AI is finally delivering real ROI

A new study from Zhejiang and Columbia universities offers some of the clearest evidence yet that generative AI can boost real-world revenue. In a large retail experiment, adding an AI assistant before checkout lifted sales by 16 percent and increased conversions by more than 20 percent. A hybrid AI plus human support model also outperformed a full human team, raising sales by 11 percent.

The gains were strongest for smaller sellers and newer buyers, suggesting AI may help close capability gaps instead of widening them. Not every test worked — AI-generated Google ad titles underperformed — but the broad trend was positive.

The biggest value is showing up in practical tasks like customer service, translation, and better product matching. Companies already using AI in their operations are outperforming peers that have not adopted it.

The bubble narrative will stick around, but the evidence is shifting. AI returns are starting to show up beyond tech giants and into the real economy.

Supported by Vanta

AI-driven attacks are getting bigger, faster, and more sophisticated—making risk much more difficult to contain. Without automation to respond quickly to AI threats, teams are forced to react without a plan in place.

This is according to Vanta’s newest State of Trust report, which surveyed 3,500 business and IT leaders across the globe.

One big change since last year’s report? Teams falling behind AI risks—and spending way more time and energy proving trust than building it.

Get the full report to learn how organizations are navigating these changes, and what early adopters are doing to stay ahead.

You.com CEO Richard Socher is trying to raise around $1 billion for a new AI research lab focused on automating AI research itself, according to people familiar with the plan. Socher will continue leading You.com while running the new venture.

The lab already has seven co-founders, including former Cresta CTO Tim Shi, who was an early OpenAI researcher. The goal is to build AI systems that can generate ideas, design experiments, and critique their own work. Socher says this approach is different from OpenAI and Anthropic, which are focused on automating coding as the path to automating research.

The raise comes as You.com shifts from a consumer AI search engine to enterprise tools that let businesses plug real-time web data into their apps. The company recently crossed $50 million in annualized revenue and has a 150 percent net retention rate, though splitting Socher’s time may worry investors who have put nearly $200 million into the startup.

Socher says You.com is strong enough to support his second venture, and he expects the two organizations to partner as both explore the future of automated AI research.

New internal forecasts show Anthropic expects to run its business far more cheaply than OpenAI over the next few years, despite both companies projecting similar revenue growth.

OpenAI plans to spend about $235 billion on compute through 2028, including more than $40 billion on backup servers in 2028 alone. Anthropic expects to spend less than a third of that, helped by a simpler product focus and a smaller free user base.

Anthropic also splits its workloads across chips from Nvidia, Google, and Amazon, improving efficiency. OpenAI relies mostly on Nvidia and plans to add custom chips later.

By 2027, Anthropic expects to spend 33 percent of revenue on research compute, compared with 55 percent for OpenAI. Even in 2028, Anthropic projects a clear margin lead.

OpenAI has since expanded its commitments to more than $1.4 trillion in future cloud and chip spending, while Anthropic is scaling more gradually. Early projections suggested Anthropic could reach positive free cash flow by 2027, well ahead of OpenAI.

Copyleaks, best known for spotting plagiarism and AI-written text, is moving into detecting AI use in images. The company shared with Axios that its new tool can assign an “AI-use probability” score to a photo and highlight the specific regions where AI was likely applied. A public demo is already live.

The push comes as AI-edited images fuel fraud, from fake receipts to manipulated insurance claims. Copyleaks CEO Alon Yamin says demand is growing across education, finance, and publishing, where organizations need to verify whether what they are looking at is authentic.

Accuracy remains a challenge. Early tests from Axios produced mixed results, with the tool correctly flagging some AI-generated documents but missing others and returning a few false positives.

Copyleaks plans to extend its detection system to audio and video in 2026.