Claude’s 85-Minute Public Outage Exposes a Widening Two-Tier Reliability Gap

Anthropic’s public AI service went dark for roughly 85 minutes on June 23. Government users? They didn’t notice a thing.

The disruption hit early that morning and touched pretty much every consumer-facing product Anthropic runs — claude.ai, the Claude API, Claude Code, the Console, and Cowork all took the hit. Error rates climbed across several models, and users started feeling it fast. Anthropic’s engineers spotted the problem and pushed a fix within 35 minutes of the initial alert, but full resolution took just over two hours from the moment the first alarm went off. Meanwhile, the government service kept running without a hiccup, operating in its own isolated environment, completely walled off from the consumer chaos.

Not a minor gap.

The numbers behind the two tiers are kind of striking when you put them side by side. The government service has held 99.93% uptime over a 90-day window. The public claude.ai tier sits at 99.1% over the same stretch. That sounds close until you do the math — 99.1% over 90 days works out to roughly 19 hours of downtime for public users. The government tier? About 90 minutes over that same period. And the June 23 outage wasn’t a one-off. Anthropic’s own status page logged more than 20 separate outages or incidents between June 9 and June 23 alone, many of them tied to its Opus 4.8 model. Paying subscribers who’ve already been squeezed by recent usage limits were not exactly thrilled.

How the Government Tier Stays Isolated

The separation isn’t accidental. Anthropic runs its government service through Palantir’s federal cloud infrastructure, which carries FedRAMP High authorization — one of the strictest security and reliability certifications the U.S. government requires for cloud services. The General Services Administration extended that service to all government branches last year, locking in a dedicated, regulated environment that doesn’t share resources with the consumer side.

That’s basically the whole point. When public systems buckle under load or hit technical snags, the federal environment stays clean because it’s not drawing from the same pool. No resource contention, no spillover. The government clients get their own lane, and it’s a fast one.

Public users don’t get that lane. And they’re starting to notice the difference.

See also: Trump Signs Quantum Computing Orders With No Budget Details Yet

Anthropic’s Infrastructure Push

To be fair, Anthropic isn’t sitting still. The company has secured agreements for up to 5 gigawatts of additional compute capacity from Amazon and is actively leasing more data center space. Those are big moves — 5 gigawatts is a serious commitment, the kind of infrastructure bet that takes months to actually come online. But here’s the catch: none of that new capacity flows into the government tier. It’s aimed at the consumer side, where the strain is most visible. The government infrastructure stays separate, stable, and unaffected by whatever growing pains hit the public service.

So the expansion plans are real, but they don’t change the fundamental architecture. Two tiers. Two sets of expectations. Two very different uptime records.

It’s worth stepping back and thinking about what 20-plus incidents in 14 days actually means for someone who relies on Claude for daily work. Developers building on the API, businesses running workflows through Claude Code, individual subscribers paying for consistent access — they’ve all been absorbing interruptions at a rate that’s genuinely hard to ignore. The June 23 outage added roughly another hour and a half to that tally.

The frustration is probably sharpest among users who’ve also run into the usage limits Anthropic rolled out recently. Getting throttled on top of outages is a rough combination, and it’s pushed some users to publicly question where they fall in the company’s priority stack.

See also: H100 Acquires Two Norwegian Firms, Eyes Top 2 Spot in European Bitcoin Holdings

The answer, based on the architecture, seems pretty clear. Federal contracts come with strict uptime obligations, regulatory requirements, and FedRAMP compliance demands that Anthropic has to meet. Consumer subscriptions don’t carry the same contractual weight, even if the users behind them are just as dependent on the service.

And so the gap holds. Government tier at 99.93%. Public tier at 99.1%. Over 20 incidents logged in two weeks. The June 23 outage lasted 85 minutes for public users and exactly zero minutes for federal clients.