Google Pushes to Double Capacity as AI Demand Outpaces Supply

At a recent company-wide meeting, Google executives laid out the scale of the challenge facing the company’s artificial intelligence infrastructure.

According to Amin Vahdat, Google’s Vice President of Machine Learning, Systems, and Cloud AI, the company needs to double its serving capacity every six months to meet rising demand. A slide from his presentation stated that Google must grow its AI compute infrastructure by a factor of 1,000 over the next four to five years.

Sundar Pichai and CFO Anat Ashkenazi addressed concerns about whether current AI investments are sustainable, especially amid growing market skepticism over a potential bubble.

Pichai acknowledged the risks but defended the company’s level of investment, saying that a failure to scale up infrastructure quickly enough could be more damaging than short-term financial strain.

Why It Matters: Google is facing mounting pressure to expand its AI capabilities in line with user demand, but the cost of doing so is enormous. As investors start questioning the long-term returns of these investments, the company must walk a fine line between growth and financial stability. The decisions made now will likely define how Google competes in the AI market for years to come.

• AI Compute Must Double Twice a Year to Keep Up: Amin Vahdat told employees that Google must double its AI serving capacity every six months, driven by the increasing load of AI applications and user demand. He showed a slide projecting a 1,000x increase in compute needs over four to five years. This growth is in part about ensuring that infrastructure can scale in power and efficiency to match rising expectations without causing unacceptable increases in cost or energy usage.

• Custom Hardware and Efficiency Are Critical to Scaling: To meet compute goals without runaway costs, Google is relying heavily on its in-house hardware developments and model optimizations. Vahdat pointed to the public launch of the Ironwood TPU, Google’s seventh-generation Tensor Processing Unit, which is reportedly nearly 30 times more power efficient than the first version from 2018. The company is not only expanding infrastructure but also improving the performance-per-watt of the technology it deploys, allowing for larger workloads without equivalent increases in energy or capital expenditures.

• Capital Spending Rises to Match Infrastructure Demands: Alphabet raised its capital expenditures forecast for 2025 to a range of $91 to $93 billion, the second increase this year. The company is also preparing for an even higher level of spending in 2026. This increase comes in parallel with similar moves from Microsoft, Amazon, and Meta. Altogether, the four companies are expected to invest over $380 billion in infrastructure this year, showing how heavily AI growth is tied to physical compute expansion across the cloud industry.

• Employees Voice Concern About AI Market Viability: During the meeting, an employee asked about how Google plans to maintain profitability if the AI market fails to meet long-term expectations. Pichai said he is aware of the broader concerns around a possible AI bubble, and acknowledged that the conversation is active in Silicon Valley and on Wall Street. However, he argued that the company’s strong financial base gives it more room than competitors to absorb potential downturns, and that holding back on spending now could lead to lost opportunities later.

• Current Compute Supply Is Already a Limiting Factor: Pichai gave a concrete example of Google’s compute shortages, pointing to the recent launch of its Veo video generation tool. He said that while the tool was well-received, the company wasn’t able to provide it to as many users as it wanted because of infrastructure constraints. In his view, this shows that demand for advanced AI services is already exceeding what the company can deliver, and expanding compute capacity is the only way to meet that need and grow adoption.

Go Deeper -> Google must double AI serving capacity every 6 months to meet demand, AI infrastructure boss tells employees – CNBC