Google's TPU Yard Sale

Google's Chips vs. Nvidia's Margin

Google and Meta have long been fierce rivals in the world of digital advertising. They fight for every second of user attention and every cent of marketing budget. But in the basement of the internet—the data center—they are fast becoming roommates.

Last week, the Wall Street Journal reported that Meta is in talks to buy billions of dollars' worth of Google’s custom AI chips, known as Tensor Processing Units (TPUs). Here's the WSJ:

Meta Platforms is in talks to use chips made by Google in its artificial-intelligence efforts, a step toward diversifying away from its reliance on Nvidia, according to people familiar with the matter.

A deal could be worth billions of dollars, but the talks are continuing and may not result in one. It is still up in the air whether Meta would use the chips, known as tensor processing units or TPUs, to train its AI models or to do inference, one of the people said. Inference, the process a trained model uses to generate the response to a query, requires less computational power than training.

While the headline moved markets, the partnership isn't entirely out of the blue. It is the natural evolution of a relationship that was formalized in August, when Meta signed a six-year, $10 billion agreement to use Google Cloud’s general compute services.

The motivation for this deepening alliance is simple: both companies find Nvidia's pricing power inconvenient.

Nvidia has achieved a level of supply chain dominance that allows it to command gross margins that make software companies jealous. Its H100 and Blackwell GPUs are the reserve currency of the AI boom. For a company like Meta, which is planning to spend tens of billions on AI infrastructure, Nvidia's monopoly is a structural risk to the balance sheet.

This creates an alignment of incentives. For Google, the logic of selling chips to its ad rival is simple: it feeds the new golden goose. Google Cloud is now the fastest-growing segment of Alphabet, with revenue jumping 34% year-over-year to $15.1 billion in Q3. Perhaps more importantly, its backlog—the money customers have contractually committed to spend—surged 46% quarter-over-quarter to $155 billion. Google has realized that hoarding its secret weapon is far less profitable than renting it out to a market desperate for alternatives.

And so, Google is opening the armory to the public.

For years, the TPU was Google's internal advantage. It was an Application-Specific Integrated Circuit (ASIC) designed solely to make Google Search, Waymo, and DeepMind run faster and cheaper than the competition. Google treated these chips like a state secret, keeping them locked inside its own data centers. After a decade of hoarding its custom silicon, Google is now holding the world's most advanced garage sale.

The shift to selling them externally began in earnest last month, when Anthropic—a company Google has invested in—announced it would buy up to one million TPUs. Now, by courting Meta, Google is positioning its custom silicon as the industry-standard alternative to Nvidia.

This creates a world where the AI models—Google's Gemini, Meta's Llama, Anthropic's Claude—are still fierce competitors, but the underlying hardware is becoming a shared utility. It suggests that AI models are becoming a commodity, and the real moat is in owning the plumbing.

Google's pitch is one of brutal efficiency. Nvidia’s GPUs are generalists; they are designed to render video games, mine crypto, and train models. They can do anything, which means they carry silicon baggage. Google’s TPUs are specialists; they are designed to do exactly one thing: matrix math for neural networks. If you are Mark Zuckerberg, staring down a $40 billion capex bill, the prospect of a chip that is more energy-efficient and potentially cheaper than Nvidia’s offering is attractive enough to expand your relationship with your biggest competitor.

For Nvidia, whose stock dropped 7% on the news, this formation of a "Not-Nvidia" alliance is a threat. The company's revenue is incredibly top-heavy: last quarter, just two customers accounted for 39% of its total sales, and the top six accounted for 85%. This is the kind of customer concentration that gives a monopoly night sweats. If those whales decide they would rather trade chips with each other than pay Nvidia's margins, the math gets ugly fast.

Google is betting that its internal tools are robust enough to become a market standard and drive that $155 billion backlog even higher. Meta is betting that it can engineer its way around Nvidia's margins. It creates a pragmatic new world where the two biggest advertising rivals in history share a silicon supply chain. You can compete for ad dollars on the screen while sharing the same circuit board behind it.

More on Cloud Compute:

• Google Cloud wins new NATO contract for sovereign cloud services (Cloudtech)
• Why Are Epic Games and Other Customers Leaving Amazon for Google? (ARPU)
• Mapping the Neocloud Landscape (ARPU)

On Our Radar

Our Intelligence Desk connects the dots across functions—from GTM to Operations—and delivers intelligence tailored for specific roles. Learn more about our bespoke streams.

The Rise of Sovereign AI

• The Headline: A growing number of countries, including South Korea, France, India, and the UAE, are launching ambitious "sovereign AI" initiatives to build domestic AI infrastructure and capabilities, driven by a fear of becoming technologically dependent on the US and China. (WSJ)
• ARPU's Take: "Sovereign AI" is the new geopolitical buzzword, and it's being backed by billions of dollars. This isn't just about national pride; it's a strategic imperative for middle-power nations to avoid becoming digital colonies in an AI world dominated by the US and China. They are racing to build their own data centers, train their own models, and control their own data.
• The GTM Question: This trend carves out a powerful new customer segment for the tech industry: national governments. For vendors like Nvidia and the major cloud providers, this creates an imperative to develop a "sovereign-in-a-box" go-to-market strategy, offering everything from chips to in-country cloud infrastructure. This shifts the sales process from a corporate one to a geopolitical one, where winning a deal is as much about national security alignment as it is about technical performance.

Tesla's Competitive Crisis

• The Headline: Tesla is facing a significant and sustained sales decline in its key global markets, with European sales falling nearly 50% in October, as its aging product lineup struggles against a wave of new, lower-priced EV models from European and Chinese rivals. (Reuters)
• ARPU's Take: The data is stark: Tesla is losing market share while the overall EV market is growing. This isn't a temporary blip; it's a structural problem. The company's once-revolutionary product lineup has grown stale, and its brand is no longer enough to fend off a flood of newer, cheaper, and more varied competition from both legacy automakers and Chinese brands.
• The Product Question: This sales data confirms that the EV market has shifted to a mature, multi-polar competition where product freshness is paramount. For product leaders at Tesla, this creates an urgent imperative to move beyond incremental software updates and accelerate the development of new vehicle models to address the widening gaps in their portfolio that rivals are now aggressively exploiting.

P.S. Tracking these kinds of complex, cross-functional signals is what we do. If you have a specific intelligence challenge that goes beyond the headlines, get in touch to design your custom intelligence.

You received this message because you are subscribed to ARPU newsletter. If a friend forwarded you this message, sign up here to get it in your inbox.