Google CEO Sundar Pichai has revealed that the company is reserving some of its limited computing capacity for artificial intelligence research, even when that means sacrificing potential short-term revenue from Google Cloud customers. The strategy reflects Google’s commitment to developing advanced AI models through its DeepMind research division as demand for computing power continues to outpace available infrastructure.
Speaking on the Big Technology Podcast with Alex Kantrowitz in October 2026, Pichai described access to computing resources for frontier AI development as a non-negotiable priority. He also warned that accommodating new customers seeking computing capacity in 2027 could be extremely difficult, highlighting the infrastructure constraints facing the technology industry.
Key Takeaways
- AI research comes first: Google is prioritizing the computing resources needed to develop advanced AI models.
- Cloud revenue trade-off: Pichai acknowledged that Google Cloud is giving up some potential short-term revenue to support internal AI research.
- Capacity shortage: New customers seeking computing resources for 2027 may face significant difficulties securing capacity.
- Strong cloud demand: Google Cloud generated $24.8 billion in revenue in Q2 2026, up 82% year over year, according to reported financial results.
- Large investment plans: Alphabet expects capital expenditure of $195 billion to $205 billion in 2026.
- AI-assisted research: Pichai expects increasingly capable AI systems to help researchers develop future generations of models.
Why Google Is Prioritizing AI Research
Google’s strategy centers on maintaining its position among the companies developing the most advanced AI models. These systems require substantial computing resources for training, experimentation, evaluation and deployment.
Pichai said ensuring Google DeepMind has the capacity needed to remain at the frontier of AI development is the starting point for allocating computing resources across the company.
In the podcast interview, he explained that Google’s different businesses still develop long-term plans based on demand forecasts and customer commitments. However, those plans operate around the company’s decision to reserve sufficient capacity for frontier AI research.
The decision reflects Google’s view that advanced AI is strategically important across its business, including Search, Cloud and other products. New model capabilities can eventually be incorporated into commercial services, potentially creating opportunities beyond the immediate revenue generated by selling computing infrastructure.
This approach also comes with an opportunity cost. Computing capacity reserved for internal research cannot always be sold to external customers, particularly when the company is operating close to its available limits.
Google Cloud Is Giving Up Potential Short-Term Revenue
Pichai acknowledged that the trade-off is already affecting Google’s cloud business.
“Our cloud business is at any given moment in time foregoing short-term revenues because we are making compute trade-offs to support frontier AI development,” he said during the interview, as reported by Big Technology.
The statement is significant because Google Cloud is a major commercial business selling infrastructure, data services and AI tools to enterprises. Reserving more capacity for internal research can limit how much infrastructure remains available to meet additional external demand.
However, Pichai did not disclose the exact amount of revenue Google is sacrificing or quantify the financial impact of these allocation decisions.
The strategy should also not be interpreted as Google abandoning cloud customers. Pichai said the company continues to plan for its different businesses, taking customer demand and existing commitments into account. The trade-off concerns how available computing resources are distributed when demand exceeds capacity.
For Google, the underlying calculation is that maintaining a competitive position in advanced AI may create substantial long-term value, even if doing so limits some immediate cloud sales.
Why the AI Computing Shortage Is Difficult to Solve
The shortage is not simply a matter of purchasing additional chips. Building AI infrastructure requires a coordinated supply of processors, servers, networking equipment, electricity, cooling systems and data-center facilities.
These components must be procured and installed together, while new facilities can take years to plan and construct. As a result, technology companies cannot immediately increase capacity whenever demand rises unexpectedly.
Pichai explained that the computing resources available today reflect investment decisions made in 2024. That illustrates how infrastructure planning can struggle to keep pace with rapidly changing AI demand.
He also warned that serving customers who want additional computing capacity in 2027 could be particularly difficult.
| Challenge | Impact on AI infrastructure |
|---|---|
| Semiconductor supply | Limits how quickly companies can obtain suitable processors |
| Data-center construction | Adds long lead times before new capacity becomes operational |
| Electricity and cooling | Determines how much computing infrastructure facilities can support |
| Unexpected AI demand | Makes earlier capacity forecasts less reliable |
| Internal research needs | Creates competition between model development and commercial cloud workloads |
Pichai said Google had previously leased short-term computing capacity to help accommodate some customers. His comments suggest that expanding supply and managing allocation are both necessary to address the industry’s constraints.
Google Cloud Is Growing Despite Capacity Constraints
Google’s decision comes at a time when its cloud business is expanding rapidly.
According to figures reported by Moneycontrol, Google Cloud generated $24.8 billion in revenue in the second quarter of 2026, an increase of 82% year over year. Its contracted revenue backlog stood at approximately $514 billion at the end of June.
The backlog represents contracted business expected to be recognized as revenue over time; it is not the same as revenue already earned or cash immediately available to Google.
These figures demonstrate the scale of demand for Google’s cloud services. They also help explain why allocating capacity is difficult: the company must support existing commercial commitments while evaluating additional opportunities and maintaining sufficient resources for internal AI development.
Alphabet’s planned capital expenditure of $195 billion to $205 billion for 2026 further demonstrates the scale of its infrastructure investment. A substantial portion of that spending is directed toward AI and cloud infrastructure, although the precise allocation can vary.
Even such large investments cannot eliminate shortages overnight. The timing of new facilities, equipment deliveries and power availability determines when additional computing capacity becomes usable.
Gemini Demand Is Adding to the Pressure
Google’s own AI products are also consuming increasing amounts of computing power.
Pichai said Gemini 4 Argon processed nearly 25 trillion tokens during its first 72 hours, approximately three times the volume processed by Gemini 3 over the same period, according to NewsBytes’ report on the interview.
Token processing is one measure of how much information a model handles, although it does not independently reveal the model’s total computing costs or profitability.
The figures nevertheless illustrate the potential scale of demand when a new AI model attracts substantial usage. Google must provide computing resources not only to train and improve models but also to serve users who interact with them.
This creates a continuing allocation challenge. More capacity can support research, consumer products and enterprise services, but each additional workload competes for resources until infrastructure expands.
Could AI Help Google Build Better AI Models?
Pichai also discussed the possibility that increasingly capable AI systems could assist researchers in developing subsequent generations of models.
This process is often called recursive self-improvement. In practice, it can involve AI systems helping with coding, research, experiment design, evaluation and testing, potentially accelerating parts of the model-development process.
Pichai sees these capabilities as another reason to invest in frontier AI research. If AI systems become more effective at assisting researchers, the resulting productivity improvements could help Google develop future models and incorporate their capabilities into commercial products.
However, this does not mean AI systems can independently and reliably improve themselves without human oversight. Progress depends on the quality of research methods, available computing resources, evaluation procedures and the ability to verify proposed improvements.
The strategy therefore remains a long-term investment in research capabilities rather than a guarantee that additional computing capacity will automatically produce better models.
The Bigger Picture
Google’s decision highlights the growing tension between selling computing infrastructure and using it to develop proprietary AI technology. Cloud providers can earn revenue by renting computing power to customers, but advanced AI research may also create products and capabilities that strengthen their broader businesses.
For Google, the challenge is balancing both opportunities while infrastructure remains constrained. Prioritizing research could support future AI products, but it also risks frustrating customers who need capacity for their own applications. The financial outcome will depend on how quickly supply expands and how effectively Google converts its research investments into commercial value.
FAQs
1. Why is Google prioritizing AI research over cloud customers?
Sundar Pichai said Google must ensure its DeepMind research division has enough computing power to develop advanced AI models. The company is willing to forgo some potential short-term cloud revenue to support that objective.
2. How much revenue is Google Cloud generating?
Google Cloud generated $24.8 billion in revenue in Q2 2026, up 82% year over year, according to reported figures. Its contracted revenue backlog was approximately $514 billion at the end of June.
3. Why is computing capacity so difficult to expand?
AI infrastructure requires specialized chips, servers, networking equipment, electricity, cooling and data-center facilities. Procuring and deploying these resources takes time, making it difficult to respond immediately to rising demand.
4. Will Google Cloud customers face shortages in 2027?
Pichai warned that accommodating new customers seeking computing capacity in 2027 could be extremely difficult. This indicates continued pressure on supply, but it does not mean every existing customer will experience a shortage.
5. What is recursive self-improvement in AI?
Recursive self-improvement describes a process in which AI systems help researchers develop more capable AI systems. This may include assisting with coding, experiments and evaluations, although the extent of such improvements remains dependent on research progress and verification.
Looking Ahead
Google’s ability to expand its computing infrastructure will be central to how it balances AI research with commercial cloud demand. New data centers, improved hardware efficiency and better allocation of existing resources could gradually ease the constraints, but long construction timelines mean that additional supply will not arrive immediately.
Investors and enterprise customers will be watching how Google’s infrastructure spending translates into AI capabilities, cloud growth and long-term returns. Pichai’s comments make the company’s immediate priority clear: Google intends to preserve the computing capacity required to compete at the frontier of AI, even when doing so means giving up some revenue it could earn in the short term.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
