MMOOZZGG.com

AI investment

AI investment data: sources and how we check the numbers

5 min read4 sources

AI investment data in our guide comes from several sources: the quarterly reports of Crunchbase, PitchBook and KPMG, the Stanford AI Index, Gartner forecasts, McKinsey surveys, Venture Guide by the Moscow Venture Fund and company reports. Their numbers differ: for the first half of 2026, PitchBook counted $407 billion invested in AI startups, while KPMG counted $560.3 billion for the whole venture market. Below: how we choose a source, check a figure and update pages.

On this page · 5 sections

Which sources we use

World, venture by quarter: Crunchbase News. The latest quarterly report came out on 5 October 2026; by it, startups worldwide received $159 billion in Q3.

World, venture by half-year: PitchBook and KPMG Private Enterprise. We take their figures from the report or from a business outlet’s article that cites the report, for example Kommersant.

The whole year: Stanford HAI, AI Index. The 2026 report gives $581.69 billion of corporate AI investment for 2025.

Forecasts and surveys: Gartner on AI spending, McKinsey on how companies get a return from AI.

Russia: Venture Guide by the Moscow Venture Fund, business outlets (Kommersant, ComNews, Vedomosti, RBC, Interfax, TASS) and studies by analytics firms. Example: AI and machine learning received $32 million in 33 deals in the first 9 months of 2026, according to Venture Guide.

Companies: annual and quarterly reports, press releases and news of closed rounds. We take a company’s valuation only from a source that states it directly.

Blogs, social media and retellings without a link to a study do not enter the guide.

How we check figures

1. We open the primary source, or an article that cites it, and find the figure in the text. If we did not open the source, the figure is not in the guide.

2. We record the figure with its source, link, publication date and date of checking.

3. We keep the wording without changing the meaning. If the source says “about”, “preliminary” or “by estimate”, so do we.

4. We name who counted. If the figure came through a retelling, we name both: for example, “Kommersant, citing PitchBook” for the half-year’s $407 billion.

An example of a caveat: the Venture Guide data for the first 9 months of 2026 is preliminary, as of 4 October 2026. We show it with this note and replace it when the final data comes out.

We rarely make our own calculations (shares, sums, comparisons in multiples), and only when they follow directly from the source’s figure.

How checking works in our own studies is described in our research method.

Why figures from different sources differ

The first difference: what counts as an AI company. PitchBook counted $407 billion invested in AI startups in the first half of 2026. In its quarterly reports, Crunchbase tags companies as AI by its own labels, and its AI share rests on a different list of companies.

Each database decides itself who counts as AI: only model developers or everyone with AI in the product, only startups or also large private companies. So you cannot extend the PitchBook sum with Crunchbase figures to get a year. We give the Crunchbase AI figures separately, in our analysis of Q3.

The second difference: the whole venture market. KPMG counted $560.3 billion for the first half of 2026, with $227.4 billion of it in Q2. For the same Q2, Crunchbase gives $212 billion.

The reasons are usual for venture statistics: different deal databases, different rules on when a deal enters a quarter (at announcement or at closing), and different treatment of deals where part of the money comes later when conditions are met.

The third difference: corporate investment. The Stanford AI Index counts not only rounds but also company acquisitions, so its $581.69 billion for 2025 is not compared directly with venture sums.

Our rule: we compare figures only within one source and always label whose count it is. We follow the trend in one database, for example in the Crunchbase reports: $104 billion in Q3 2025 and $159 billion in Q3 2026.

How often we update

Once a month, in the first week. We add new deals to the tracker and new figures to the statistics, and publish the monthly review.

Quarterly reports change the figure tiles on the guide’s pages. The Crunchbase report for Q3 came out on 5 October 2026, and the next one will come after the end of Q4.

The Stanford report comes out once a year: the 2026 report with figures for 2025 came out in April.

We do not edit an old figure. A new figure gets a new row, the pages switch to it, and the checking date at the bottom of each page shows when we last checked the numbers.

To get the changes without checking the site, subscribe to our monthly review.

How to report an error

Contact us: tell us which page, which figure and where you saw a different one. A link to the source speeds up the check.

We will open the primary source, fix the figure if the error is confirmed and record a new checking date.

If the source itself updated the data, as happens with the preliminary Venture Guide figures as of 4 October 2026, we add the new figure as a separate row and switch the pages to it.

All figures of the section, with sources and dates, are on the statistics page, and the whole section starts from the guide’s main page.

Sources and fact-checking

Checked by the editors on 14 October 2026.

  1. Crunchbase News, Gené Teare, 5 October 2026 (news.crunchbase.com)
  2. Kommersant, citing PitchBook, Venera Petrova, 13 August 2026 (kommersant.ru)
  3. Stanford HAI, AI Index 2026, chapter “Economy”, 13 April 2026 (hai.stanford.edu)
  4. ComNews, citing Venture Guide (Moscow Venture Fund), 5 October 2026 (comnews.ru)

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