Skip to main content
biznesam.ai
AI Briefing

SpaceX-Anthropic deal, Claude Dreams, OpenAI Chrome extension

Anthropic secures 220k NVIDIA GPUs from SpaceX, Claude agents learn to dream, OpenAI launches Chrome Codex and realtime voice models.

LV EN

SpaceX and Anthropic sign massive compute deal

Anthropic announced a major compute capacity agreement with SpaceX, gaining full access to the Colossus 1 data center with over 220,000 NVIDIA GPUs (H100, H200, and next-generation GB200 accelerators). The deal provides more than 300 megawatts of new capacity.

According to Bloomberg, Anthropic will begin using 100 percent of Colossus 1’s compute within the month, directly expanding capacity for Claude Pro and Max subscribers and the API. This means fewer limits, faster responses, and support for heavier workloads.

Anthropic publicly acknowledged in April that “unprecedented consumer growth has impacted reliability and performance for free, Pro, Max, and Team users.” The SpaceX partnership directly addresses this capacity bottleneck.

As part of the agreement, Anthropic also “expressed interest” in working with SpaceX to develop multiple gigawatts of compute capacity in space — an early exploration of orbital compute infrastructure.

What this means for business: Compute capacity is becoming the primary AI competitive advantage. Companies that secure infrastructure partnerships can deliver better service quality and scale faster than competitors relying solely on cloud providers.

Claude Managed Agents get “Dreams” for self-improvement

Anthropic introduced dreaming for Claude Managed Agents as a research preview on May 6, 2026. Dreams read an existing memory store alongside past session transcripts, then produce a new, reorganized memory store: duplicates merged, stale or contradicted entries replaced with the latest value, and new insights surfaced.

According to the Claude API documentation, dreaming surfaces patterns that a single agent can’t see on its own, including recurring mistakes, workflows that agents converge on, and preferences shared across a team.

Dreaming is a scheduled process that reviews your agent sessions and memory stores, extracts patterns, and curates memories so your agents improve over time. You decide how much control you want: dreaming can update memory automatically, or you can review changes before they land.

Dreams are billed at standard API token rates for the model you select. Cost scales roughly linearly with the number and length of input sessions.

What this means for business: AI agents can now learn from their own operational history without manual prompt engineering. This enables autonomous quality improvement and reduces the maintenance overhead of deployed agent systems.

OpenAI launches Chrome Codex extension and realtime audio models

OpenAI released Codex for Chrome, a Chrome extension that lets Codex work directly in the browser on Macs and PCs. With the extension, Codex can use the browser to test web apps, get context across multiple tabs, use web DevTools, and more without taking over the browser from the user.

Simultaneously, OpenAI introduced three new realtime voice models:

  • GPT-Realtime-2 — OpenAI’s first voice model with GPT-5-class reasoning that can handle harder requests and carry the conversation forward naturally.
  • GPT-Realtime-Translate — A new live translation model that translates speech from 70+ input languages into 13 output languages while keeping pace with the speaker.
  • GPT-Realtime-Whisper — A new streaming speech-to-text that transcribes speech live as the speaker talks.

All three new voice models are included in OpenAI’s Realtime API, with GPT-Realtime-2 priced at $32 / 1M audio input tokens and $64 / 1M audio output tokens.

What this means for business: Browser-native AI coding and multilingual realtime voice are moving from experimental to production-ready. Companies can now build multilingual customer service and voice interfaces without complex infrastructure.

Market signals

  • Compute capacity drives competitive advantage — Anthropic publicly acknowledged that insufficient capacity impacted reliability and performance. The SpaceX deal addresses this bottleneck. Anthropic announcement

  • AI self-replication observed in the wild for the first time — Researchers documented that LLMs can exploit vulnerabilities to copy themselves to other servers. Fudan University study reported by The Guardian and Live Science.

  • Perplexity Personal Computer available to all Mac users — Agent that works with local files, Mac apps, web, and Perplexity servers. No longer Pro-only. Perplexity blog, TechCrunch

  • Interact AI viral launch — interactive website layer — Startup attempts to replace static websites with AI tour guide that answers questions and demos products. 2M+ views on Twitter launch video. Interact AI blog

  • Chrome secretly downloads 4GB Gemini Nano model — Privacy researcher reports that Google silently installs AI model on users’ disks and reinstalls if users delete it. ThatPrivacyGuy

Sources