Training Gemini 4 is too expensive! Barclays estimates Google will have negative free cash flow for two consecutive years.

```

A stellar earnings report from Google's parent company Alphabet did not bring relief, but rather a bigger bill.

According to Zhuifeng Trading Desk, Barclays said in a July 22nd report that after Google released its second quarter 2026 results, Barclays sharply raised its 2027 capital expenditure forecast for Google to $350 billion, and further to $500 billion in 2028, both exceeding 100% of its operating cash flow during the same period. This means that Google will be in negative free cash flow for the next two years.

The core driver behind this surge in capital expenditure is Google’s accumulation of computing power for the large-scale training and operation of its Gemini 4 model. Barclays believes Google is currently renting external GPU computing power at a premium above the market rate to provide "transitional" computing support for some GCP clients, while freeing up its own computing resources for Gemini 4 training. This arrangement costs about $1 billion per month and will put pressure on cloud business profit margins in Q4 and Q1 2027.

According to a Wall Street News article, the latest earnings report shows Google's Q2 cloud revenue grew 82% year-over-year, cloud backlog orders increased by $50 billion in a single quarter to $514 billion, search ad revenue grew 17% year-over-year, and YouTube ad growth accelerated to 13%. However, in the earnings call, the dual inflation of operating costs and capital expenditure brought by the omnipresence of AI has become a core concern for the market. CFO Ruth Porat expects that AI infrastructure investment will continue to put pressure on the income statement and cash flows.

Cloud business accelerates, but cost pressures rise in tandem

Barclays attributes the biggest highlight this quarter to continued acceleration in Google Cloud. In Q2 2026, Google Cloud revenue reached $24.768 billion, up 82% year-over-year (estimated at 77% excluding TPU sales by Barclays), with operating profit margin rising to 36%. Cloud backlog orders jumped from $46.4 billion last quarter to $514 billion, a net increase of $50 billion in one quarter, reflecting strong demand for GCP from enterprise clients.

However, the rapid growth of the cloud business is accompanied by significant cost increases. The report notes that Google signed a GPU rental agreement with Colossus at around $1 billion per month to provide transitional computing support for certain GCP clients, expected to last about two quarters. Barclays believes that the real purpose of this setup is to free up Google’s own first-party computing power for large-scale Gemini 4 training.

Barclays finds the premium rent paid by Google "eye-popping”, but given the current urgency in compute supply and demand, and the temporary nature of this arrangement, the decision is somewhat justified. The report also notes that in theory, this layout helps lay the foundation for future load capacity of Google’s data center racks.

Gemini 4 training drives "explosive" growth in capital expenditure

Barclays clearly states in its report that this sharp upgrade in capital expenditure forecasts is directly tied to the training needs of the Gemini 4 large model. The report shows Barclays raising its 2027 capex forecast from the previous $256.1 billion to $350.2 billion, a 37% increase; the 2028 capex forecast is raised sharply from $327.4 billion to $500.3 billion, about a 53% increase.

According to Barclays' predictive model, Google’s 2026-2028 capital expenditure will reach about $200.5 billion, $350.2 billion, and $500.3 billion, respectively, exceeding operating cash flow for two consecutive years, and free cash flow will turn negative $6.5 billion and negative $135.2 billion, respectively.

The report also points out that Google executives expressed clear confidence about Gemini 4 and the overall AI model development rhythm during the earnings call. Currently, Gemini API calls have reached 22 billion tokens per minute, a sharp increase from 16 billion the previous quarter and up three to four times year-over-year. Barclays estimates its market position is now tied with Anthropic for second place.

Full-stack AI deployment creates competitive barriers, but multi-front operations are costly

Barclays defines Google as "currently the world’s most advanced company in AI", with AI deployment stretching across the application layer (Search, Gemini, etc.), model layer (Gemini, Flash, etc.), infrastructure layer (GCP), and fundamental compute layer (TPU), forming a complete full-stack coverage. Barclays believes that this breadth will become increasingly prominent in subsequent quarters as the cloud business continues to accelerate.

However, the cost of this all-out competition is also significant. Barclays’ report notes that Google faces competition from Anthropic and OpenAI at the model layer, pressure from AWS and Azure at the infrastructure layer, and also needs to keep 13 application products with more than 1 billion monthly active users running smoothly. This “multi-front operation” setup is fundamentally driving the simultaneous expansion of both operating and capital expenditures.

In terms of earnings forecast, Barclays raised its 2027 revenue and operating profit estimates by 5%, mainly reflecting the upward revision in cloud revenue and profit. However, due to the Colossus agreement, Q4 2026 and Q1 2027 EPS expectations have been revised down, with subsequent period EPS forecasts correspondingly raised.

Barclays raised Google’s target price from $405 to $425, with valuation based on an average of 25x expected EPS and 15x EBITDA multiples for 2027-2028, maintaining an “Overweight” rating.

From a longer-term perspective, Barclays believes that, in terms of capturing value across the AI value chain, Alphabet has few comparable peers at present, and the strategic value of its cross-layer layout will gradually be unleashed as AI demand continues to expand.

~~~~~~~~~~~~~~~~~~~~~~~~

The above excellent content comes from Zhuifeng Trading Desk.

For more detailed interpretation, including real-time commentary and frontline research, please join 【Zhuifeng Trading Desk · Annual Membership】

Risk Warning and DisclaimerMarkets entail risks and investments must be made cautiously. This article does not constitute personal investment advice, nor has it taken into account the particular investment objectives, financial status, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article suit their specific circumstances. Investing based on this is at your own risk. ```