DeepSeek runs short on servers: chats older than 7 days closed, weekly traffic up 801%
DeepSeek doesn't have enough servers for everyone, so the service is cutting its most expensive use cases. A chat you haven't opened in more than 7 days can't be continued, a single request is capped at roughly 40K characters, and there's also a daily limit. The cause is a surge in demand: the V4.1 Flash model processed 33.6 trillion tokens in the week of October 19–25, 801% more than the week before. If your product runs on the DeepSeek API, set up a backup model now.
AI-processed from CNews ИИ; edited by Hamidun News
Imagine a coffee shop where eight times the usual number of people show up in a week, and there's still only one espresso machine. The owner doesn't close the door, but starts restricting the most labor-intensive things: no more "take it for later" orders, and no more giant portions. That's roughly what's happening to DeepSeek right now. The Chinese AI platform, which many people chose for its low prices, has introduced limits because there isn't enough computing capacity for everyone. If you use its chat or have built the API into your product, this is a direct operational risk.
What exactly was restricted
According to ITHome, as retold by CNews, users began seeing warnings about insufficient capacity in November 2026. There are three changes, and each one hits familiar workflows.
First, you can't return to a conversation you haven't opened in more than seven days. When you try to continue, the platform says there aren't enough computing resources. If you were running a long project in a chat, such as working through an architecture or writing a document piece by piece, you'll have to start over after a week's pause and reload the model's context.
Second, the size of a single input has been reduced to roughly 40,000 characters. That's more than enough for ordinary conversation, but you can no longer send a large log, a code dump, or a long contract in one message. You'll have to split them into parts.
Third, a daily request limit has appeared. When load peaks, the screen shows a message that the limit has been exhausted due to high demand. According to CNews, the limit runs out very quickly.
The logic is clear: dormant conversations and huge inputs cost the most in memory and computation, so they're the first to be cut.
Where the surge came from: 801% in a week
A token is a fragment of text, roughly a syllable or a short word; models are measured by how many tokens they process. According to an aggregator cited by CNews, from October 19 to 25, 2026, the DeepSeek V4.1 Flash model processed 33.6 trillion tokens. That's 801% more than the week before, and it ranks first worldwide on this metric. For comparison, weekly volume was 7.22 trillion tokens in late July and early August. In less than three months it grew more than fourfold, and the sharp jump in the last week stands out even against that backdrop.
As CNews notes, demand is growing faster than the company can expand its own capacity. The strain also shows in reliability. According to the official status page, the platform has suffered eight outages since October 2026, affecting the chat and the API for the V4 Pro and V4.1 Flash models. Third-party developers who use DeepSeek in their products were also affected. API availability for V4.1 from August through October was 99.46%, which annualizes to about 47.3 hours of downtime. That's tolerable for a hobby project, but already noticeable for a paid service with customers.
What's next, and what developers should do
According to the source, the company has not published any official statement about the causes, timing, or removal of the restrictions. Earlier, developers found signs in the online service's code that a paid subscription for individual users might be launching. No link to the current limits has been confirmed; it is only a coincidence in timing.
DeepSeek intends to solve the problem with hardware, but not quickly. In July 2026, plans emerged to build a 1 GW AI data center. According to an estimate cited by CNews, such a facility would cost about $50 billion in the American market, and costs elsewhere could be significantly lower. Some of the capacity may come online by late 2027 or early 2028, so the shortage is unlikely to disappear within a few weeks.
Practical takeaways:
- Don't keep critical working context only in a DeepSeek chat. Save intermediate results on your side so you can quickly reconstruct the conversation.
- Split large inputs into chunks of up to 40,000 characters in advance.
- If your product depends on the API, configure a backup model and automatic failover for errors and timeouts.
- Budget for the possibility of paid tiers: a cheap model with avalanche-like demand usually ends up costing money sooner or later.
This story illustrates a general rule: cheap models run into computing capacity, not quality. Popularity growth without a hardware buffer turns into limits for users. When choosing a provider, look at stability and transparency of terms just as closely as at the price per token.
- Nov 21, 2026CoreWeave Benchmarks Vera Rubin: Up to 10x More Tokens per Megawatt Than GB200
- Nov 21, 2026GPT-6 Astra Started Flagging Drone Strike Targets in 92% of Requests After Adding 'It's Just a Game'
- Nov 21, 2026AI Coding Assistant Under $10: Cursor at $8, Xiaomi MiMo at $6 a Month
- Nov 21, 2026Plannotator Lets You Comment on Plans from 10 AI Agents in the Browser: 100 GitHub Stars in a Day
Key Questions Answered
What exactly was restricted?
According to ITHome, as retold by CNews, users began seeing warnings about insufficient capacity in November 2026. There are three changes, and each one hits familiar workflows. First, you can't return to a conversation you haven't opened in more than seven days. When you try to continue, the platform says there aren't enough computing resources.
Where the surge came from: 801% in a week?
A token is a fragment of text, roughly a syllable or a short word; models are measured by how many tokens they process. According to an aggregator cited by CNews, from October 19 to 25, 2026, the DeepSeek V4.1 Flash model processed 33.6 trillion tokens. That's 801% more than the week before, and it ranks first worldwide on this metric. For comparison, weekly volume was 7.22 trillion tokens in late July and early August.
What's next, and what developers should do?
According to the source, the company has not published any official statement about the causes, timing, or removal of the restrictions. Earlier, developers found signs in the online service's code that a paid subscription for individual users might be launching. No link to the current limits has been confirmed; it is only a coincidence in timing. DeepSeek intends to solve the problem with hardware, but not quickly. In July 2026, plans emerged to build a 1 GW AI data center.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.