What happened to Deepseek?
DeepSeek shocked global markets in January 2025 with a powerful, low-cost AI model that briefly became the most downloaded app in the United States, then faded from headlines due to chip shortages, infrastructure outages, and the simple fact that markets had already priced in the...
A thread on Reddit's LocalLLaMA forum, submitted by user Mr_Moonsilver, captured what a lot of AI watchers have been quietly wondering for months: What actually happened to DeepSeek? Meta clawed back attention with a string of new releases. OpenAI kept shipping. But DeepSeek, the Chinese AI lab that briefly broke the internet and rattled Wall Street, seemed to vanish. The answer is more complicated than "they failed," and more honest than "they're fine."
Why This Matters
DeepSeek's January 2025 launch was not a fluke or a PR stunt. It was a genuine forcing function that made Western investors and policymakers reassess their assumptions about Chinese AI capabilities and the cost of training frontier models. The company proved that a non-American lab could match OpenAI's output at a fraction of the reported training cost, and markets responded by wiping billions from AI-adjacent stocks in a single week. That repricing has permanent consequences, even if DeepSeek itself never makes another headline.
Daily briefing from 50+ sources. Free, 5-minute read.
The Full Story
DeepSeek released its R1 model in the first week of January 2025, which happened to coincide with Donald Trump's return to the White House. The timing was almost cinematic. Within days, the app became the most downloaded free application in the United States on Apple's App Store, a result no one in Silicon Valley had predicted for a Chinese AI product. The model demonstrated reasoning capabilities that matched or exceeded OpenAI's ChatGPT at the time, and DeepSeek's reported training costs suggested American labs had been dramatically overspending, or at minimum, that their cost advantages were not as wide as assumed.
The market reaction was swift and brutal. As Gartner analyst Haritha Khandabattu later explained to CNBC in January 2026, the initial shock repriced "global beliefs about frontier-model cost curves and China's competitiveness" in a single event. AI stocks entered a freefall. Investors who had bet on American dominance in the space suddenly had to reckon with the fact that the race had at least one more serious entrant than they had accounted for.
But here is where the story gets more nuanced. The shock was so effective precisely because it was a surprise. Once markets absorbed that surprise and adjusted their models, no subsequent DeepSeek release could recreate the same reaction. By the time CNBC revisited the story in January 2026, one year after the initial wave, the outlet reported that DeepSeek's newer releases had generated no comparable investor frenzy. The element of surprise, which is not a technology you can simply rebuild, had been spent.
U.S. export controls then began to bite. Reporting from The Information in June 2025 indicated that DeepSeek's development had slowed specifically because American restrictions on advanced semiconductor exports were cutting off the lab's access to Nvidia's latest processors. Training a frontier AI model without top-tier chips is like trying to win a Formula 1 race with a smaller fuel tank. You can still race, but the gap widens over time. This hardware bottleneck is arguably the single most significant reason DeepSeek's technical momentum slowed.
The company also faced operational problems that made it look less like a juggernaut and more like a startup struggling to scale. According to Reuters, DeepSeek's AI chatbot experienced its longest outage since its viral debut, in March 2026. Rapid user adoption had strained the company's infrastructure in ways that were visible to everyday users, and that kind of reliability problem does real damage to public perception in a category where ChatGPT and Gemini simply stay . Despite all of this, DeepSeek is not gone. Reuters reported in April 2026 that the company's upcoming V4 model will run on chips designed by Huawei Technologies, a direct adaptation to the export control restrictions. The company is not sitting still. It is engineering around its constraints, which is itself a significant technical and strategic choice.
Key Details
- DeepSeek R1 launched in the first week of January 2025, during the same week as Trump's second inauguration.
- The app became the number one free download in the United States on the Apple App Store within days of release.
- AI stocks dropped sharply in January 2025, with Gartner analyst Haritha Khandabattu citing a broad repricing of frontier model cost assumptions.
- The Information reported in June 2025 that U.S. chip export controls had stalled DeepSeek's development progress.
- Reuters reported on March 30, 2026 that DeepSeek suffered its longest chatbot outage since its viral rise in early 2025.
- DeepSeek V4 is confirmed by Reuters as of April 3, 2026, and will run on Huawei-designed chips rather than Nvidia hardware.
- Meta recruited three researchers from OpenAI in June 2025 to work specifically on reasoning models, a direct competitive response to the capabilities DeepSeek demonstrated.
What's Next
Watch for DeepSeek V4's release and benchmark results closely, because its performance on Huawei chips will be the first real test of whether Chinese domestic semiconductor alternatives can sustain frontier AI development without access to Nvidia. If V4 performs competitively, it changes the calculus on whether export controls are actually slowing Chinese AI research in a meaningful way. If it underperforms, it validates the chip restriction strategy and signals a widening capability gap.
How This Compares
Meta is the most instructive comparison here. Where DeepSeek faded, Meta came back with a string of model releases in 2025 and 2026 that recaptured developer attention, particularly in the open-weight space. Meta's advantage is infrastructure: the company can absorb the cost of compute at a scale DeepSeek cannot match right now. Mr_Moonsilver's original post acknowledged Meta's comeback and noted the open-source debate around it, but the real point is that Meta never had its access to cutting-edge chips restricted. That asymmetry matters enormously.
Compare DeepSeek's situation to what happened with Huawei itself in the smartphone space. Huawei lost access to advanced chips in 2020, fell off global market share charts, then spent years developing domestic alternatives and came back with the Mate 60 Pro in 2023, surprising analysts who had assumed the restrictions were a knockout blow. DeepSeek may be on a similar timeline. The V4 announcement using Huawei chips suggests the lab is following a similar playbook: adapt, build around the constraint, and return when ready rather than disappearing permanently.
OpenAI and Anthropic, by contrast, have had an uninterrupted run of hardware access and talent. In June 2025, Meta specifically recruited three OpenAI researchers to work on reasoning models, a domain where DeepSeek's R1 had demonstrated surprising strength. That recruitment shows how much DeepSeek's January 2025 release shaped the competitive agenda in ways that are still playing out. The lab may be quiet, but the ripple effects from its debut are still reshaping how every major AI company thinks about cost, efficiency, and Chinese competition. For the latest on how these competitive dynamics are shifting, it is worth tracking all three storylines together.
FAQ
Q: Is DeepSeek still available to use right now? A: Yes, DeepSeek's chatbot is still accessible and the company continues to operate. It experienced a notable outage in March 2026, but service has continued. The company is actively developing its V4 model, which is expected to run on Huawei-designed chips rather than Nvidia hardware.
Q: Why did DeepSeek stop making news after January 2025? A: The January 2025 launch was so surprising that it repriced global assumptions about Chinese AI capabilities in a single event. Once markets adjusted to the new reality, subsequent releases could not generate the same shock. U.S. chip export controls also slowed the company's development pace, making it harder to ship releases that could match the R1's impact.
Q: What is DeepSeek V4 and when does it come out? A: DeepSeek V4 is the company's next major model release, confirmed by Reuters in April 2026. It will run on chips designed by Huawei Technologies, bypassing the Nvidia hardware that U.S. export controls have restricted. No firm release date has been officially announced as of this writing.
DeepSeek's story is not over. It is just in a slower, harder chapter, one where the company has to prove it can build at the frontier without the world's best chips and without the element of surprise that made January 2025 such a defining moment. How V4 performs will tell us a great deal about the real-world effectiveness of U.S. semiconductor policy and the resilience of Chinese AI research. Subscribe to the AI Agents Daily weekly newsletter for daily updates on AI agents, tools, and automation.
Get stories like this daily
Free briefing. Curated from 50+ sources. 5-minute read every morning.
