the curve," said Sundar Pichai, Google's chief executive.

Sundar Pichai, Chief Executive Officer at Alphabet

Google's eighth-generation TPUs claim 80% better performance per dollar on inference tasks compared to previous generations, a metric that directly addresses one of enterprises' biggest concerns: the total cost of ownership for AI infrastructure.

What Makes Google's Approach Different From NVIDIA's?

NVIDIA's strategy centers on ecosystem dominance. The company's CUDA software platform has become the de facto standard for AI development, creating a powerful moat around its hardware business. NVIDIA's Spectrum-X networking solution already annualizes over $8 billion in revenue and has added Google Cloud as a customer, further entrenching the company's position.

Google, by contrast, is pursuing what analysts call a "full-stack" strategy. Rather than selling just chips, Google offers:

• Custom Silicon: TPU chips optimized for AI workloads, now available to external customers in their own data centers
• Frontier AI Models: Gemini and other large language models that Google develops in-house, giving it direct insight into what hardware optimizations matter most
• Cloud Infrastructure: Google Cloud services that integrate seamlessly with TPUs, reducing the friction of deploying AI systems

This vertical integration means Google controls the entire customer experience from silicon to software. When a company buys a TPU from Google, they are not just getting a chip; they are getting access to Google's AI expertise, cloud infrastructure, and ongoing optimization work.

What Are the Key Risks and Uncertainties?

For Google, the critical test will be whether TPUs can escape what some call "the Google garden." The company has a history of building impressive technology that remains largely internal. If external TPU deployments fail to gain traction, Google's silicon advantage becomes a cost center rather than a revenue driver.

NVIDIA, meanwhile, faces its own headwinds. The company's free cash flow declined 46.63%, a sharp drop driven by massive capital expenditures to support the "largest infrastructure expansion in human history," as CEO Jensen Huang described it. Additionally, NVIDIA has lost access to the Chinese market, which represents roughly $50 billion in total addressable market, due to U.S. export restrictions.

"The largest infrastructure expansion in human history," declared Jensen Huang, describing NVIDIA's current trajectory.

Jensen Huang, Chief Executive Officer at NVIDIA

NVIDIA's Blackwell GPU, which the company is ramping at unprecedented speed, will face increasing competition from Google's TPUs as more enterprises gain access to them. The question is not whether Google can build competitive silicon; it clearly can. The question is whether Google can convince enterprises to trust a new supplier for such a critical piece of infrastructure.

How to Evaluate These Companies as AI Infrastructure Bets

• Monitor TPU Adoption Rates: Watch Google's quarterly earnings reports for updates on external TPU deployments and whether the company is converting its $462 billion Cloud backlog into recognized revenue from silicon sales
• Track Blackwell Yields and Demand: For NVIDIA, the critical metric is whether Blackwell 300 chips are shipping on schedule and whether customer bookings exceed the company's $119 billion in supply commitments
• Assess Margin Sustainability: NVIDIA's 75% gross margin is historically high; watch whether this compresses as competition intensifies or whether the company's ecosystem moat holds firm

Google trades at a price-to-earnings ratio of 16, while NVIDIA's market capitalization has reached $4.91 trillion. Google has gained 103.78% over the past year, yet still trades like a search utility despite owning the silicon, the AI models, and the cloud infrastructure. For investors seeking exposure to the AI infrastructure buildout, NVIDIA offers the cleanest expression through its dominant market position and $80 billion share buyback program. For those betting on disruption, Google's quiet vertical integration presents a more interesting risk-adjusted opportunity, assuming the company can successfully commercialize its TPU advantage outside its own data centers.