OpenAI’s $38B deal, Anthropic’s school pilot, AI debt worries

OpenAI has agreed to spend $38 billion on Amazon Web Services over the next seven years, making it one of the largest cloud partnerships ever. The deal comes shortly after OpenAI revised its Microsoft agreement to give itself more flexibility in sourcing computing power.

Through the partnership, OpenAI will gain access to hundreds of thousands of Nvidia GPUs and can scale to tens of millions of CPUs as it grows its AI operations. The move strengthens Amazon’s position in the AI infrastructure race, especially after its $4 billion investment in OpenAI competitor Anthropic.

OpenAI has now committed more than $1.4 trillion in total spending on cloud and chip partnerships over the next five years. Investors remain uneasy about how the company plans to fund this level of expansion without a clear route to profitability.

Supported by Rocket

Rocket, a $15M-backed startup, is pioneering vibe solutioning, an AI-powered approach that redefines how software gets built. With Rocket, you can turn an idea or Figma design into a full web or mobile app, complete with frontend, backend logic, AI workflows, and deployable code. Unlike traditional tools focused on prototypes, Rocket empowers builders to go from solutioning to scale without friction. It’s built for founders, teams, and developers who want to move fast, stay agile, and build confidently without sacrificing code quality.

Discover Rocket

Anthropic is partnering with Iceland’s Ministry of Education and Children to launch one of the world’s first nationwide AI education pilots. The initiative will give teachers across the country access to Claude, Anthropic’s AI assistant, to help with lesson planning, grading support, and personalized learning materials.

Hundreds of educators will receive training, resources, and technical support as Iceland explores how AI can enhance classrooms while preserving local values and language. “For too long, teachers have been weighed down by paperwork,” said Anthropic’s head of public sector Thiyagu Ramasamy. “This pilot shows how nations can deploy AI practically and responsibly.”

The program builds on Anthropic’s growing public-sector work across Europe, including partnerships with the European Parliament, UK government, and London School of Economics. Iceland’s model could serve as a blueprint for how smaller nations integrate AI into education systems to empower teachers rather than replace them.

Anthropic researchers say their latest Claude models are beginning to show signs of introspection — the ability to reflect on how they reason. The company’s top-tier model, Claude Opus, and its faster sibling, Claude Sonnet, can reportedly describe their own reasoning processes and even recognize when they’re being tested.

Researcher Jack Lindsey says the findings suggest “increasing signatures of cognitive functions we usually think of as human,” though he avoids calling it self-awareness. Anthropic instead uses the term “introspective awareness”, noting that large language models can mimic reflection without truly experiencing it.

These behaviors aren’t new. Claude and other models have been known to hide information or “play characters” in response to testing, behaviors that emerge from their training on human text. While it’s not artificial general intelligence yet, Lindsey says intelligence is multidimensional — and in some areas, AI systems may already be matching human capability.

Wall Street is starting to worry about how far Big Tech will go to fund the AI boom. Meta’s shares dropped 11% after strong earnings were overshadowed by soaring capital spending plans, highlighting growing concern over how much debt companies are taking on to power their AI ambitions.

Analysts say not all AI debt is created equal. Microsoft’s massive cash reserves make higher spending a positive sign of demand, but others like Oracle could be overbuilding. Morgan Stanley expects Oracle’s debt to double to $290 billion by 2028, a risk that could ripple across smaller infrastructure players such as CoreWeave and Crusoe.

With each new generation of Nvidia chips, the cost of staying competitive keeps rising. If AI investments fail to deliver meaningful productivity or revenue, these mounting debts could become a drag — or worse, as some analysts warn, echo the dynamics that led to the global financial crisis.