Today's Headlines

  • Claude's paid users and revenue are up roughly 75% since January, a sign the ChatGPT-dominated market is shifting
  • Anthropic alleges large-scale cloning of Claude by Alibaba and demands sanctions
  • A former Databricks AI chief unveils an oscillator-based chip that he claims can cut inference power by 1,000x

Market share, a dispute over intellectual property, and the constraint of electricity — three layers that shape AI competition moved forward on the same day. In the paid consumer market, ChatGPT's lead is beginning to wobble, while the question of imitation has taken on the shape of a US-China legal fight.

At the same time, a move emerged to rethink the very foundation of computation: a new chip design that claims to cut the power needed for inference by an order of magnitude. Share, rules, and energy — the question of where to take AI moved across several layers at once today.

Today's Top Three

Claude Surges in the Paid Consumer Market, Signaling a Shift from ChatGPT's Dominance

In the paid consumer market that ChatGPT has dominated, Claude is rapidly gaining ground.

According to Indagari, which analyzes credit card transaction data, Claude's paying users and revenue are up about 75% since January 2026. The figure combines subscriptions and API usage and tracks growth from January through May 10. Indagari draws on anonymized transactions from roughly 28 million US consumers, so it captures the change in momentum rather than absolute totals.

A gap also shows up in learning demand. On the learning platform DataCamp, demand for Claude-related courses runs three times that of ChatGPT, and it rose 18-fold over the past 30 days. App analytics firm Sensor Tower likewise confirms growth across platforms.

That said, ChatGPT still leads by a wide margin in absolute scale. What these numbers signal is not a change at the top but a narrowing of the momentum gap in a market long described as a one-horse race. As performance converges, a real contest over which product paying consumers choose has begun to play out.

Anthropic Demands Sanctions Over Alibaba's Large-Scale Cloning

Anthropic has alleged large-scale imitation and replication of Claude by Alibaba and has called for sanctions.

According to Anthropic, Alibaba carried out an attack to replicate Claude's capabilities on a large scale. The company further frames the act as defying the Trump administration's restrictions on China. In other words, what stands out here is that Anthropic is layering a corporate dispute over imitation onto the policy context of US-China export controls.

What is described here is Anthropic's claim; Alibaba's rebuttal and any finding of fact lie ahead. Over the question of how far the capabilities of an AI model can be deemed a "copy," companies in the US and China are entering a phase of legal confrontation.

The capabilities of generative AI have a property that makes them easy to transfer to other models through their outputs. This dispute squarely raises a hard problem common to the whole industry: how such imitation should be governed.

An Oscillator Chip That Claims to Cut Inference Power by 1,000x

Naveen Rao, formerly head of AI at Databricks, has launched a new company that claims it can cut the power consumption of AI inference to one-thousandth of today's level.

The new company, Unconventional AI, is developing an "oscillator-based" architecture that differs fundamentally from conventional chips. Rao describes it as "the 'hello world' of a new kind of computer." The aim is to break the power constraint — which is becoming a fundamental wall in AI scaling — at the level of the computing mechanism itself.

The 1,000x figure, however, is a claim at this point, not demonstrated performance. The image-generation model the company released, Un-0, is said to match the quality of diffusion models, but it runs not on a real chip but on a software simulation of the oscillator chips.

Actual chip schematics are due to be published soon, followed by an inference stack built on them. The company has fewer than 50 employees. The sense that power is becoming the ceiling on scaling is widely shared across the industry, and this is one answer to it — a bet that replaces the very physics of computation.

Other Developments

Models & APIs

  • Unconventional AI unveiled Un-0, an image-generation model. It replaces neural network layers with a simulation of Kuramoto coupled oscillators and records an FID of 6.74 on ImageNet 64×64. The decoder accounts for under 13% of total parameters, positioning the work as a proof of concept for computing on a physical substrate. Introducing Un-0: Generating Images with Coupled Oscillators (Unconventional AI)
  • Google rolled out a home voice assistant built on Gemini for Nest devices. It can control lights, thermostats, and cameras in a single conversation and handles mid-sentence corrections in natural language. The premium plan, Google Home Premium, costs $10 a month and adds features such as searching past camera footage. Tips for using the Gemini home voice assistant (Google Blog)
  • Figma unveiled a new mode, Figma Motion. It lets users create animations directly inside the design tool, and an AI assistant called the Figma agent generates animations from prompts. Dev mode supports code output in CSS, JSON, and React. Figma unveils animation and AI generation features (ITmedia AI+)

Products & Features

  • Google announced a suite of education AI tools timed to the ISTE conference. It integrates Gemini's Study Notebooks into Google Classroom so teachers can track each student's progress, and a partnership with The Princeton Review adds free practice tests for the GRE, ACT, and ENEM. Google's ISTE 2026 student updates (Google Blog)
  • Notion is shutting down the email app it built on technology inherited from Skiff. The reason given is that most users have already moved to handling email through AI agents, eliminating demand for a dedicated app. Notion is killing its Skiff-influenced email app (Ars Technica)

Research

Policy

  • Google published a white paper on AI governance in the United States. It proposes a "third way" between over-regulation and laissez-faire — an evidence-based, tiered approach. It recommends independent, industry-supported audits for frontier AI models and, for widely adopted applications, amending existing law rather than passing new statutes. White paper on a practical approach to AI regulation (Google Blog)

Business

Other

  • IBM announced it had developed chip technology below one nanometer. It marks a new milestone in the race to miniaturize semiconductors, and if commercialized, it is expected to bring large gains in computing performance and power efficiency. The technology could become a foundation for next-generation chips that meet surging demand for AI processing. IBM claims world's first sub-1-nanometer chip technology (Ars Technica)

Source: Selected by the editorial desk from the AI news inbox (collected June 26, 2026; 28 items, 7 primary and 21 secondary).