MMOOZZGG.com

AI investment

Google AI investment: Alphabet’s infrastructure, stakes and deals

6 min read9 sources

Google AI investment in 2026 goes mostly into data centers and its own chips: in the second quarter alone, Alphabet spent $44.9 billion on equipment and buildings. Below, based on company reports and media reports, we show how much it invests, which stakes it holds and which deals it closed in 2026.

On this page · 5 sections

How much it invests

Alphabet’s capital expenditure, that is, money for equipment, buildings and data centers, reached $80.6 billion in the first half of 2026. Of this, $44.9 billion fell in the second quarter, the company said in its report for that quarter.

Sources give the forecast for the whole of 2026 differently. We show both versions with dates and do not merge them into one figure:

1. On 22 July 2026 Alphabet raised its capital expenditure forecast to $195–205 billion, according to Bloomberg. The previous forecast, by the same source, was up to $190 billion.

2. In October 2026 Forbes gave $180–190 billion of 2026 capex for Alphabet.

When you use one of these figures in your own calculation or presentation, state the date and the source. The check date of all numbers is at the bottom of the page next to the links.

For the scale of the market: by Forbes’ estimate, the four largest companies, Amazon, Microsoft, Alphabet and Meta, will together spend $700–725 billion in 2026 against $410 billion in 2025, about +75%.

Alphabet raised part of the money for this spending on the capital market. In June 2026 the company placed shares and convertible preferred shares and received $49.6 billion net; among the purposes it named capital expenditure on AI infrastructure and computing capacity.

The signal for a finance director: even a company of this size raises outside money for AI construction and does not get by on its own funds alone.

Into what: infrastructure and models

Alphabet’s main spending item is data centers and its own AI chips, TPUs, Google’s tensor processors for training and running models. On 22 April 2026 Google presented the eighth generation of these chips.

The generation is split into two chips for different tasks:

1. TPU 8t for training models. One superpod links 9,600 chips and delivers 121 exaflops.

2. TPU 8i for running finished models. According to Google, it gives 80% better performance per dollar than the previous generation.

Both chips, the company said, will become generally available by the end of 2026. According to Google’s technical description, the new Virgo network makes it possible to build a training cluster of more than 1 million TPU chips.

Google also rents its chips to outside customers through the cloud. By an estimate quoted in an industry review, Anthropic bought from Google Cloud about a million TPUs.

Splitting into a chip for training and a chip for running shows where the money is moving. A model needs to be trained once, but customers will pay for its answers every day, so Google makes the cost of running a model per dollar a separate feature of the chip.

For a company that buys an AI solution, the conclusion is the same: the budget must cover not only the launch cost but also the monthly cost of running the model.

Stakes in other companies

Alphabet’s main known stake in another AI company is Anthropic, a developer of AI models. In 2026 Alphabet is increasing it.

On 24 April 2026 Alphabet announced that it will invest up to $40 billion in Anthropic, Reuters reported. The company gives $10 billion in cash at once at an Anthropic valuation of $350 billion, and another $30 billion depends on whether Anthropic meets its targets.

In May 2026 Anthropic raised $65 billion at a valuation of $965 billion, according to Bloomberg Law. Google put several billion dollars into this round under an earlier commitment; the lead investors were Altimeter Capital, Dragoneer, Greenoaks and Sequoia Capital.

The valuations of $350 billion and $965 billion come from different deals and different sources. We give them as they are, with dates, and do not recalculate the value of Alphabet’s stake.

The link here goes both ways. Alphabet gives Anthropic money and gets a stake, and Anthropic, by an estimate from an industry review, buys chips and capacity in Google Cloud.

The conditional part of the deal, $30 billion of $40 billion, shows the method of staged financing: most of the money comes only after targets are met.

Why big tech buys stakes in AI startups and what risks such schemes carry is covered on the page about corporate investment in startups.

Deals of 2026

Alphabet’s deals in 2026 by date:

1. 11 March: Google closed the purchase of Wiz, a developer of cloud security systems, for $32 billion in cash (TechCrunch, date confirmed by an Alphabet press release).

2. 24 April: Alphabet announced an investment of up to $40 billion in Anthropic, $10 billion of it at once (Reuters).

3. May: Google put several billion dollars into Anthropic’s $65 billion round under an earlier commitment (Bloomberg Law).

4. June: Alphabet placed shares and convertible preferred shares and received $49.6 billion net, including for AI infrastructure (company report).

The Wiz purchase closed a year after it was announced. Plan for this time frame in any acquisition: terms are fixed at the announcement, while money and control pass later.

Wiz does not develop models; it protects cloud systems. The $32 billion purchase shows that Google invests not only in computing but also in the security of the cloud through which it sells its AI capacity.

All large deals of the year, with sums, dates and sources, are collected in the deal tracker.

What a Russian company can learn

By Forbes’ estimate, the four largest companies will spend $700–725 billion on capital expenditure in 2026. A Russian company’s budget is on another scale, but Alphabet’s methods work for any sum:

1. Keep your AI budget as a range and review it every quarter. In July 2026 Alphabet raised its capex forecast to $195–205 billion, though the previous one was up to $190 billion.

2. Count the cost of running the model, not only the launch. Google released a separate chip for running finished models: according to the company, it gives 80% better performance per dollar than the previous generation.

3. Pay for results in stages. Of the $40 billion for Anthropic, Alphabet promised $30 billion only after targets are met, and the same principle works in a contract for an AI solution of any size.

4. Check your dependence on the vendor. If a contractor sells you both the solution and the computing, find out whether you can move the solution to another vendor without extra payment.

Start not with buying capacity but with finding the processes where AI will return money. The AI solution map does this: it shows where the company loses money, which solutions close those gaps and what they will bring in money.

Sources and fact-checking

Checked by the editors on 14 October 2026.

  1. Alphabet, press release on Q2 2026 results (SEC, Form 8-K), 22 July 2026 (sec.gov)
  2. Reuters (Deborah Sophia, Zaheer Kachwala), reprinted by CP24, 24 April 2026 (cp24.com)
  3. Google, official blog, Amin Vahdat, 22 April 2026 (blog.google)
  4. Google Cloud, official blog, Diwakar Gupta, Sabastian Mugazambi, 22 April 2026 (cloud.google.com)
  5. Bloomberg News, Davey Alba, Julia Love (Bloomberg Law), 23 July 2026 (news.bloomberglaw.com)
  6. TechCrunch, Rebecca Bellan; date confirmed by Alphabet’s press release of 11 March 2026, 11 March 2026 (techcrunch.com)
  7. Bloomberg Law, 28 May 2026 (news.bloomberglaw.com)
  8. Yandex Cloud blog, 24 July 2026 (yandex.cloud)
  9. Forbes, Jason Kirsch, 7 October 2026 (forbes.com)

Found a mistake? Tell us

Publications

Mike Mozg

Agentic AI use cases at AT&T: nearly 500 processes now run on agents

AT&T no longer uses AI only to help staff work faster: it is rebuilding whole business processes around agents. SiliconANGLE reported on 14 November 2026 that nearly 500 agentic AI use cases at AT&T are already in production. Here is how the company rebuilds them, where a human still checks the agent, and what other large companies can take from it.

News4 min
Contact us