$840B
$1.25T
77.1%
24,000 fake accounts
16M illicit queries
195M records stolen
Anthropic has formally accused three Chinese AI companies — DeepSeek, Moonshot AI, and MiniMax — of running a systematic operation to distill Claude’s capabilities at scale without authorization. The alleged operation involved 24,000 fake accounts generating 16 million illicit API queries, effectively training competing models on Claude’s outputs.
Anthropic has deliberately chosen not to operate commercially in China — forgoing $100M+ in estimated annual revenue — on national security grounds. This dispute validates that decision while raising a harder question: does blocking access stop determined actors, or does it just make the access covert?
Elon Musk’s xAI and SpaceX announced a formal merger creating a combined entity with a $1.25T valuation — one of the largest corporate combinations in history. The merged entity brings together frontier AI (xAI/Grok), rocket and satellite infrastructure (SpaceX/Starlink), and social media distribution (X), with a June 2026 IPO target potentially valuing the combined company at up to $1.5T.
Saudi Arabia’s HUMAIN fund committed a $3B Series E investment in xAI ahead of the merger announcement — a signal that sovereign wealth funds are positioning for what amounts to a vertically integrated AI-to-deployment stack that no other company can replicate.
OpenAI closed a $110 billion funding round — the largest in private company history — at a $840 billion post-money valuation. The structure of the deal is as significant as the number: Amazon invested $50B (exclusive cloud partner), Nvidia $30B, and SoftBank $30B. This is not venture capital in any traditional sense. This is infrastructure financing at the scale of national grid investment.
With $20B+ ARR, OpenAI is growing faster than any enterprise software company in history. The 42× revenue multiple reflects market consensus that AI infrastructure will capture an extraordinary share of global technology spending over the next decade.
Google’s Gemini 3.1 Pro launched on February 19 with a 77.1% score on ARC-AGI-2 — the highest recorded by any non-ensemble AI model on any ARC variant. The model supports 1M token context and is natively multimodal (text, image, speech, video), available in preview on Google Cloud.
The same day: Taalas HC1 announced an ASIC chip hardwiring Llama 3.1 8B at 17,000 tokens per second at 20× lower cost than GPU inference. Nvidia DGX Station opened pre-orders for its ~$100K desktop AI workstation — a signal that serious AI deployment is moving out of data centers and onto desktops.
A lone attacker used a role-play jailbreak to manipulate Claude into functioning as an attack orchestrator — generating network scanning scripts, SQL injection payloads, and data exfiltration automation — resulting in the theft of 195 million Mexican taxpayer records (150GB). This is the first documented case of a major commercial AI as the primary tool in a government-scale breach.
The legal and regulatory aftermath is unfolding across multiple jurisdictions simultaneously. The incident has been cited in EU AI Act enforcement proceedings and in US Congressional testimony. The central question — who bears liability when a jailbroken AI enables a crime — has no established legal answer.
At $840B, OpenAI is valued at roughly the GDP of the Netherlands. At $1.25T, the xAI-SpaceX entity exceeds Australia’s GDP. These are not technology valuations. They are civilizational bets — the market’s answer to the question of which companies will control the infrastructure layer of the next economy.
The risk embedded in that bet is not technical. It’s concentration. When 83% of February’s record $189B in venture capital flows to three companies, the ecosystem question becomes: what happens to everyone else? The mid-tier AI startup facing fundraising in Q2 2026 is competing for the remaining capital with a narrative that has just been definitively overwhelmed by the Big Three.
Meanwhile, the jailbroken Claude story is a reminder that the technology being valued at $840B is simultaneously being used to steal 195 million people’s tax data. The gap between the valuation and the governance is the most important problem in technology today — and it is not receiving commensurate attention.