An AI-themed bar in Beijing's Zhongguancun tech hub hands out free, unlimited DeepSeek tokens with its drinks, running inference locally on two Nvidia DGX Spark mini-PCs kept on display. The AGI Bar, opened last year on Haidian district's Inno Way startup street by Song De, an independent AI developer in his thirties who runs it in his spare time, sells a signature 9.9 yuan ($1.50) glass of foam, also named AGI, and lets anyone on the WiFi code use its house AI agent at no charge. Song has told Reuters the bar is "completely losing money," with roughly 10 times more drinks given away than sold.
The venue sits a short walk from Tsinghua and Peking universities and the Beijing offices of DeepSeek and Microsoft, has hosted parties for Chinese AI labs including Z.ai, and keeps the gong Z.ai struck for its January Hong Kong listing outside the front door. Its registered Chinese name translates to "knowledge distillation," a pun that works equally well for liquor and LLMs.
According to the report, much of the bar's operations have been automated, with AI agents taking care of inventory, reservations, and memberships. The owner is set to introduce humanoid robots later this year.
Each Spark pairs a 20-core Arm CPU with a Blackwell GPU on Nvidia's GB10 and carries 128GB of unified LPDDR5X, enough, per Nvidia, for models up to 200 billion parameters at FP4. Linking two units over their ConnectX-7 NICs pools 256GB. DeepSeek's V3 and R1 weigh in at 671 billion parameters, and V4 runs to 1.6 trillion, so whatever flows over the bar's WiFi is a distilled or aggressively quantized cut of the model rather than the real thing. The showpiece hardware isn't cheap either: after Nvidia raised the Founders Edition price 18% to $4,699 in response to memory shortages, a matched pair costs about $9,400 before a single free token is poured.
DeepSeek suspended its second fundraising round in late July, days after remarks attributed to founder Liang Wenfeng went viral on Chinese social media. The round had targeted a pre-money valuation of roughly 480 billion yuan, or about $71 billion.
Reports citing Chinese outlet Yicai say the leaked meeting minutes had Liang discussing DeepSeek's continued reliance on Nvidia chips and estimating that China trails leading U.S. labs by 12 to 18 months on around one-twentieth of their compute; Bloomberg, which originally covered the leaked transcript, said it hasn't verified its authenticity.
Those admissions reflect badly on Beijing's push to wean its AI sector off American silicon, an effort that last year saw DeepSeek reportedly urged to train R2 on Huawei's Ascend hardware before repeated failures sent training back to Nvidia GPUs. Two American Blackwell boxes displayed as a Beijing bar's main attraction make for a slightly off-message shrine.
Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.
