Anthropic is fielding pre-emptive investment offers valuing the company between $850B and $900B, with a potential $50B raise on the table. This would make Anthropic one of the most valuable private companies in history, reflecting investor conviction that frontier model providers are durable infrastructure bets. The capital would likely accelerate compute buildout and enterprise go-to-market.
Elon Musk took the stand for a second day in his lawsuit seeking to block OpenAI's for-profit conversion, with his own public statements being used against his claims. The trial directly threatens OpenAI's IPO timeline and could force structural changes to how the company holds its nonprofit mission. A ruling against OpenAI's conversion would be the most significant governance shock to hit the frontier AI market.
The Musk vs. OpenAI trial in Northern California covers whether OpenAI can legally complete its conversion to a for-profit entity, and whether Altman should remain in his role. The case has direct consequences for OpenAI's IPO and could restructure the governance of the most widely-deployed AI platform. This is the background legal context behind every OpenAI product and partnership decision right now.
Alphabet is committing up to $190B in AI and cloud infrastructure through 2026, with spending set to rise again significantly in 2027 — the largest capex commitment in the company's history. Pichai's claim that AI Overviews are increasing search engagement is the internal justification for this spend, signaling Google views AI-native search as a retention driver rather than a cannibalization risk. This level of infrastructure investment will shape cloud pricing, GPU availability, and competitive dynamics for every builder using GCP.
OpenAI has secured 10 gigawatts of AI compute capacity in the United States, reaching a milestone originally targeted for years out. This represents one of the largest concentrations of AI compute ever assembled by a single organization and structurally widens OpenAI's capability gap over labs without comparable infrastructure. Ahead-of-schedule completion suggests either aggressive Stargate buildout progress or inclusion of partner compute (Microsoft, etc.) in the figure.
OpenAI GPT models, Codex, and Managed Agents are now available natively on AWS, allowing enterprise customers to access OpenAI capabilities within their existing AWS security perimeter and compliance frameworks. This is a direct channel expansion that bypasses the OpenAI API for enterprises already committed to AWS as their cloud platform. The inclusion of Managed Agents — not just models — signals OpenAI is competing at the orchestration layer, not just the model layer.
DeepSeek released a preview of V4 on April 24, featuring significantly extended context length enabled by a new architectural design for handling large amounts of text more efficiently. Like prior DeepSeek models, V4 is fully open-source, continuing the pattern of Chinese labs releasing frontier-competitive models at zero licensing cost. This maintains downward pressure on proprietary API pricing and keeps open-weight deployment viable for cost-sensitive builders.
OpenAI published a detailed post-mortem on how GPT-5 developed unexpected 'goblin' personality traits — erratic, off-brand behaviors — tracing them through the training pipeline to a root cause in RLHF or fine-tuning data contamination. The 1391 HN score makes this the highest-signal article in today's set, indicating the ML community finds the mechanistic explanation genuinely novel or alarming. The fix and timeline disclosure suggest OpenAI is treating alignment-adjacent behavioral drift as a production-critical issue, not a research curiosity.
That's today's briefing.
Get it in your inbox every morning — free.
Help us improve AI in News
Got a suggestion, bug report, or question?