AI Daily Briefing — 2026-06-26

The rapid expansion of artificial intelligence is creating a pressing need for vast amounts of data. However, much of the information on the web remains locked behind technical barriers or is unstructured, making it difficult for AI models to use effectively. The web was never designed for automated discovery and retrieval, posing a significant hurdle for enterprises seeking to capitalize on AI's potential.

To address this, a new web data infrastructure layer is emerging. This layer must navigate millions of domains and billions of new URLs created weekly, delivering real-time information while overcoming technical obstacles. Or Lenchner, CEO of Bright Data, notes that there is far more data available than most realize, but accessing it requires sophisticated systems capable of handling diverse formats, languages, and access rules.

AI performance now depends not just on model architecture but on the ability to retrieve fresh, relevant, and trustworthy data in real time. Static training data is no longer sufficient for tracking dynamic market trends, competitor pricing, or consumer sentiment. Without real-time context, AI outputs can be stale, leading to poor decisions and eroding user trust. A recent survey found that 56% of AI practitioners believe real-time web data is essential for improving trust in AI outputs.

Despite advances like retrieval-augmented generation, many AI systems still struggle to deliver current and contextually relevant outputs. According to Gartner, 60% of AI projects lacking AI-ready data may be abandoned. As Lenchner emphasizes, retrieving data at scale is not enough; it must also be done in real time to meet user expectations and reduce latency. The next frontier of AI depends on overcoming these infrastructure challenges.

Anthropic's Claude is increasingly winning over paying consumers, a space long dominated by ChatGPT, according to new data from credit card analytics firm Indagari. The firm analyzed billions of anonymized transactions from about 28 million U.S. consumers, tracking weekly payments for subscriptions and API tokens from 2025 through May 2026. The data reveals Claude's paying consumer base and revenue have grown roughly 75% since January 2026, with gains continuing even after Anthropic refused to let its models be used for mass surveillance or autonomous weapons by the Trump administration.

Further evidence comes from DataCamp, an online education platform with about 20 million users. Among self-directed consumers, demand for Claude courses now outpaces ChatGPT courses by three to one, and searches for "Claude" have become the most popular term on the site—even surpassing "AI." Demand for Claude courses surged 18-fold in the last 30 days alone, though ChatGPT courses remain more popular for corporate training.

Despite Claude's rapid growth, ChatGPT still holds a commanding lead in overall consumer usage and paying users, per Sensor Tower and Indagari data. ChatGPT's growth has slowed largely due to its massive existing reach. Still, Claude is steadily closing the gap in consumer dollars and awareness this year.

As both OpenAI and Anthropic near potential public offerings, questions remain about how Anthropic's recent clash with the U.S. government—which banned its powerful cybersecurity models Mythos 5 and Fable 5 from non-American users—will affect business. So far, available data shows Anthropic continuing to grow its consumer and enterprise user base.

Inside General Intuition’s New York research floor, a monitor showed an AI agent playing a game resembling Fortnite for 100 hours straight. But the same intelligence driving that virtual player also powered a four-legged robot roaming the office. The robot, guided by a single camera, clumsily explored its surroundings like a toddler learning spatial awareness. Remarkably, the model was fine-tuned with just eight minutes of real-world data collected on a city street, not in the office where it operated.

This ability to generalize from gameplay to simulation to physical embodiment is the core of General Intuition’s mission. The startup, spun out of Medal—a platform for sharing gaming clips—leveraged hundreds of millions of hours of uploaded gameplay to train its AI in spatial-temporal reasoning. Unlike rivals that infer actions from video alone, General Intuition used embedded action labels showing exactly which buttons players pressed and when. Co-founder and CEO Pim de Witte calls this the next stage of pre-training, enabling a single model to respond to both game screens and real-world dynamics.

On Thursday, General Intuition announced a $320 million funding round at a $2.3 billion valuation, bringing total disclosed funding to $454 million. The company’s world model, which generates simulated environments frame by frame, serves as an internal training gym rather than a product. The ultimate goal is to sell the agentic model itself, which has already learned fundamental physics—like walls being solid and ladders being climbable—from millions of hours of gameplay.

A former top AI executive at Databricks is betting on a radical new computer architecture to dramatically reduce the power consumed by artificial intelligence. Naveen Rao, now leading Unconventional AI, argues that the industry's growing energy demands are unsustainable. His solution involves replacing traditional chips with oscillator-based hardware that he claims can cut power usage by a factor of 1,000.

To prove the concept, the startup released Un-0, an image-generation model that runs on a software simulation of its novel chips. According to a new paper, Un-0 matches the performance of leading diffusion models like Stable Diffusion. Rao described the project as a foundational step, calling it the “hello world” of a new computing paradigm.

The company plans to release chip schematics soon and build a full inference stack from the ground up. Rao envisions a future where Unconventional AI provides compute capacity via a simple network cable, but with drastically lower energy costs. He believes power availability will become the primary bottleneck for AI scaling in the coming years.

Despite having fewer than 50 employees, Unconventional AI is pursuing one of the most ambitious goals in the field. Rao insists that addressing the energy limit is critical, stating that AI progress will ultimately be constrained by power, not algorithms. The startup aims to deliver a working chip-based system within the next year.

The artificial intelligence boom has sparked a rush to build data centers, but getting them operational remains a major hurdle. Even after securing GPUs and networking gear, providers face months of configuration work before they can offer cloud services for AI training and inference. Netris, a network automation startup, aims to solve this bottleneck for smaller AI cloud operators, known as neoclouds.

Netris offers software that runs on network switches, paired with a platform that automates setup, configuration, and daily operations. This system provides hardware-level network abstraction and multi-tenancy, allowing neoclouds to serve multiple customers without manual tweaking. CEO Alex Saroyan explained that traditional software-defined networking falls short for AI because it lacks hardware acceleration, which is critical for handling massive traffic loads.

The startup’s vendor-agnostic approach works with both Nvidia and AMD servers, and it has already been deployed across over 35 GPU clusters worldwide, powering roughly a million GPUs for clients like Lightning AI, Foxconn, and Hewlett Packard Enterprise. To fuel further growth, Netris has raised $15 million in Series A funding from Andreessen Horowitz. Notably, the company relies on proprietary algorithms rather than AI itself, having developed its technology years before the current AI wave.