The Omniscient Bulletin · 2026-05-09
The Omniscient Bulletin — May 9, 2026
Sutskever and Nadella take the stand in Oakland, OpenAI's voice models go GA with GPT-5-class reasoning, and Anthropic's $200B commitment now anchors 40% of Google Cloud's revenue backlog.
Trial week three opens Monday with the witnesses both sides have been waiting for: Sutskever, the founder who voted to fire Altman, and Nadella, whose reaction to that firing forced the reversal. Beyond the courtroom, the labs spent the week locking in the next phase of compute and capital - Anthropic's $200B Google commitment, OpenAI's voice models going GA, Meta's quiet Android-style move into humanoid robotics - while a contested ARR figure puts Anthropic ahead of OpenAI on revenue for the first time.
Industry
Trial week 3 hits the marquee witnesses: Sutskever and Nadella
Week three of Musk v. Altman opens Monday in Oakland with two marquee witnesses - former OpenAI chief scientist Ilya Sutskever, who voted in November 2023 to fire Altman before publicly expressing regret and calling for his reinstatement, and Microsoft CEO Satya Nadella, whose $13B+ investment and reaction to the firing reportedly forced the reversal. Sutskever is the trial's closest thing to direct evidence on OpenAI's mission drift; Nadella will be probed on the precise contours of Microsoft's leverage over Altman.
Frontier
OpenAI's Realtime API goes GA with GPT-5-class voice reasoning
OpenAI took its Realtime API out of beta Thursday with three new audio models. GPT-Realtime-2 is the first voice model with GPT-5-class reasoning and a 128K context window (up from 32K). Companions: GPT-Realtime-Translate handles live speech translation across 70+ input languages and 13 output languages at $0.034/min, and GPT-Realtime-Whisper streams transcription at $0.017/min with controllable latency. The GA - and pricing transparency - reframe voice as primary surface area rather than a wrapper around text. Worth watching which agent vendors fold these in fastest.
Industry
Anthropic's $200B Google commitment now anchors 40% of Cloud's revenue backlog
The Information reported Tuesday that Anthropic has committed $200B to Google Cloud over five years, on top of an April deal with Google and Broadcom for multiple gigawatts of TPU capacity coming online in 2027. The spend reportedly accounts for more than 40% of the revenue backlog Google disclosed to investors - a single AI customer now anchors Google Cloud's growth narrative. Combined with Amazon's $25B, the SpaceX/Colossus 1 capacity locked last week, and CoreWeave, Anthropic has more compute optionality than any frontier rival - and Google has more concentration risk than any hyperscaler.
CNBC (citing The Information)→
Industry
Meta buys Assured Robot Intelligence in Android-style humanoid platform play
Meta closed its acquisition of Assured Robot Intelligence on May 1, folding co-founders Lerrel Pinto (NYU, formerly Fauna Robotics) and Xiaolong Wang (UCSD, formerly Nvidia) into Superintelligence Labs. The pitch: rather than ship a Meta-branded humanoid, license robotics software as an Android-style platform other manufacturers run. ARI's specialty is whole-body humanoid control - the layer that lets a robot walk, balance, and adapt without scripted gaits. Meta is betting the platform layer beats the device layer for owning the next compute surface, the same wager Google won with Android.
Industry
Anthropic's $30B ARR reportedly tops OpenAI - but the accounting is contested
Anthropic's annualized revenue reportedly cleared $30B last month, ahead of OpenAI's $24B for the first time, with ~80% enterprise customers and enterprise spend over $1M/year doubling from roughly 500 to 1,000 in about two months. OpenAI disputes the figure, arguing Anthropic overstates by roughly $8B by recording AWS and Google Cloud revenue at gross rather than net of partner share. Anthropic counters that it's the principal in those transactions: it sets the price, the clouds are distribution. The accounting question matters now that an IPO is reportedly months away, not years.