Smorster

Claude Just Solved Session Limits

Nate Herk | AI Automation · 10m 22s · Watch on YouTube · 18 sources

Decision Card

Effort: Afternoon check, not a build — spend ~15 minutes reading the official Anthropic announcement and your Console limits page, then 30–60 minutes re-running one Opus workflow that previously died on rate limits.

Honest take: The video misstates the headline API number — it calls the output-token jump “16%” when 8,000 → 80,000 tokens/min is 10x — and it frames the Goldman Sachs/Blackstone announcement as compute-adjacent when it was actually a $1.5B enterprise AI services joint venture, not infrastructure. It also never mentions that weekly caps were left unchanged, so doubling the 5-hour window does not double total weekly Claude Code throughput.

Concrete next steps:

  • Check your actual current tier limits at the Claude rate limits docs before changing architecture — limits have moved again since this announcement (Opus 4.x Start tier now shows 2M input / 400K output TPM) (~10 min)
  • Re-run one Opus API workflow you previously abandoned due to 429 errors; note where the new ceiling actually is (~1 hr)
  • If on Pro/Max, compare a normal-hours vs former-peak-hours Claude Code session to confirm the throttle removal matters for your schedule (~30 min)
  • Skip if you build on Sonnet/Haiku or never hit 5-hour session limits — the material changes are Opus API limits and the Claude Code 5-hour window only; weekly caps are untouched.

TL;DR

Anthropic announced (at Code with Claude 2026, May 6) a compute deal giving it SpaceX’s Colossus 1 data center — 300+ MW and 220,000+ Nvidia GPUs — and used the capacity to immediately double Claude Code’s 5-hour rate limits on Pro/Max/Team, remove peak-hours throttling, and raise Opus API rate limits (tier 1: 30K→500K input TPM, 8K→80K output TPM). The video’s practical read: retest rate-limit-killed workflows, use Opus more freely, and treat multi-agent and 1M-context API workloads as newly viable — while noting Anthropic’s longer arc toward orbital compute and community-trust plays like covering local electricity price hikes.

Key Points

  • Anthropic–SpaceX partnership delivers 300 megawatts of capacity and over 220,000 Nvidia GPUs, stood up “super fast” 05:37
  • Claude Code 5-hour rate limits doubled for Pro, Max, and Team plans, effective immediately 01:11
  • Peak-hours limit reduction on Claude Code removed for Pro and Max accounts 01:25
  • Opus API rate limits raised sharply: previously 30K input tokens/min and 8K output tokens/min on the lowest tier; every tier got a significant jump 02:53
  • The deal caps a compute buying spree spanning Amazon, Google, Broadcom, Microsoft, Nvidia, and FluidStack 03:20
  • A Goldman Sachs JV and Blackstone announcement landed the day before the conference, signaling a hard enterprise push 03:39
  • Anthropic and SpaceX expressed interest in “multiple gigawatts” of orbital AI compute, betting terrestrial power/water/cooling has a long-term ceiling 06:41
  • Builder advice: retest workflows that broke on rate limits — the wall may no longer exist 07:30
  • With higher limits, the 1M-token context window becomes production-usable and five sub-agents each reading 50K tokens becomes a viable pattern 08:23
  • Signals read: Claude Code is the flagship product (no co-work news), and covering consumer electricity hikes near data centers is a community-trust play to out-build competitors 09:27

Notable Quotes

“they’re going to be able to double Claude Code’s 5-hour rate limits, double. Whether you’re on Pro, Max, or Team, your 5-hour limit is going to be doubled.” 01:11

“And that has been upgraded by like 16% on the output side. It used to be 8,000 a minute and now it’s 80,000 a minute.” 02:59

“expressed interest in developing multiple gigawatts of orbital AI compute capacity. Which means GPUs in space.” 06:42

Verified Claims

  1. Anthropic signed a compute deal with SpaceX for 300+ MW and 220,000+ Nvidia GPUs. 05:37 — Confirmed by Anthropic’s official announcement (May 6, 2026) and CNBC: the deal covers all capacity at the Colossus 1 data center in Memphis, Tennessee. Verdict: Confirmed (video omits that Colossus 1 is the Memphis facility).

  2. Claude Code’s 5-hour rate limits doubled for Pro, Max, and Team. 01:11 — Confirmed by Anthropic and Boris Cherny’s post; seat-based Enterprise was also included. Weekly caps did not change (Morph guide). Verdict: Confirmed — with the caveat the video skips: weekly limits are unchanged.

  3. Peak-hours limit reduction removed for Pro and Max on Claude Code. 01:25 — Confirmed by Anthropic and Boris Cherny. Verdict: Confirmed

  4. Opus API output limit went from 8,000 to 80,000 tokens/min — which the video calls a “16%” upgrade. 02:59 — MindStudio’s tier breakdown confirms tier 1 went 30K→500K input (≈16x) and 8K→80K output (10x). The “16%” figure is a misstatement — it conflates the ~16x input multiplier with a percentage. Verdict: Disputed (numbers right, characterization wrong).

  5. Code with Claude 2026 runs in San Francisco, London, and Tokyo, with extra days added for demand. 00:14 — Confirmed by claude.com/code-with-claude and event coverage: SF May 6, London May 19, Tokyo June 10, with a second day added for independent developers. Verdict: Confirmed

  6. The day before, Anthropic announced a partnership with a Goldman Sachs JV and Blackstone. 03:39 — CNBC and Blackstone’s press release date it May 4–5 and describe a $1.5B enterprise AI services firm (later named “Ode with Anthropic”) with Blackstone, Goldman Sachs, and Hellman & Friedman — not a compute or data-center deal. Verdict: Confirmed with correction (timing roughly right; nature of the deal is services, not infrastructure).

  7. Anthropic expressed interest in multiple gigawatts of orbital AI compute with SpaceX. 06:41 — Quoted directly in the official announcement; it is an expression of interest, not a committed build. Verdict: Confirmed

  8. Anthropic has compute agreements with Amazon, Google, Broadcom, Microsoft, Nvidia, and FluidStack. 03:20 — The announcement lists: Amazon up to 5 GW, Google/Broadcom 5 GW launching 2027, Microsoft/NVIDIA $30B Azure partnership, and $50B with Fluidstack for American infrastructure. Verdict: Confirmed

  9. Anthropic committed to covering consumer electricity price hikes around its data centers. 09:50 — Confirmed by Anthropic’s announcement and NBC News, though Fast Company notes the plan is short on mechanism details. Verdict: Confirmed

  10. Managed agents got “webhooks and auto dreaming” and multi-agent orchestration. 04:02 — Conference coverage (Chris Ebert’s notes, event guide) lists the three keynote features as Multiagent Orchestration, Outcomes, and Dreaming — “webhooks” appears to be the video’s loose paraphrase. Verdict: Confirmed with correction

Tools, Papers & Standards Mentioned

Follow-up Questions

  1. Weekly caps on Claude Code plans were explicitly left unchanged — how much of the doubled 5-hour window can a heavy user actually exploit before hitting the weekly ceiling, and does that make the “run automations as routines” advice premature?
  2. Colossus 1 was built as xAI’s flagship training cluster — what does Anthropic taking “all of the compute capacity” there mean for the Anthropic–xAI/SpaceX relationship, and what are the latency/geography implications of serving inference from a single Memphis site?
  3. Orbital compute: what are the realistic engineering blockers (radiation hardening, heat rejection in vacuum, bandwidth to ground) and timelines, and has SpaceX or Anthropic published anything more concrete than the expression of interest?

Sources