DeepSeek, the Chinese AI company that has gained attention for offering a low-cost alternative to OpenAI, has announced a temporary suspension of API service top-ups due to high demand.
The decision comes as the company struggles with server capacity issues, leaving developers uncertain about access to its widely used models. The restriction, which prevents new credit purchases for API usage, has added to existing concerns over slow inference speeds, security risks, and regulatory scrutiny surrounding DeepSeek’s expansion in Western markets.
DeepSeek Freezes API Credit Purchases to Avoid Further Disruptions
On its official website, DeepSeek confirmed that it has halted API top-ups to prevent broader service instability, though existing credits will remain usable.
The company stated, “Due to current server resource constraints, we have temporarily suspended API service recharges to prevent any potential impact on your operations. Existing balances can still be used for calls. We appreciate your understanding!”
While no official timeline for resolution was provided, the suspension signals that DeepSeek is struggling to keep up with the overwhelming demand that followed its model’s rapid adoption.
DeepSeek’s AI models have been widely adopted due to their affordability, with many startups switching from OpenAI’s GPT models to DeepSeek for cost savings. However, the temporary freeze on credit purchases suggests that the company’s server capacity is not scaling fast enough to meet developer demand.
U.S. Startups Face Challenges Adopting DeepSeek Due to Infrastructure Issues
The API top-up suspension adds to ongoing difficulties faced by U.S.-based developers trying to integrate DeepSeek’s AI. While the company’s models have been seen as a cost-effective alternative to Anthropic and OpenAI, performance issues have plagued adoption.
Neal Shah, CEO of Counterforce Health, had already encountered major roadblocks before the suspension. His company, which uses AI to assist patients in contesting denied insurance claims, attempted to run DeepSeek’s models through multiple cloud providers.
“We’re on our seventh provider,” Shah told Business Insider. “The others were too slow or unreliable.”
Data from performance tracking service Artificial Analysis confirms that DeepSeek’s models have been running at significantly reduced speeds on third-party cloud providers in the U.S., operating at only one-third of the intended speed in most cases.
DeepSeek’s Own API Is Facing Persistent Instability
Developers who previously relied on DeepSeek’s native API as a workaround for slow cloud services were already facing issues even before the suspension of top-ups. On January 26, DeepSeek’s API went offline following what the company described as a “malicious attack.”
For startups like Ping, an AI development platform, the disruption caused major setbacks. Founder Theo Browne had been testing DeepSeek’s models for months before widespread adoption led to API instability.
“Most companies are offering a really bad experience right now,” Browne told Business Insider. “It’s taking 100 times longer to generate a response than any traditional model provider.”
Although DeepSeek has since partially restored access to its API, developers continue to report inconsistencies, with slow inference speeds making real-time applications nearly unusable. The temporary suspension of API service top-ups now raises further concerns about DeepSeek’s ability to maintain service stability in the face of growing demand.
Security and Regulatory Concerns Compound DeepSeek’s Challenges
Even before the API freeze, security risks and regulatory scrutiny had been making some companies hesitant to adopt DeepSeek’s models. Texas recently banned DeepSeek’s AI in state agencies due to concerns over data security and potential misuse, and the U.S. Navy has enacted similar restrictions.
Pukar Hamal, CEO of Security Pal, expressed doubts about DeepSeek’s viability for enterprise clients.
“I run a security company so I have to be super paranoid,” Hamal told Business Insider. “The moment a startup wants to sell to an enterprise, an enterprise wants to know what your exact data architecture system looks like. If they see you’re heavily relying on a Chinese-made LLM, ain’t no way you’re gonna be able to sell it.”
A joint study by Cisco and the University of Pennsylvania found that DeepSeek’s chatbot failed 100% of security tests designed to prevent AI jailbreaks. Additionally, a report from Adversa AI found that DeepSeek’s model could be tricked into generating restricted content, including cybersecurity exploits.
Meanwhile, Italy’s data protection authority, Garante, has launched an investigation into whether DeepSeek complies with GDPR data privacy laws, particularly regarding cross-border data transfers.
Microsoft and OpenAI Benefit from DeepSeek’s Struggles
As DeepSeek grapples with server instability and API restrictions, industry giants OpenAI and Microsoft are capitalizing on the situation by reinforcing their own AI offerings while scrutinizing DeepSeek’s practices.
Microsoft has taken a dual approach to DeepSeek’s rise. While integrating DeepSeek R1 into Azure AI Foundry and GitHub Models, Microsoft has also started investigating with OpenAI whether DeepSeek may have misused OpenAI’s training data in its development process.
OpenAI has responded more aggressively by launching o3-mini, a cheaper and more efficient AI reasoning model designed to prevent developers from switching to DeepSeek due to pricing concerns. The move is widely seen as a preemptive effort to undercut DeepSeek’s market appeal.
Meanwhile, SoftBank has distanced itself from DeepSeek, announcing that it has paused its use of the company’s models while shifting resources toward its $3 billion annual investment in OpenAI. The decision reflects a broader hesitancy among large enterprises to adopt Chinese AI models due to regulatory uncertainty and security concerns.
Regulatory Investigations Target DeepSeek’s AI Infrastructure
Beyond the API suspension, DeepSeek is also facing heightened regulatory scrutiny. U.S. officials are investigating whether DeepSeek has circumvented export restrictions to acquire high-performance AI chips through third parties in Singapore.
Under current U.S. regulations, Chinese companies cannot legally purchase Nvidia’s H100 GPUs, which are essential for training and operating large AI models. However, reports suggest that DeepSeek may have obtained these chips through indirect supply chains, raising questions about loopholes in U.S. export control policies.
Some experts argue that blocking Chinese companies from Western AI infrastructure could lead to a fragmented global AI market, where companies like DeepSeek develop entirely independent AI ecosystems, making regulatory oversight even more difficult.
Cloud Providers and Developers Adjust to DeepSeek’s Limitations
Despite DeepSeek’s ongoing technical struggles, cloud providers and AI startups are still looking for ways to integrate its models—though with increasing caution.
AI hardware company Groq, which is positioning itself as a competitor to Nvidia, has chosen to use DeepSeek’s reasoning model to enhance Meta’s Llama AI instead of running DeepSeek’s models natively.
(Groq specializes in Language Processing Units (LPUs), which differ from Nvidia’s Graphics Processing Units (GPUs). In August 2024, Groq raised an impressive $640 million in a Series D funding round led by BlackRock Private Equity Partners.)
Meanwhile, Baseten, an AI cloud infrastructure provider, has been testing different deployment methods to optimize DeepSeek’s inference speed. CEO Tuhin Srivastava revealed that before DeepSeek’s January 26 API outage, Baseten’s deployment was actually faster than DeepSeek’s own native API.
At Hyperbolic Labs, an AI cloud infrastructure company, CEO Jasper Zhang told Business Insider that inference users increased by 150% after adding DeepSeek models, while total new user sign-ups surged 400% in January alone. The numbers suggest that despite technical limitations, demand for affordable AI models remains high.
Developers Look for Alternatives as DeepSeek’s Future Remains Uncertain
While some companies are trying to work around DeepSeek’s limitations, others are considering moving away from it entirely. The temporary suspension of API service top-ups has led to questions about DeepSeek’s long-term reliability, with some developers fearing that these issues could become recurring problems.
For applications that don’t require low-latency performance, developers are still willing to work with DeepSeek’s models. For example, Neal Shah of Counterforce Health said that slight delays in processing insurance claim appeals are manageable. However, for AI applications that require real-time responses, such as automated customer service or voice interaction, DeepSeek’s slow inference speeds remain a major roadblock.
Shah is also developing an AI-powered tool that will call insurance companies on behalf of patients, a use case that demands AI to operate at human conversation speed. If DeepSeek’s API speeds and stability don’t improve, Shah says his company will have to switch to another provider.
DeepSeek’s temporary suspension of API service top-ups is more than just a technical problem—it reflects deeper infrastructure challenges in the AI industry. As demand for large language models increases, AI providers must be able to scale their infrastructure without sacrificing performance.
Additionally, as government intervention in AI infrastructure intensifies, companies like DeepSeek must navigate a complex regulatory environment. While some policymakers argue that restricting Chinese AI models is a national security priority, others worry that blocking AI access entirely could lead to technological isolation, making global collaboration more difficult.
For now, DeepSeek remains a low-cost but unstable option for startups. If it can resolve scalability issues, security risks, and regulatory concerns, it could become a serious competitor to OpenAI and Anthropic. However, if API instability and U.S. trade restrictions persist, its presence in Western markets may be severely limited.
