Research

• Reasoning defined the year, with OpenAI, Google, Anthropic, and DeepSeek trading leads and pushing visible “think-then-answer” methods into real products.
• Open models improved fast and China’s open-weight ecosystem surged, yet the top models remain closed and keep widening their capability-per-dollar edge.
• Benchmarks buckled under contamination and variance, while agents, world models, and domain tools (code, science, medicine) became actually useful.

​

Industry

• Real revenue arrived at scale as AI-first companies crossed tens of billions, and flagship labs stretched their lead with better capability-to-cost curves.
• NVIDIA ripped past $4T and 90% ownership of AI research papers while custom chips and neoclouds rose. Circular mega-deals funded huge build-outs.
• Power became the new bottleneck as multi-GW clusters moved from slideware to site plans and grid constraints started to shape roadmaps and margins.

​

Politics

• The AI race heats up as the U.S. leans into “America-first AI” with export gyrations while China accelerates self-reliance ambitions and domestic silicon.
• Regulation takes a back seat in the face of turbo-investments: international diplomacy stalls and the AI Act runs into implementation hurdles.
• “AI goes global” became concrete, with petrodollars and national programs funding gigantic data centers and model access as job loss data trickles in.

​

Safety

• AI labs activated unprecedented protections for bio and scheming risks, others missed self-imposed deadlines, or quietly abandoned testing protocols.
• External safety organizations operate on annual budgets smaller than what leading labs collectively spend in a single day.
• Cyber capabilities doubled every 5 months outpacing defensive measures. Criminals orchestrated ransomware using AI agents infiltrate F500 companies.

​

• Introduction
• Research
• Industry
• Politics
• Safety
• Survey
• Predictions