Anthropic is simultaneously the most wanted AI company on Earth and the one that cannot stop making its own customers angry. The contradiction is getting absurd.

In the span of two weeks, Anthropic has cut off power users from flat-rate Claude subscriptions, throttled usage limits during peak hours, suffered noticeable service interruptions, and, according to a growing chorus of developers, quietly degraded Claude's core capabilities. Fortune, VentureBeat, and Slate all ran stories this week documenting the backlash. The developer community is furious.

"Claude has regressed to the point it cannot be trusted to perform complex engineering," one developer wrote in a widely circulated complaint. An AMD Senior Director filed a detailed GitHub issue cataloging performance degradation. Reddit threads and X posts from paying subscribers describe a model that has become slower, dumber, and less reliable in real-time.

The $380 Billion Company That Cannot Keep Its Customers Happy

Here is what makes this genuinely bizarre. While users are posting screenshots of Claude refusing to complete basic coding tasks, investors are throwing money at Anthropic faster than the company can say no. Business Insider reported this week that VCs have offered valuations as high as $800 billion. The company tripled its revenue to $30 billion. It just put a Novartis CEO on its board. It is winning the enterprise market at a pace that is making OpenAI investors nervous.

And yet the people actually using the product every day say it is getting worse.

Slate nailed the paradox: Anthropic stood up to Sam Altman and the Pentagon, earning massive public goodwill. Then it turned around and started treating its own users like an afterthought. It blocked OpenClaw, the open-source agent framework, from using flat-rate Claude subscriptions. It throttled heavy users. When people complained, the company responded with corporate non-answers that made things worse.

The OpenClaw creator, who now works at OpenAI, put it bluntly. When someone suggested he should have joined Anthropic instead, he replied: "One welcomed me, one sent legal threats."

What Is Actually Happening

The most likely explanation is the boring one: Anthropic is running out of compute. The company is distributing Mythos to governments and banks for cybersecurity. It is serving enterprise customers who are burning through tokens at unprecedented rates. Meta employees alone consumed 60 trillion tokens from Anthropic in a single month. Claude Code is being used by developers for everything from vibe coding to full application architecture.

Something has to give. When demand outstrips supply, the company has two choices: raise prices or degrade service for the users generating the least revenue. Anthropic appears to be doing both, but without telling anyone that is the strategy.

That is the mistake. Users can handle "we are capacity-constrained and working on it." What they cannot handle is pretending nothing changed while the product gets worse. Anthropic is running the same playbook that destroyed trust in every other tech company: gaslight the users, serve the enterprise clients, and hope nobody notices.

The OpenAI Opportunity

OpenAI is paying attention. It just launched a $100 ChatGPT Pro subscription targeting the exact developers Anthropic is alienating. Sam Altman declared Code Red on Anthropic internally and told his team to copy the features that made Claude dominant in coding workflows. OpenAI even hired the OpenClaw creator. That move now looks less like a talent grab and more like a strategic middle finger.

The window for OpenAI to recapture defecting Claude users is narrow but real. If Anthropic keeps treating its power users like a cost center while chasing government contracts and billion-dollar valuations, those users will find alternatives. They already are.

Anthropic has six months to fix this before the goodwill from standing up to the Pentagon evaporates entirely. Based on the last two weeks, that clock is moving fast.