Google Cloud CEO Thomas Kurian made a striking claim this week about the scale of Google’s custom AI chip business, telling investors at a Goldman Sachs conference that the company’s Tensor Processing Unit (TPU) operation is “more than twice” the size of the next-largest hyperscaler’s comparable business.
The statement, made Tuesday at the Goldman Sachs Communacopia + Technology Conference, offers a rare public glimpse into how Google views its competitive position in the rapidly expanding market for AI accelerators—the specialized chips that power artificial intelligence workloads.
For investors watching Google stock, the claim raises important questions: How significant is Google’s TPU business really? Can it meaningfully contribute to Alphabet’s revenue and profits? And does it pose a credible challenge to Nvidia’s dominance in AI hardware?
This analysis breaks down what Kurian said, what it means for Google’s cloud and AI strategy, and what investors should consider as they evaluate Alphabet’s position in the AI infrastructure race.
Thomas Kurian, who has led Google Cloud since 2019, spoke at length about Google’s AI infrastructure strategy during his appearance at the Goldman Sachs conference on September 8.
The key claim:
Kurian stated that Google’s accelerator business—specifically its TPU business—is “more than twice the size of the next-largest hyperscaler.” Investor’s Business Daily, which reported on the remarks, interpreted this as a reference to Amazon’s custom AI chip business, Trainium, and Microsoft’s Maia chips.
Three monetization models:
Kurian explained that Google generates revenue from TPUs through three distinct channels:
Investment payback:
According to the transcript cited in reporting, Kurian said Google’s overall AI server investment payback period is less than two years, while systems using Google’s proprietary chips have a payback period of approximately half that time—less than one year.
Performance claims:
Kurian stated that Google’s chips offer approximately 2.7 times better price-performance in training compared to competitors, 80% better in inference, and 30% better in CPU performance.
Important context: These are Kurian’s statements about Google’s own business. The specific comparison to competitors has not been independently verified by third-party market research firms.
Tensor Processing Units, or TPUs, are custom-designed chips that Google developed specifically for machine learning and artificial intelligence workloads.
What is a TPU?
A TPU is an application-specific integrated circuit (ASIC) designed to accelerate the mathematical operations that underpin AI models. Google first began developing TPUs in 2015, and the chips have since gone through multiple generations.
Why Google developed TPUs:
Google created TPUs to handle the massive computational demands of its own AI services—including search, translation, and later, its Gemini AI models. By designing chips optimized for specific AI tasks, Google aimed to achieve better performance and efficiency than general-purpose hardware could provide.
How TPUs differ from CPUs and GPUs:
The role of TPUs in Google Cloud:
Google Cloud customers can access TPU computing power through the platform, allowing them to train and run AI models without purchasing their own hardware. This “compute-as-a-service” model has become increasingly important as AI workloads grow.
The latest generation:
Google unveiled its eighth-generation TPUs in May 2026, releasing “TPU 8t” for training workloads and “TPU 8i” for inference.
The market for AI accelerators has become one of the most strategically important segments in technology. Understanding why requires looking at how AI models are built and deployed.
AI model training:
Training large language models and other AI systems requires enormous computational resources. The process involves running massive datasets through neural networks repeatedly until the model learns to perform its intended task. This training phase is computationally intensive and time-consuming.
AI inference:
Once a model is trained, it must be deployed to respond to user queries—a process called inference. Inference workloads are growing rapidly as AI-powered products reach millions of users.
Data center infrastructure:
Both training and inference require specialized hardware housed in data centers. The efficiency of this hardware directly affects the cost of delivering AI services.
Cloud customers:
Enterprises that want to build AI applications need access to AI infrastructure. Cloud providers that offer competitive AI hardware can attract and retain these customers.
Google’s Gemini ecosystem:
Google’s TPUs power its own AI products, including Gemini. By controlling the hardware layer, Google can optimize performance and potentially reduce costs for its AI services.
Infrastructure economics:
If Google’s TPUs offer better price-performance than alternatives—as Kurian claims—this could translate into lower costs for Google and more competitive pricing for cloud customers.
Important caveat: Different AI workloads may favor different hardware. GPUs remain essential for many applications, particularly those requiring flexibility across diverse computing tasks.
Alphabet’s cloud division has become a significant growth driver for the company, and the latest quarterly results illustrate why investors are paying attention.
Q2 2026 results:
According to Alphabet’s second-quarter 2026 earnings report, Google Cloud generated $24.8 billion in revenue, representing 82% year-over-year growth. This marked the segment’s strongest growth in recent quarters, accelerating from 63% growth in Q1 2026 and 48% growth in Q4 2025.
Operating income:
Google Cloud operating income reached $8.8 billion in Q2 2026, up from $2.8 billion in the same period a year earlier—a 212% increase.
Operating margin:
The segment’s operating margin improved to 35.6%, up 14.9 percentage points year-over-year.
Backlog:
Google Cloud’s backlog—signed contracts for future revenue—reached $514 billion, crossing the $500 billion threshold for the first time.
AI infrastructure demand:
Alphabet management attributed the strong cloud growth to demand for AI infrastructure and enterprise AI solutions.
TPU revenue recognition:
In Q2 2026, Alphabet recognized TPU hardware sales revenue for the first time, having begun delivering TPU systems to customer data centers.
The cloud computing market is dominated by three players: Amazon Web Services (AWS), Microsoft Azure, and Google Cloud. Each has developed custom AI chips to complement offerings from Nvidia.
Market position comparison:
| Provider | Q2 2026 Revenue | Year-over-Year Growth | Market Share |
|---|---|---|---|
| AWS | $42.2 billion | 37% | 28% |
| Microsoft Intelligent Cloud | $39.3 billion | 32% | 20% |
| Google Cloud | $24.8 billion | 82% | ~13% |
Custom AI chips:
Kurian’s comparison:
Kurian’s claim that Google’s TPU business is “more than twice” the size of the next-largest hyperscaler’s comparable business would suggest Google leads Amazon and Microsoft in custom AI chip sales. However, this comparison is based on Google’s characterization and has not been independently verified.
Context on market share:
AWS remains the largest cloud provider by revenue and market share, followed by Microsoft and then Google Cloud. Google Cloud’s faster growth rate reflects its smaller starting base and strong AI demand.
| Factor | Google TPU | Nvidia GPU |
|---|---|---|
| Main purpose | AI/ML acceleration | Broad accelerated computing and AI |
| Provider | Google (internal and external) | Nvidia (sold to multiple customers) |
| Cloud availability | Google Cloud | Multiple cloud providers |
| Key strength | Optimized for Google’s AI stack | Broad ecosystem, flexibility, developer familiarity |
| Market position | Growing; primarily internal and select external sales | Dominant market leader |
Why both can coexist:
Nvidia’s GPUs are used across virtually all cloud providers and by companies building their own AI infrastructure. Nvidia reported Q2 2026 data center revenue of $89 billion, up 117% year-over-year. This scale dwarfs Google’s TPU business, even by Kurian’s characterization.
TPUs are optimized for specific workloads and Google’s own infrastructure. GPUs offer greater flexibility and are supported by a broader software ecosystem. Many organizations use both types of hardware depending on the task.
No universal winner:
The AI hardware market is not a zero-sum competition. Demand for AI computing is growing rapidly enough that multiple hardware providers can succeed simultaneously.
For investors evaluating Google stock (ticker: GOOGL), the TPU development is one factor among many that could influence the company’s financial performance.
Potential positives:
Risks to consider:
For investors, the development could be viewed as: A signal that Google is positioning itself competitively in AI infrastructure, with the potential for cloud growth to offset heavy capital expenditures. However, the path from AI investment to shareholder returns involves numerous variables that cannot be predicted with certainty.
Alphabet’s AI ambitions require enormous capital investment, and the company has significantly increased its spending plans.
2026 capital expenditure guidance:
Alphabet raised its 2026 capex guidance to $195–205 billion, up from $180–190 billion previously. This represents a substantial increase from prior years.
Q2 2026 capital expenditure:
The company spent $44.9 billion on capital expenditures in Q2 2026 alone.
Allocation:
Of the Q2 capex, 60% was allocated to servers and 40% to data centers and networking equipment.
CFO commentary:
Alphabet CFO Anat Ashkenazi stated during the company’s earnings call that “demand still outpaces that investment,” indicating that the company would continue spending to meet customer demand. She also noted that capital expenditure is expected to increase significantly in 2027.
Financial pressure:
Alphabet’s total debt has risen from $16 billion to approximately $100 billion as the company funds its infrastructure buildout.
The relationship between AI opportunity and AI costs:
Strong AI demand does not automatically translate into higher profits. Infrastructure investment is expensive, and the returns depend on factors including:
Google’s TPU development has prompted questions about whether it could challenge Nvidia’s dominance in AI chips.
Why Google building its own accelerators matters:
By designing custom silicon, Google can optimize hardware for its specific AI workloads and potentially reduce dependence on external suppliers. This gives Google more control over its infrastructure costs and roadmap.
Why cloud providers want custom silicon:
Cloud providers including Google, Amazon, and Microsoft have all developed custom AI chips. The motivations include:
Why Nvidia remains important:
Nvidia’s GPUs remain the industry standard for AI computing. Key factors include:
Why Google TPUs do not automatically eliminate Nvidia GPUs:
How the market could evolve:
The AI hardware landscape is likely to include a mix of GPUs and custom accelerators. Different customers will make different choices based on their specific needs, existing infrastructure, and strategic priorities.
The potential pathway from AI investment to shareholder value involves several steps, each with uncertainties.
The chain:
AI investment → Cloud demand → TPU usage → Cloud revenue → Operating income → Alphabet financial performance
Where the chain could break:
Analyst perspective:
Morgan Stanley estimated in a report that Google Cloud’s first-party TPU revenue could reach $84 billion next year and $108 billion by 2028. These projections reflect one analyst’s view and are not guarantees of future performance.
For investors monitoring Google stock, several metrics and developments will provide insight into the company’s AI progress.
Next Alphabet earnings report:
The Q3 2026 earnings report will provide updated cloud revenue, margin, and capex figures.
Google Cloud revenue growth:
The 82% growth rate in Q2 was exceptional. Sustainability of this growth will be closely watched.
Cloud operating margin:
Margin improvement reflects operating leverage as the cloud business scales.
Capital expenditure:
How much Alphabet invests in AI infrastructure—and whether returns justify the spending.
AI infrastructure demand:
Commentary on customer demand and backlog trends.
TPU adoption:
Signs of external customer traction for TPU systems.
Gemini monetization:
Revenue from Google’s AI products and services.
Search advertising trends:
Any impact of AI on Google’s core search business.
YouTube advertising:
Another key revenue driver for Alphabet.
Regulatory developments:
Ongoing antitrust cases and potential remedies.
A balanced assessment of Alphabet’s investment case requires acknowledging the risks.
AI spending risk:
Alphabet may need to continue heavy spending to compete effectively. If AI returns disappoint, this could weigh on the stock.
Competition risk:
Microsoft, Amazon, and Nvidia remain major competitors with significant resources.
Regulatory risk:
Alphabet faces antitrust scrutiny in the U.S. and Europe. Potential outcomes include fines, business practice changes, or structural remedies.
Monetization risk:
Strong AI usage does not automatically translate into proportional revenue. Consumer AI products may be difficult to monetize directly.
Technology risk:
AI hardware and models are evolving rapidly. Architectural shifts could affect the value of current investments.
Valuation risk:
A strong company can still have a stock that performs poorly if expectations are already priced in.
Bull case:
Bear case:
Important note: This analysis does not provide a specific price target. Investors should conduct their own research and consider their individual circumstances before making investment decisions.
Google stock reflects investor sentiment about Alphabet’s businesses, including Google Cloud, advertising, and AI investments. Recent developments include Thomas Kurian’s comments about Google’s TPU business and strong Q2 2026 cloud results.
Kurian said Google’s TPU business is “more than twice the size of the next-largest hyperscaler.” He also described three monetization models: cloud leasing, direct sales, and joint venture sales.
A Tensor Processing Unit is a custom AI chip developed by Google for machine learning workloads. Google began developing TPUs in 2015 and is now in its eighth generation.
Yes, Google’s TPUs compete with Nvidia GPUs for AI workloads, but the market is large enough for multiple providers. Nvidia remains the dominant player.
Google claims better price-performance for TPUs in certain workloads. However, GPUs offer broader flexibility and a larger software ecosystem. The best choice depends on specific use cases.
Google Cloud generated $24.8 billion in Q2 2026 revenue, up 82% year-over-year, with operating income of $8.8 billion.
AI growth could potentially benefit Alphabet through cloud revenue growth and TPU sales. However, heavy capital expenditures create near-term financial pressure.
Key risks include AI spending, competition, regulatory pressure, monetization uncertainty, and valuation.
Yes, Google designs TPUs for internal use and external sales to cloud customers and enterprises.
Google is a major cloud provider offering both TPUs and Nvidia GPUs, with a growing position in custom AI chips.
This article does not provide personalized investment advice. Investors should evaluate factors including valuation, growth prospects, competitive position, and their own financial circumstances.