Google hoards its TPU chips to not lose the race to AGI

What happened?

During the second quarter 2026 earnings presentation, Alphabet, Google's parent company, revealed that it is prioritizing internal use of its tensor processing units (TPUs) to advance toward artificial general intelligence (AGI). CEO Sundar Pichai responded to questions from Goldman Sachs and Bernstein analysts, stating that the first priority is to allocate the necessary computing capacity to compete on the frontier of AGI development, and only then distribute the rest among Search, YouTube, Google Cloud, and other areas.

The news, initially reported by The Register and picked up by Xataka, highlights that Google is willing to sacrifice immediate revenue from its cloud business to secure an edge in the race to AGI. According to Alphabet's earnings report, the company generated $12.3 billion in Google Cloud revenue in the second quarter, up 28% year-over-year, but growth may have been limited by the decision to hoard TPUs. Pichai emphasized that AGI is the foundation of everything, indicating the company is willing to divert resources even from its most profitable segments.

Why is this important?

Historically, the cloud business has been based on selling infrastructure to third parties. That Google decides to serve itself first indicates that AGI is not just a marketing goal but a real strategic priority. This has far-reaching implications:

• For Google Cloud customers: They could face restrictions on TPU availability, slowing their AI projects or forcing them to seek alternatives on AWS or Azure. Companies like Character.AI or generative AI startups that relied on Google's TPUs might be forced to migrate to NVIDIA H100/B200, increasing their costs and development times.
• For the chip market: Google's decision could pressure other AI hardware makers, like NVIDIA, to ramp up production, and competitors like Amazon (with its Trainium chips) and Microsoft (with Maia) to adopt similar strategies. NVIDIA has already seen increased demand for its AI GPUs, and this news could accelerate adoption of TPU alternatives.
• For the AI industry: Concentrating computing resources in the hands of a few companies could delay decentralized innovation and increase dependence on big tech players. Recall that during the cloud boom of 2010-2020, AWS, Azure, and Google Cloud competed to offer the best infrastructure; now, Google's internal priority could reshape that balance.

Consequences and context

The race to AGI has intensified in recent years, with companies like OpenAI, DeepMind, and Anthropic competing to develop increasingly capable models. Google, with its vast TPU infrastructure, has a significant advantage, but the decision to hoard these chips could have side effects:

• Friction with customers: Companies that rely on Google Cloud to train their AI models could feel sidelined and seek other providers. For example, a startup planning to scale its language model on TPU v4 could face months of delays, as happened in 2023 when GPU demand outstripped supply.
• Regulation: Regulators could view this practice as anticompetitive, especially if Google uses its cloud dominance to favor its own AI products. The European Commission is already investigating similar practices in the cloud market, and the U.S. FTC has shown interest in the concentration of AI resources.
• Internal innovation: By concentrating resources on AGI, Google might neglect other areas of its business, such as improving its search engine or expanding YouTube. However, Pichai argues that AGI will enhance all products, though in the short term there could be a slowdown in incremental improvements.

'Our first priority is to make sure we are allocating what we need to compete on the frontier of AGI development. That is the foundation of everything we do,' said Sundar Pichai.

The Bernstein analyst also asked about the impact on cloud revenue. Pichai avoided giving specific figures but acknowledged that internal allocation could reduce capacity available for external customers. The Register notes that Google has been increasing its TPU capacity, but internal demand is growing faster. In the past, Google has already prioritized its own projects, such as when it reallocated TPU resources to train PaLM in 2022, leaving some customers with less capacity.

What should readers know?

This news affects not only investors and Google Cloud customers but anyone interested in the future of AI. Google's decision underscores that AGI is seen as an existential goal for big tech, and they are willing to take drastic measures to get there first. For developers and startups relying on Google's infrastructure, it's time to diversify providers. For investors, the signal is that Google prioritizes the long term over quarterly revenue, which could affect stock price in the short term but strengthen its position in the long term. And for the general public, it's a reminder that the race to AGI is being fought with increasingly concentrated resources, which could define the balance of technological power in the coming decades. As with the space race of the 20th century, concentrating resources in a few players can accelerate progress but also create dependencies and geopolitical risks.