Why Claude Keeps Going Down — Inside the April Outages and Where the LLM Race Stands

The Claude Outage Pattern That Emerged in April

Claude went down — properly down — more than a few times in April. Sessions stalled mid-prompt, logins failed, and Claude Code workflows broke partway through. For anyone running daily work through Anthropic's stack, the Claude outage pattern stopped feeling like coincidence and started feeling like a trend.

This post lays out what actually happened in April, how Anthropic explained it, and what OpenAI and Google were doing in parallel. The framing is deliberately neutral — no team to root for, just the picture as it stands. If your workflow leans heavily on one model, an earlier piece on AI tool dependency pairs well with this one.

April Claude Outage Timeline

Pulling from Anthropic's own status page (status.claude.com), the major April incidents looked like this:

Date	Scope	Notes
April 15	Claude.ai and Claude Code login failures, API briefly down	Downdetector peak 9,000+ reports
April 16	Opus 4.6 elevated errors	~1h 23m duration
April 20	File upload failures, broad user impact	~6,500 users affected
April 22–23	API structured outputs degraded	~9h 49m duration
April 28	Claude.ai down, API auth errors	Major Outage, ~78 minutes
April 29	Haiku 4.5 and Opus 4.7 elevated errors	Multiple short incidents

Status aggregator IsDown counted 138 Claude incidents in the trailing 90 days — 43 major outages and 95 minor — with a median resolution time of about an hour. Anthropic's own published 90-day uptime sat at 99.03% for the API and 99.2% for Claude Code.

Anthropic's Own Explanation for the Claude Outage Streak

In a statement reported by Fortune, Anthropic said demand for Claude had grown at an unprecedented rate and that infrastructure had been stretched, particularly during peak hours. The company added that compute was an industry-wide constraint and pointed to expanded partnerships with Amazon and Google as the path to bringing new capacity online.

The context behind that statement:

• Anthropic's annualized revenue run rate was reported at around $30 billion — more than triple where it sat at the end of last year
• In late February, after OpenAI's $200 million Pentagon contract announcement drew user backlash, a wave of users moved toward Claude, briefly pushing the Claude app to #1 in the U.S. App Store
• OpenAI's chief revenue officer characterized Anthropic's compute posture as a "strategic misstep" in an internal memo reported by CNBC. Anthropic acknowledged the compute squeeze but framed it as an industry condition rather than an isolated mistake

The compressed version: Claude's runaway popularity outpaced its infrastructure, and April was when that gap became publicly visible.

What Users Can Do About the Claude Outage Risk

Stability will likely take time. For anyone whose workflow depends on Claude, a few practical steps are worth setting up:

• Subscribe to status.claude.com — email and Slack webhook alerts surface recovery times faster than refreshing manually
• Keep API key authentication ready — the API often stayed up even when Claude.ai itself was unreachable
• Pre-stage a backup LLM — having prompts and context ready to run on GPT-5.5 or Gemini 3.1 Pro means outages cost minutes, not hours

A previous comparison of Codex vs Claude Code is worth revisiting when choosing a backup.

Meanwhile at OpenAI

OpenAI shipped GPT-5.5 and GPT-5.5 Pro on April 23. With GPT-5.4 having launched only about six weeks earlier, the release cadence had clearly tightened. OpenAI positioned GPT-5.5 around agentic coding, computer use, data analysis, and document and spreadsheet work, citing internal benchmarks like 82.7% on Terminal-Bench 2.0 (versus 75.1% for GPT-5.4) and 81.8% on CyberGym (versus 73.1% for Claude Opus 4.7).

The company also disclosed scale figures alongside the launch: more than 900 million weekly active ChatGPT users, over 50 million paying subscribers, and 4 million active Codex users. The framing across the briefing leaned toward a future "super app" — a unified surface across ChatGPT, Codex, and the AI browser.

Why Google Stock Took Off

Alphabet (GOOGL) moved sharply in the same window. As of late April, the stock was up roughly 78% year-to-date and 118% over the trailing twelve months. The Q1 2026 earnings released on April 29 supported the rally:

• Quarterly revenue of about $109.9 billion, up 22% year over year
• Google Cloud revenue of around $20 billion, up 63% year over year
• First-party models processing roughly 16 billion tokens per minute, up 60% quarter over quarter
• 2026 capex guidance of $180 to $190 billion — close to double the prior year

On the model side, Gemini 3.1 Pro launched in February and scored 77.1% on ARC-AGI-2, more than doubling Gemini 3 Pro's mark. Google Cloud Next 2026 on April 22 added a deeper Gemini integration across Workspace, two new TPU chips (TPU 8t and 8i), and a confirmed Gemini–Apple Intelligence partnership for the upcoming Siri overhaul.

The Google-Anthropic Deal Worth Up to $40 Billion

The single biggest market signal was the Google-Anthropic deal worth up to $40 billion. Reports indicated that Google committed $10 billion upfront with another $30 billion tied to performance milestones, alongside an agreement giving Anthropic multiple gigawatts of TPU capacity. In effect, Anthropic's compute shortage was being converted directly into Google infrastructure revenue.

What the Claude Outage Pattern Really Showed — Compute Is the Gap

The variable that mattered most across the month was not model intelligence — it was who had stable compute and who didn't. Anthropic was visibly working to close the gap between Claude demand and its infrastructure. OpenAI was pushing rapid model iteration to keep users locked in. Google was positioning itself as the underlying capacity layer for both, while shipping its own frontier model.

For users, the safest short-term move was reducing single-model dependency. Working alongside AI like a colleague rather than a vending machine only works when the colleague is reachable — and April made it clear that no single provider could guarantee that yet. The Claude outage pattern was, in that sense, a fairly direct signal about where the market stands.