Realtime AI News
Nikkei: DeepSeek hit by five-hour outage as V4-Pro and V4.1-Flash go down
Nikkei Asia reports that DeepSeek suffered an outage of around five hours on Thursday, with its V4-Pro and V4.1-Flash models and chat services unusable from 8 a.m. to 1 p.m. The company has not identified a cause, some point to insufficient computing resources, and paid subscribers received credit refunds for services affected between Oct. 8 and Oct. 14.
DeepSeek suffered an outage of about five hours on Thursday, local time, according to Nikkei Asia: from 8 a.m. to 1 p.m., the Chinese AI startup's V4-Pro and V4.1-Flash models and its chat services were unusable.
The company has not identified a cause, but some point to insufficient computing resources as a likely factor. Chinese media have also reported cases where computing resources did not match usage volumes, and where problems stemmed from how processing power was allocated.
For users, the most direct remedy has been compensation. Paid subscribers received partial refunds in the form of credits for services affected between Oct. 8 and Oct. 14, which can be used for future services, and Chinese media reported examples of refunds worth 3,000 yuan, or about $450.
This is not the first stability problem for DeepSeek. Nikkei notes that when the company launched its R1 model, which it said offered high performance at low cost and which briefly sent global technology stocks lower, it suffered a wave of issues in January and February 2025. In 2026, problems increased noticeably after the July and August official releases of V4-Flash and V4-Pro, and this week's outage appears to be among the more serious cases.
A rise in usage after the Sept. 10 launch of the lower-cost V4.1-Flash may also have played a role, the report says, as cheaper models drive call volumes and make any mismatch between compute and services more visible during peak hours.
There is a compute reality behind the outage as well. DeepSeek is expected to soon complete a round of outside fundraising that could reach 100 billion yuan, with Chinese investment funds and internet conglomerate Tencent Holdings among the investors. The money will likely go toward computing resources, including chip purchases, which must come from domestic suppliers because of U.S. restrictions on exports of high-performance AI semiconductors to China.
So the five-hour outage is not only a product reliability incident; it exposes the tension between surging demand and constrained access to compute.
What to watch next is whether DeepSeek publishes a root cause, whether fresh compute capacity after the fundraising eases pressure at peak times, and whether credit-based refunds become a standing part of its subscription service.
Why it matters
The outage and the credit refunds turn 'not enough compute' into a visible service-quality problem, and until export controls and its fundraising land, DeepSeek's reliability will track the pace of chip and capacity allocation.
Nearby Updates
All10/16, 23:17
17-year-old Kimi researcher Chen Guangyu says he has left Moonshot AI
Chen Guangyu, a post-2000 developer known for co-authoring Kimi's Attention Residuals technical report at 17, said on social media on Oct. 15 that he had left Moonshot AI's Kimi team, according to a Southern Metropolis Daily report on Oct. 16. The report does not say why he left or where he is going next.
10/16, 23:33
Vercel data: TypeSafe AI's Jev is its fastest-adopted model ever
Data published by Vercel on Oct. 15 shows that Jev, a decision model built by TypeSafe AI, was used by 35% of paid teams within two weeks of its limited release, nearly double the reach of the second most popular model, Sonnet 5.5. The same dataset shows open source models now handle more than 70% of tokens on the gateway, while frontier lab spending grew tenfold.
10/16, 22:49
Microsoft Finishes Pre-Training MAI-Base-2, a 2.8T-Parameter Model It Says Beats DeepSeek on Efficiency
Microsoft AI said on October 15 that it completed pre-training of MAI-Base-2, a mixture-of-experts model with 2.8 trillion total parameters and 80 billion active per token, trained on 16,000 Nvidia GB200 GPUs over 38 days. Microsoft claims its recipe is 4.2 to 12.3 times more FLOPs-efficient than DeepSeek-V4-Pro-Base, but the model is still in post-training with no release announced.
10/17, 00:00
Nanchang's Donghu District Opens 'Token' Industry Center, Offering AI Compute on Demand
A Token industry operations center in Nanchang's Donghu District has officially opened, with local reporting describing AI compute as usable on demand in the way water and electricity are. The facility is framed as industrial infrastructure for large-model workloads.