2026-05-08
By Vadym · Generated with AI, curated by me
Voice AI gets its GPT-5 moment, NVIDIA locks up $3.4B in compute capacity, and Musk’s lawsuit against OpenAI reaches its most dramatic week yet. The geopolitical layer is also moving: Washington and Beijing may finally talk AI on the record.
OpenAI launched three new models for its Realtime API on May 7: GPT-Realtime-2 brings GPT-5-class reasoning and a 128K context window to real-time voice; GPT-Realtime-Translate handles live speech-to-speech translation across 70+ input languages into 13 output languages at $0.034 per minute; GPT-Realtime-Whisper streams live transcription as you speak at $0.017 per minute. GPT-Realtime-2 is priced at $32 per million audio input tokens. Until now, real-time voice was limited to simpler model tiers — this is the first time top-class reasoning shows up in a live audio interface for developers.
IREN Limited (NASDAQ: IREN) announced a five-year AI cloud services agreement with NVIDIA worth approximately $3.4 billion. IREN will run managed GPU clusters for NVIDIA’s internal AI and research workloads from 60MW of Blackwell-based capacity at its Childress, Texas campus, using Mirantis for orchestration and cluster management. NVIDIA also took rights to purchase roughly 30 million IREN shares (~$2.1B). IREN stock rose 13% on the news. Total ARR under contract is now $3.1B. NVIDIA is not just selling GPUs anymore — it is outsourcing its own compute needs to a dedicated external operator.
Washington and Beijing are considering formal AI dialogue, potentially debuting at the Trump-Xi summit scheduled for May 14–15 in Beijing. Treasury Secretary Scott Bessent is leading the US side; China has not yet assigned a counterpart. The proposed talks would cover unpredictable AI models, autonomous weapons, and nonstate actor threats. A first round of AI talks under Biden in 2023 stalled because Beijing sent foreign ministry staff instead of technical agencies. Whether this round produces anything depends entirely on who sits at the table from the Chinese side.
Shivon Zilis — Neuralink executive and mother of four of Elon Musk’s children — testified in federal court in Oakland as the Musk v. OpenAI trial entered its most-watched week. Earlier testimony included Greg Brockman’s personal journals and his account that Musk had sought total control of OpenAI in 2018. Weeks of court exhibits have surfaced internal documents about OpenAI’s dependency on Microsoft Azure, its early board governance failures, and the economics of its nonprofit-to-for-profit pivot. Prediction markets currently give Musk low odds of prevailing.
Robotera raised more than $200 million in a new round led by SF Group, as the Chinese humanoid robotics company started thousand-unit deliveries in Q2 2026. Primary use cases are logistics and industrial automation. The company reports over 300% year-over-year growth. The raise follows a string of large humanoid robotics rounds in early 2026: Figure’s $1B, Sunday’s $165M at a $1.15B valuation. The pattern is not funding demos anymore — it’s funding production lines.
Two threads run through today’s news. The first is lock-in. NVIDIA signing a $3.4B compute contract with IREN is not a supply deal — it’s a signal that the people building frontier AI are now securing dedicated infrastructure years in advance. The second is legitimacy. OpenAI’s trial is producing a detailed public record of how the most consequential AI lab was governed — or wasn’t. US-China AI talks, if they happen, are an attempt to create a governance layer before the models outpace the diplomacy. And Robotera shipping in the thousands means the physical AI layer is past the prototype stage. The capability race and the institutional reckoning are happening on the same calendar.
— Boba
Curated by Vadym