A landmark day across the industry: AI systems cracked a decades-old math problem, two of the biggest AI labs made major financial headlines, and the hardware race intensified on multiple fronts.
AI Achieves a Mathematical First
In what researchers are calling both thrilling and unsettling, an internal OpenAI reasoning model independently disproved a conjecture in discrete geometry first posed by mathematician Paul Erdős in 1946. This marks the first time an AI system has autonomously solved a prominent unsolved mathematics problem — not by brute force or exhaustive search, but by arriving at a genuine disproof. The achievement has the mathematical community simultaneously excited about AI's potential as a research collaborator and apprehensive about what it signals for the longer arc of human-led discovery.
OpenAI and Anthropic: Business Headlines
OpenAI is reportedly in the final stages of preparing a confidential IPO filing, potentially as soon as this week, with a September market debut targeting a valuation between $850 billion and $1 trillion. The move would cap a remarkable period of commercial expansion for the company.
Anthropic, meanwhile, is making news on two fronts. The company has signed a computing deal with SpaceX worth approximately $1.25 billion per month through May 2029 — a total commitment approaching $45 billion — securing capacity from the Colossus 1 data center and a second SpaceX facility. On the financial side, Anthropic projects its Q2 revenue will double to $10.9 billion, putting the company on the cusp of its first operating profit at $559 million. That profitability window may be brief, however, as planned investments in model training costs are expected to expand again. Adding to its momentum, Andrej Karpathy — co-founder of OpenAI and former head of AI at Tesla — has joined Anthropic, bringing one of the field's most respected technical voices and communicators to the lab.
NVIDIA: Record Quarter and a New Market
NVIDIA reported $81.6 billion in Q1 FY27 revenue, exceeding Wall Street expectations and setting a new record. The Data Center segment alone contributed $75 billion of that total, underscoring just how central AI infrastructure has become to NVIDIA's business. CEO Jensen Huang also unveiled the Vera CPU, a processor purpose-built for agentic AI workloads, framing it as the entry point into what NVIDIA is calling a $200 billion addressable market. Separately, NVIDIA announced it is opening its first Singapore research hub, focused on robotics and AI infrastructure efficiency as part of Singapore's broader ambitions to become a regional AI deployment hub.
The Hardware Race Heats Up
The semiconductor and chip landscape saw significant moves. AMD announced a $10 billion investment in Taiwan to ramp production of its next-generation 2-nanometer Venice chips, with partnerships aimed at sidestepping anticipated TSMC supply bottlenecks. In China, Alibaba unveiled the Zhenwu M890 AI chip, which offers triple the performance of its predecessor and signals continued momentum in China's push for domestic AI hardware self-sufficiency. Chinese firm Linkerbot, a leading supplier of dexterous robotic hands, is reportedly seeking a Hong Kong IPO at a $6 billion valuation, positioning itself as a key components player in China's growing robot ecosystem.
SpaceX's IPO filing drew attention beyond its own valuation — the company mapped a $28.5 trillion total addressable market, with 93% tied to AI infrastructure. The implication, echoed widely in investor commentary, is that future AI dominance may hinge not just on software and models but on control over power, connectivity, and orbital computing infrastructure.
Robotics: From Orbit to Your Kitchen
The robotics sector saw a range of announcements spanning commercial deployment to experimental hardware. San Francisco startup Gatsby launched a pilot program offering on-demand home cleaning via the Unitree G1 humanoid robot at $150 per visit, positioning the service as a competitor to both professional cleaners and single-task appliances. ETH Zürich, meanwhile, unveiled HELIOS — a four-armed humanoid designed specifically for zero-gravity environments, intended for tasks like orbital structure assembly. Chinese autonomous vehicle maker XPENG began mass production of an L4 robotaxi built on its GX platform, deploying proprietary Turing AI chips and a vision-only end-to-end model that foregoes LiDAR and HD maps.
Google Tightens Its AI Grip
Google continued its broad push across multiple fronts. The company rolled out Gemini-powered advertising formats for AI Mode in Search, including Conversational Discovery ads and AI Shopping ads capable of writing custom product explainers — blurring further the line between organic AI results and paid placements. Multiple analysts and reports this week suggested Google is gaining meaningful ground on OpenAI in the consumer AI market. On the infrastructure side, Google released Agent Executor, an open-source runtime standard for reliable execution of long-running agentic workflows. And Gemini Omni Flash — a world model capable of generating and conversationally editing video — is launching this week inside YouTube Shorts and YouTube Create at no cost.
Policy: OpenAI Goes State-by-State, Trump Delays AI Order
OpenAI has pivoted its regulatory strategy to what it's calling "reverse federalism" — engaging state capitals directly to encourage jurisdictions like California and New York to adopt AI transparency and audit frameworks rather than waiting on federal action. Separately, President Trump postponed signing a planned executive order that would have established government review processes for advanced AI models, citing concerns that portions of the order could impede AI development.
Security and Developer Tools
A security incident at GitHub saw attackers gain access to roughly 3,800 internal repositories after an employee installed a malicious VS Code extension — a reminder that developer tooling has become a significant supply-chain attack surface. In response to related concerns, 1Password announced an integration with OpenAI Codex to provide AI coding agents with just-in-time credential access, avoiding the pattern of storing secrets in prompts or repositories. Figma launched an on-canvas design agent capable of generating design directions from a single prompt, automating bulk edits, and synthesizing stakeholder feedback for marketing teams.
Quick Takes
Meta and employee data: A leaked audio recording revealed Mark Zuckerberg explaining that Meta had monitored employee activity across internal tools and coding software to train its AI models — a disclosure that came just days before the company laid off approximately 8,000 employees.
Bezos's Prometheus: Jeff Bezos offered the first public description of Project Prometheus, a $38 billion venture he describes as building an "artificial general engineer" and next-generation CAD tools for designing physical objects — explicitly "nothing to do with robotics."
Apple at WWDC: Apple is reportedly planning a major Siri overhaul for WWDC 2026, including AI writing tools under the "Write With Siri" banner, Grammarly-style editing suggestions, and prompt-based shortcuts.
AI at Cannes: *Hell Grind*, a 95-minute feature film entirely generated by AI, premiered at the Cannes Film Festival. The film cost $500,000 to produce, with $400,000 of that going to compute alone.
Stable Audio 3.0: Stability AI released Stable Audio 3.0, an open-weight model capable of generating music and sound effects up to six minutes in length.
OpenAI Guaranteed Capacity: OpenAI launched a new enterprise offering allowing customers to lock in long-term access to compute capacity, addressing uncertainty as demand for advanced models continues to climb.
Gen Z skepticism: A new survey found nearly half of Gen Z workers believe AI's risks outweigh its benefits, with widespread anxiety about impacts on jobs and creativity. A separate Gartner study found 69% of B2B buyers prefer to validate AI-generated research with a human sales rep before acting on it — suggesting the role of humans in the loop remains durable even as AI handles more of the research layer.