AI may already be adding billions to the economy—without showing up in the data

Hello and welcome to Eye on AI. In this edition…The AI economy has a measurement gap…Anthropic files IPO paperwork…Meta bets on subscriptions and enterprise to monetize AI…and AI-generated fake citations are infiltrating scientific literature.

When it comes to measuring the economic impact of AI, no one can agree on where to start.

Listen to the narrative coming from the Big Tech firms, and AI is already supposed to be transforming everything from how we work to how companies are organized and even how entire industries, such as software engineering, function. But if you look at the official economic statistics, like productivity numbers or GDP growth rate, you’d be hard-pressed to find the numbers to back this up. Some argue the technology is transforming the economy in a way our statistics simply cannot keep up with. Others say that, for all its hype, AI has yet to show up in firm-level productivity in any systematic way. 

That gap between the hype and the numbers is what a new policy brief from the Peterson Institute for International Economics is trying to explain.

Measuring AI’s economic impact

The brief, written by Anton Korinek, a nonresident senior fellow at Peterson and head of Transformative AI Economic Studies at the Anthropic Institute, and Patrick McKelvey, a senior data scientist at the Bank of Canada, argues that AI is already growing at extraordinary speed, but that official statistics are simply not built to track it.

The brief points to two problems. First, AI activity is scattered across dozens of different industries in the official accounts—cloud services, software, data processing—so there’s no single place where you can see the AI economy as a whole. Second, the stats have no good way to account for how fast AI is improving.

They estimate that AI generated roughly $250 billion in economic activity in 2025, comparable in size to the entire U.S. airline industry, and that the amount of AI output the industry can produce is growing at around 2,600% a year. The authors also estimate that the cost of getting the same level of AI performance has fallen by about 94% a year—meaning each dollar spent on AI today buys vastly more than it did a year ago.

To arrive at those figures, they build their own estimates from scratch—using data on GPU rental rates, electricity consumption, AI inference prices, and the pace of algorithmic progress in AI training—rather than relying on official statistics. They also calculate that if official statistics accounted for AI’s rapid improvement in capability, U.S. economic growth in 2025 would appear about 4 percentage points higher. (The authors caution that the estimate is an upper bound, meaning it represents the maximum plausible impact rather than their central estimate.)

Their proposed fix: give AI its own dedicated statistical track—the same way governments separately account for energy or international trade—that aggregates AI activity across industries and adjusts for how quickly the technology is improving. Build that now, they argue, or the gap in the data risks becoming a gap in policy, meaning governments could find themselves making decisions about taxes, labor markets, and public spending without being able to see what the AI economy is actually doing. Or as they put it: “What cannot be measured cannot be steered.”

What the data is missing

However, not everyone is convinced. Diane Coyle, Bennett Professor of Public Policy at the University of Cambridge, told Fortune that while she agrees that the measurement gap is real, she disputes the scale of what Korinek and McKelvey are claiming. One of her objections is that AI is mostly used to help create other products and services rather than being a product in its own right. GDP measures the final goods and services that reach consumers. If AI is mostly an ingredient rather than a finished product, its economic impact only matters if it actually makes the end product better.

(Notably, Korinek and McKelvey actually acknowledge that AI is mostly an intermediate input—they cite it as the reason their 4-point GDP estimate is a ceiling, not a prediction.)

According to Coyle, there is also still little systematic evidence that AI is increasing productivity at the firm level, and even where individual workers are going faster, that doesn’t always feed through to the organization as a whole. If one department speeds up but the next hasn’t adopted AI, the gains hit a bottleneck and disappear.

“I think AI is a significant technology. It will have these great effects, but I think both the speed and scale that this paper is claiming for it are overdone,” she said.

At its core, the question is really about how much AI is transforming the way we work and whether the tools we have to answer that question are fit for purpose. As Coyle puts it, the challenge isn’t just that we’re measuring badly—it’s that we haven’t even agreed on what we should be measuring.

With that, here’s more AI news.

Beatrice Nolan
beatrice.nolan@fortune.com
@beafreyanolan

FORTUNE ON AI

Anthropic leapfrogs OpenAI with a record $965 billion valuation and says its ‘Mythos’ AI model is coming soon — Beatrice Nolan

Anthropic confidentially files its S-1 first—but the IPO race with OpenAI is just beginning — Allie Garfinkle 

Asana was battered by the AI boom. Now it’s betting its future on humans and agents working together. — Beatrice Nolan

Snowflake CEO says monster quarter shows why software firms need new pricing models to thrive in AI age — Sebastian Herrera

AI IN THE NEWS

Anthropic files confidentially for IPO. Anthropic submitted a confidential S-1 draft registration statement to the SEC on June 1, setting the stage for a potential initial public offering of its common stock. No pricing or share count has been set yet. The filing comes less than a week after Anthropic raised $65 billion in a Series H round that pushed its valuation to $965 billion—up from $183 billion as recently as December 2025. The company’s annualized revenue run rate hit $47 billion in May. The move puts Anthropic ahead of rival OpenAI in a closely watched race to reach public markets. Reports suggest OpenAI’s financials are comparatively weaker and that secondary market demand for its shares has softened—while competition for Anthropic secondary shares has been described as fierce. Anthropic’s latest round was led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital, with Blackstone, Brookfield, General Catalyst, and GIC among a broader group of participants. Read more in Fortune. 

Anthropic opens Mythos access to EU cybersecurity body. Anthropic is set to grant ENISA, the EU’s cybersecurity agency, access to Claude Mythos, according to a Bloomberg report, making it the first European body to join Project Glasswing, the AI lab’s restricted preview program for the model. Mythos is capable of finding and exploiting vulnerabilities in computer systems, and access has been tightly limited since its April debut over fears it could fall into criminal hands. Commission officials reportedly flew to San Francisco last Thursday to press Anthropic executives directly, following weeks of unproductive negotiations. Anthropic informed the Commission of its decision over the weekend, but conditions are still being finalized. Project Glasswing participants—which already include firms like Amazon, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks—are permitted to use Mythos for defensive purposes only. Read more in Bloomberg.

Meta bets on subscriptions and enterprise to monetize AI. Meta is selling consumer subscriptions to its Meta AI chatbot for the first time, per Bloomberg, in a bid to offset hundreds of billions of dollars in AI investment. Two tiers are on offer: Meta One Plus at $7.99 a month, covering image and video generation and extended reasoning, and Meta One Premium at $19.99, with higher usage limits. The rollout begins in Singapore, Guatemala, and Bolivia, with more countries set to follow. Meta is also making a direct enterprise push, according to an internal memo obtained by The Information. In the memo, Meta’s Head of Product Naomi Gleit outlined a new unit called Enterprise Solutions that will embed product managers and software engineers inside large corporate clients—mirroring the forward-deployed engineer model used by companies like Anthropic. CTO and Head of Reality Labs Andrew Bosworth called 2026 “a critical year” for Meta’s transformation. Read more in The Information.

Cognition raises $1bn at a $26bn valuation. Cognition AI’s $1 billion funding round is the latest sign of strong demand for companies using AI for software development. The round was led by Lux Capital, General Catalyst, and 8VC, with participation from Founders Fund and others. The valuation has more than doubled since a previous round in September. Cognition’s flagship product, Devin, is an AI agent designed to automate the programming process. Revenue has grown from $37 million last May to a $492 million run rate, with the company targeting $1 billion by year-end, and approximately 89% of Cognition’s own internal code is now written by Devin. CEO Scott Wu said the funding is intended to keep the company independent. Read more in Bloomberg.

Trump signs scaled-back AI executive order. On Tuesday, President Trump signed an executive order requiring AI companies to voluntarily submit powerful new models to government review 30 days before public release. This is down from the 90-day window in an earlier draft that the industry had resisted. The order also directs the Treasury Department to establish a cybersecurity clearinghouse within 30 days and creates a classified NSA-overseen benchmarking process to assess the national security implications of advanced models. The signing came after weeks of reversals and was driven in part by alarm over Anthropic’s Mythos model, which researchers say has already identified vulnerabilities in widely used computer systems. Read more in Politico.

EYE ON AI RESEARCH

AI-generated fake citations are infiltrating scientific literature. A team of researchers from Cornell University and UCLA has found that an estimated 146,900 hallucinated citations appeared in scientific papers published in 2025 alone.  The researchers audited 111 million references across 2.5 million papers on arXiv, bioRxiv, SSRN, and PubMed Central, and found an estimated 146,900 hallucinated citations in papers published in 2025 alone. The fake references appeared across many papers, each containing a small number of fabricated sources, suggesting researchers may have used AI to write and failed to check the output. Early-career scientists and small teams were most likely to include fake citations. The errors also followed a pattern: hallucinated references disproportionately credited already-prominent and predominantly male scholars, potentially reinforcing existing inequalities in scientific recognition. Existing safeguards are also not keeping pace with the new technology. The researchers found an estimated 78.8% of non-existent citations passed through arXiv moderation undetected. 

AI CALENDAR

June 8-10: Fortune Brainstorm Tech, Aspen, Colo. Apply to attend here.

June 17-20: VivaTech, Paris.

July 6-11: International Conference on Machine Learning (ICML), Seoul, South Korea.

July 7-10: AI for Good Summit, Geneva, Switzerland.

Aug. 4-6: Ai4 2026, Las Vegas.

BRAIN FOOD

Will investors accept Anthropic’s corporate structure? Anthropic’s IPO filing will put its unusual corporate structure in front of public investors for the first time. The company is incorporated as a Delaware Public Benefit Corporation, meaning its board must balance shareholder financial interests against its stated mission of responsible AI development for the long-term benefit of humanity. But Anthropic’s Long-Term Benefit Trust, an independent corporate governance body created by the lab, holds a special class of shares giving it the power to elect a majority of Anthropic’s board of directors, with five trustees required to have expertise in AI safety, national security, public policy, and social enterprise, and explicitly excluded from holding any financial interest in Anthropic.

OpenAI’s for-profit arm is also a public benefit corporation controlled by a nonprofit foundation whose directors must balance shareholder returns against a broader mission—a structure whose instability was laid bare when OpenAI’s board fired and briefly ousted Sam Altman in 2023. Anthropic may find itself facing some hard questions in public markets. The LTBT currently holds a majority of board seats, but it is unclear that public market investors will indefinitely tolerate a governance structure in which their capital is subordinate to a trust they cannot elect or remove.

Fortune AIQ Special Digital Issue: The AI Economy

From global corporations to local entrepreneurs, artificial intelligence is changing the way businesses operate, compete, and succeed. Explore all of Fortune AIQ, and read the latest collection of stories below:

–After AI stole his clients, one Big Tech ghostwriter is using AI to get them back

–Outnumbered: At $4 billion ClickUp, a 3:1 agent-to-human ratio is rewiring work itself

–How a mom-and-pop car wash chain went from sticky notes to AI-powered operations that are upleveling every part of the company

–Solo founders are using AI to do the work of entire teams—but going it alone has limits

–How EarthRanger uses AI to help protect endangered species—and boost the wildlife tourism industry

–The smartphone’s days are numbered. Meet the device that could come next

#adding #billions #economywithout #showing #data