ENFR
8news

Tech • IA • Crypto

TodayMy briefingVideosTop articles 24hArchivesFavoritesMy topics

Musk–Anthropic GPU deal, GPT Realtime 2, Gemini Android shift

AIThursday, May 14, 2026· 13 videos

Briefing

0:00 / 0:00

Musk–Anthropic compute deal shocks market

Elon Musk agreed to supply Anthropic with access to the Colossus 1 cluster, featuring 220,000 Nvidia GPUs and 300 MW capacity. The deal could generate $3–4 billion in revenue and up to $2.5 billion in profit, signaling massive demand for compute. It marks a sharp reversal after Musk’s earlier criticism of the company. The move underscores how infrastructure alliances now outweigh ideological divides in AI competition.

Compute replaces models as bottleneck

Access to large-scale compute is now the निर्णing constraint across AI development. Revenue growth among top labs increasingly tracks directly with available GPU capacity. Even advanced models face strict usage limits without sufficient infrastructure. The shift elevates data centers, energy, and chips to the core of competitive strategy.

Altman testimony intensifies OpenAI lawsuit

Sam Altman testified in federal court, denying he misled Elon Musk about OpenAI’s transition from nonprofit to for-profit structure. He acknowledged “misunderstandings” but rejected claims of intentional deception. The lawsuit also implicates Microsoft, alleging its partnership enabled the shift. The case now centers on whether commercialization was necessary or a breach of founding principles.

Musk control claims face contradictions

Testimony revealed Elon Musk explored integrating OpenAI into Tesla, complicating his legal stance. Discussions included potential leadership roles for Sam Altman within Tesla’s AI efforts. These disclosures challenge Musk’s portrayal as solely defending nonprofit governance. The القضية now reflects deeper conflicts over control, not just mission.

OpenAI launches GPT Realtime 2

OpenAI introduced GPT Realtime 2, a voice-first model with ~200 ms latency and 128,000-token context. It supports 70+ input languages and integrates real-time reasoning with tool use. The system replaces traditional speech pipelines with a native voice-to-voice architecture. This enables more natural, continuous conversations and real-time task execution.

Google embeds Gemini across Android

Google is transforming Android into an AI-first platform powered by Gemini Nano. High-end devices with around 12 GB RAM can run on-device models for privacy and speed. New capabilities include autonomous multi-step task execution and contextual screen understanding. Smartphones are repositioned as active AI agents rather than passive interfaces.

DeepMind taps EVE Online sandbox

Google DeepMind has taken a stake in CCP Games, now rebranded as Fenris, to use EVE Online as a training environment. The MMO’s complex economy and large-scale coordination make it ideal for AI experimentation. Research will occur in controlled environments, not the live game. The initiative targets long-term planning and multi-agent decision-making capabilities.

Hermes agent overtakes OpenClaw usage

Hermes, developed by Noose Research, reached 224 billion daily tokens on May 10, 2026, surpassing OpenClaw. The agent grew rapidly after its February 2026 launch, gaining 147,000 GitHub stars. Its design centers on a self-improving loop with layered memory systems. The surge signals rising demand for persistent, learning AI agents over stateless assistants.

Videos covered

Previous briefings · AI