State of AI Report
Last reviewed
Jun 8, 2026
Sources
9 citations
Review status
Source-backed
Revision
v1 · 1,537 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 8, 2026
Sources
9 citations
Review status
Source-backed
Revision
v1 · 1,537 words
Add missing citations, update stale details, or suggest a clearer explanation.
The State of AI Report is a free, annual review of progress in artificial intelligence, published every year since 2018. It is compiled by the investor Nathan Benaich, founder and general partner of the London-based venture firm Air Street Capital, and was co-authored in its early years with the entrepreneur and investor Ian Hogarth. [1][2] The report is distributed as a slide-deck-style document, has grown to several hundred slides per edition, and synthesizes the year's most significant developments across AI research, industry, geopolitics, and safety. Its stated aim is "to trigger an informed conversation about the state of AI and its implication for the future." [2]
The report is best known for two recurring features: a set of explicit, falsifiable predictions about the year ahead, and a scorecard the following year that grades how many of the prior predictions came true. [1][3] It is accompanied by an online "Compute Index" that tracks the world's large computing clusters and the use of AI accelerator chips in published research. [4] The report is released under a Creative Commons Attribution 4.0 International license and is widely read by AI investors, researchers, operators, and policymakers. [2]
The first State of AI Report was published in 2018 by Nathan Benaich and Ian Hogarth, who described themselves as AI investors writing "a good old fashioned report" on the field. [5] At the time, the report covered topics such as research, talent, industry, and politics, in an era still dominated by reinforcement learning and narrow, task-specific systems rather than the large generative models that would later define the field. [1]
The report has been issued annually each autumn (typically October) since then. Hogarth co-authored the editions from 2018 onward, but stepped back from active production after he was appointed in 2023 to lead the UK Government's Frontier AI Taskforce (originally the Foundation Model Taskforce). That body became the UK AI Safety Institute in November 2023, with Hogarth as its founding chair; it was renamed the AI Security Institute in February 2025. [6][7] From the 2023 edition onward, the report has been produced by Benaich together with an Air Street Capital research team, described in the report itself as "a Team Air Street production," with input from external reviewers and companies that contribute data. [8]
To mark its fifth anniversary in 2023, the authors reviewed the cumulative archive of 934 slides produced across the report's history to that point and compiled the field's major storylines, illustrating how the report had ballooned in scope as AI accelerated. [8] By the 2025 edition, the eighth in the series, the report had become, in its own description, "the most widely read and trusted independent, annual report on global progress in artificial intelligence." [2]
| Aspect | Detail |
|---|---|
| First published | 2018 |
| Frequency | Annual (released in autumn, typically October) |
| Producers | Nathan Benaich and Air Street Capital |
| Original co-author | Ian Hogarth (2018 onward; stepped back after 2022) |
| 2025 edition | Eighth annual report |
| License | Creative Commons Attribution 4.0 International |
| Companion tool | State of AI Report Compute Index |
Modern editions of the report are organized into a consistent set of thematic sections, each summarizing the most important developments of the prior year. The core sections are: [1][2]
The 2025 edition added a large-scale practitioner Survey as a further section, described as one of the largest open-access surveys of its kind, gathering responses from more than 1,200 AI practitioners on how AI is used in work and daily life. The report stated that 95 percent of surveyed professionals used AI at work or at home. [2] Editions are typically reviewed before publication by working AI practitioners in industry and academia, and the deck is supplemented by a video walkthrough and live online survey results. [2]
A defining feature of the report is its set of year-ahead predictions, which the authors deliberately frame as concrete and checkable rather than vague. Past predictions have addressed topics such as forthcoming model releases and capabilities, the scale of AI funding rounds and valuations, the pace and shape of regulation, and developments in AI safety research and governance. [1][3]
The following year, each prediction is revisited in a scorecard that marks it as correct, incorrect, or partially correct, an unusual practice of public self-accountability that has become one of the report's signatures. [1][3] This "keep them honest" mechanism, as the authors put it, distinguishes the State of AI Report from many other industry surveys, which rarely audit their own forecasts. [1]
Alongside the main report, the team maintains the State of AI Report Compute Index, an online resource that tracks the global supply and use of AI computing power. It monitors the size of public, private, and national high-performance computing clusters, and it measures how frequently specific AI accelerators, such as NVIDIA GPUs, Google TPUs, and AMD chips, are cited in published AI research papers, using data from the Zeta Alpha research-discovery platform. [4][9]
In its mid-2025 update, the Compute Index reported that nearly 49,000 open-source AI research papers mentioned the use of a specific AI accelerator, up about 58 percent year over year, with NVIDIA chips appearing in roughly 90 percent of cited compute mentions, down from a peak near 94 percent in 2023 as AMD's MI300-class accelerators gained share. [9] The index serves as a quantitative complement to the narrative report, giving readers a running view of the hardware underpinning AI progress.
The State of AI Report is widely circulated within the AI community and is regularly summarized by technology and business media on release. Coverage by outlets such as DeepLearning.AI's The Batch has highlighted the report's synthesis of the year's major trends in research, investment, regulation, and safety. [3] Its annual appearance has become a fixture of the autumn AI calendar, read by investors evaluating the field, researchers tracking the frontier, operators inside AI companies, and policymakers seeking an accessible overview. [2][3]
The report's prominence is reinforced by its authors' positions in the field: Benaich runs an AI-focused venture firm, while co-founder Hogarth's appointment to lead the UK's national AI safety body underscored the report's policy relevance. [6][7] Its open-access, slide-based format and willingness to make and grade predictions have helped it reach a broad technical, business, and policy audience rather than a purely academic one. [1][2]
The State of AI Report is often discussed alongside two other major efforts to chronicle the field. The most prominent comparison is to the AI Index, produced annually by the Stanford Institute for Human-Centered AI (HAI). The Stanford report is generally more academic, data-heavy, and comprehensive, drawing on large datasets and a steering committee of researchers, whereas the State of AI Report is more opinionated, more closely tied to industry and investment, and distinctive for its forward-looking predictions and scorecard. [1][2]
The report is also complementary to the work of Epoch AI, a research group that builds quantitative datasets and forecasting models for trends in machine learning, such as the growth of training compute and the scaling of large models. Where Epoch AI focuses on rigorous longitudinal data and modeling of specific trends, and the Stanford AI Index aims for breadth and neutrality, the State of AI Report aims to compile and interpret the year's most interesting developments and to stake out testable claims about what comes next. [4][9] Together, the three are frequently cited as leading independent sources for understanding the trajectory of AI.