Key Notes
- Google adds two TPU systems aimed at sensitive data and demanding AI workloads.
- Analysts forecast TPU sales could exceed $135 billion in 2028, excluding Cloud rentals.
- Alphabet’s Cloud revenue doubled in the third quarter, while capital spending exceeded operating cash flow.
Google introduced two new artificial intelligence computing systems Monday, broadening its challenge to Nvidia (Nasdaq: NVDA) as it turns technology developed for its own services into a business selling hardware to outside customers.
The TPU 8tx and TPU 8i Max give parent Alphabet (Nasdaq: GOOGL) more ways to compete for the budgets flowing into AI infrastructure. For shareholders, the opportunity extends beyond renting computing power: Google can also earn revenue from systems installed in customers’ own data centers.
An Ambitious Sales Forecast
Google says its TPU business now serves nearly 100 customers, including AI laboratories, trading firms, retailers and universities.
Bloomberg Intelligence estimates cited by Yahoo Finance put TPU sales at about $12 billion in 2026 and more than $135 billion by 2028, excluding chips rented through Google Cloud. That would be an increase of more than elevenfold in two years. These are analyst projections, rather than Alphabet guidance or a separately reported revenue figure.
The forecast suggests a steeper growth curve than Nvidia’s data-center business achieved when revenue rose from about $15 billion in fiscal 2023 to $115 billion in fiscal 2025. The comparison has limits: Nvidia’s figures are historical results for an entire segment, while the Google estimates concern future TPU sales.
Security and Memory Drive the New Designs
TPU 8tx targets governments and regulated businesses that need to keep sensitive information protected while AI models process it. TPU 8i Max combines two chips in one package, expanding computing power and memory bandwidth for models handling large amounts of information at once.
Both are new configurations of Google’s existing eighth-generation silicon. Yahoo Finance reported that the systems will ship in 2027. Google is adapting the equipment around its chips to reach different buyers, without waiting for an entirely new semiconductor design.
Cloud Growth Meets Heavy Spending
Alphabet’s third-quarter results already show the scale of demand. Google Cloud revenue rose 103% to $30.71 billion, while its operating income reached $11.12 billion. Alphabet described TPU system sales as a meaningful contributor to growth, but did not disclose their revenue separately.
The expansion is expensive. Alphabet’s earnings release shows $51.36 billion of capital expenditures against $42.29 billion of operating cash flow in the quarter, leaving free cash flow at negative $9.07 billion. Even with Cloud growing rapidly, the group spent more on property and equipment than its businesses generated in operating cash.
Software Opens the Door to More Buyers
Google is also working to reduce the effort required to move AI workloads onto its hardware. Its open-source TorchTPU software, released in October, lets developers run existing PyTorch workloads on TPUs with minimal code changes, according to Google Cloud.
That matters to procurement decisions. A cheaper processor can lose its appeal if adopting it requires a costly software overhaul. Making familiar tools work across different chips gives customers more room to negotiate and choose suppliers. For Alphabet, repeat orders from those customers would offer stronger evidence of a durable hardware business than another performance benchmark.
