AI News Roundup: OpenAI Revenue Miss Rattles Markets, Google Bets $40B on Anthropic, China Blocks Meta-Manus Deal

OpenAI Misses Revenue Targets as CFO Warns of Funding Crunch

OpenAI has missed its own projections for user growth and revenue, sending shockwaves through the AI supply chain. An internal milestone of one billion weekly active ChatGPT users by year’s end went unmet, and monthly revenue fell short on several occasions as Google’s Gemini and Anthropic’s Claude gained ground in enterprise and coding markets.

CFO Sarah Friar has warned colleagues that if revenue growth doesn’t accelerate, the company could face difficulty funding its roughly $600 billion in locked-in data-center contracts. Oracle, which holds a $300 billion five-year compute partnership with OpenAI, dropped 4% on the news. Chipmakers Broadcom and AMD fell 3–4% in sympathy.

OpenAI pushed back hard, calling the report “ridiculous” and insisting the company is “totally aligned on buying as much compute as we can.” The miss comes just days after OpenAI launched GPT-5.5 and amid reports the company is taking early steps toward a public listing, potentially as soon as late 2026.

Google Commits Up to $40 Billion to Anthropic

Google announced it will invest up to $40 billion in Anthropic — $10 billion in cash now at a $350 billion valuation, with another $30 billion tied to performance milestones. The deal deepens a relationship between two companies that are simultaneously partners and rivals in the AI race.

Prior to the deal, Google had invested over $3 billion and held a reported 14% stake in Anthropic. The investment mirrors a separate agreement Anthropic struck with Amazon just weeks earlier: $5 billion upfront with up to $20 billion more contingent on commercial milestones. Anthropic’s annualized revenue has now topped $30 billion, up from roughly $9 billion at the end of 2025, and investors have recently sought to back the company at valuations of $800 billion or more.

GPT-5.5 Launches as OpenAI’s First Agent-Native Model

OpenAI released GPT-5.5 on April 23, positioning it as the company’s first flagship model built primarily as an agent runtime rather than a chat model. It excels at writing and debugging code, researching online, analyzing data, creating documents, operating software, and chaining tools until a task is finished.

API pricing lands at $5/$30 per million tokens for standard and $30/$180 for GPT-5.5 Pro. The model is available to Plus, Pro, Business, and Enterprise ChatGPT users, with the Pro tier restricted to paying subscribers. The launch comes on the heels of GPT-5.4, which scored 75% on OSWorld-Verified — surpassing the average human score of 72.4% on desktop computer-use tasks.

China Blocks Meta’s $2 Billion Acquisition of Manus

China’s National Development and Reform Commission ordered the cancellation of Meta’s $2 billion acquisition of agentic AI startup Manus, the Chinese-founded company that relocated to Singapore in mid-2025. Beijing cited concerns over export controls, technology transfers, and the loss of key AI talent and technology to a U.S. tech giant.

The decision creates a practical nightmare for Meta: after announcing the deal in December 2025, the company had already integrated Manus into its internal systems and onboarded the startup’s executives. Meta maintains the acquisition “complied fully with applicable law.” The move underscores the growing geopolitical dimension of the AI race, with both Washington and Beijing increasingly treating AI talent and technology as national security assets.

SpaceX Secures $60 Billion Option to Buy Cursor

SpaceX, which absorbed Elon Musk’s xAI in a $1.25 trillion combined-valuation merger in February, struck a deal on April 21 giving it the right to acquire Anysphere’s Cursor for $60 billion later this year — or pay $10 billion for their joint work without completing the acquisition.

Cursor’s revenue run rate hit $2 billion in February and is projected to reach $6 billion by year-end. The deal gives Cursor access to xAI’s 550,000-GPU Colossus cluster and a foundation model partner that isn’t also a competitor — a rarity in today’s AI landscape where most model providers also sell coding tools.

ICLR Paper Reveals “The Reasoning Trap” in AI Agents

A paper presented at ICLR 2026 in Rio de Janeiro found that training language models for stronger reasoning through reinforcement learning increases tool-hallucination rates in lockstep with task gains. The authors built a diagnostic benchmark called SimpleToolHalluBench to measure hallucination in two failure modes: when no tool is available, and when only distractor tools are available.

The finding has major implications for the 96% of enterprises now running AI agents in production. The researchers showed that the model layers responsible for restraining bad tool calls are “exactly what gets trained away” during reasoning RL. Prompt engineering and DPO mitigate the problem partially, but neither closes the reliability gap — revealing a fundamental capability-versus-reliability trade-off.

19 New AI Laws Passed Across U.S. States

Since mid-March, 19 new AI bills have been signed into law across U.S. states, bringing the 2026 total to 25, with another 27 bills having passed both chambers. Notable new laws include Idaho’s framework for generative AI in K-12 education, New York’s regulatory framework for frontier AI developers, and new regulations in Indiana, Utah, and Washington prohibiting health insurers from using AI as the sole basis for denying claims.