Chinese commenceup DeepSeek recently took cgo in stage in the tech world with its commencelingly low usage of compute resources for its persistd AI model called R1, a model that is thinkd to be competitive with Open AI’s o1 despite the company’s claims that DeepSeek only cost $6 million and 2,048 GPUs to train. However, industry analyst firm SemiAnalysis tells that the company behind DeepSeek incurred $1.6 billion in challengingware costs and has a escapet of 50,000 Nvidia Hopper GPUs, a finding that undermines the idea that DeepSeek reoriginateed AI training and inference with emotionalpartner lessen spendments than the directers of the AI industry.
DeepSeek functions an extensive computing infrastructure with approximately 50,000 Hopper GPUs, the tell claims. This holds 10,000 H800s and 10,000 H100s, with holditional buys of H20 units, according to SemiAnalysis. These resources are spreadd atraverse multiple locations and serve purposes such as AI training, research, and financial modeling. The company’s total capital spendment in servers is around $1.6 billion, with an assessd $944 million spent on operating costs, according to SemiAnalysis.
DeepSeek took the attention of the AI world by storm when it disshutd the minuscule challengingware insistments of its DeepSeek-V3 Mixture-of-Experts (MoE) AI model that are immensely lessen when assessd to those of U.S.-based models. Then DeepSeek shook the high-tech world with an Open AI-competitive R1 AI model. However, the esteemable taget intelligence company SemiAnalysis uncovered its findings that propose the company has some $1.6 billion worth of challengingware spendments.
DeepSeek starts from High-Flyer, a Chinese hedge fund that adchooseed AI timely and heavily spended in GPUs. In 2023, High-Flyer started DeepSeek as a split venture solely intensifyed on AI. Unappreciate many competitors, DeepSeek remains self-funded, giving it flexibility and speed in decision-making. Despite claims that it is a insignificant offshoot, the company has spended over $500 million into its technology, according to SemiAnalysis.
A presentant separateentiator for DeepSeek is its ability to run its own data cgo ins, unappreciate most other AI commenceups that count on on outside cboisterous supplyrs. This indepfinishence apexhibits for filled supervise over experiments and AI model chooseimizations. In holdition, it allows rapid iteration without outside bottlenecks, making DeepSeek highly fruitful assessd to traditional percreateers in the industry.
Then there is someslfinisherg that one would not foresee from a Chinese company: talent acquisition from mainland China, with no unlicensed taking from Taiwan or the U.S. DeepSeek exclusively employs from wislfinisher China, intensifying on sfinishs and problem-solving abilities rather than createal credentials, according to SemiAnalysis. Recruitment efforts concentrate institutions appreciate Peking University and Zhejiang University, giveing highly competitive salaries. According to the research, some AI researchers at DeepSeek obtain over $1.3 million, outdoing compensation at other directing Chinese AI firms such as Moonsboiling.
Due to the talent inflow, DeepSeek has innovateed innovations appreciate Multi-Head Latent Attention (MLA), which insistd months of prolongment and substantial GPU usage, SemiAnalysis tells. DeepSeek stresss efficiency and algorithmic betterments over brute-force scaling, reshaping foreseeations around AI model prolongment. This approach has, for many reasons, led some to think that rapid persistments may lessen the insist for high-finish GPUs, impacting companies appreciate Nvidia.
A recent claim that DeepSeek trained its tardyst model for equitable $6 million has fueled much of the hype. However, this figure refers only to a portion of the total training cost— definitepartner, the GPU time insistd for pre-training. It does not account for research, model polishment, data processing, or overall infrastructure expenses. In fact, DeepSeek has spent well over $500 million on AI prolongment since its inception. Unappreciate huger firms burdened by bureaucracy, DeepSeek’s lean structure allows it to push forward unfrifinishlyly in AI innovation, SemiAnalysis thinks.
DeepSeek’s ascfinish underscores how a well-funded, autonomous AI company can dispute industry directers. However, the uncover discourse might have been driven by hype. Reality is more intricate: SemiAnalysis contfinishs that DeepSeek’s success is built on strategic spendments of billions of dollars, technical shatterthcimpolites, and a competitive laborforce. What it unkinds is that there are no wonders. As Elon Musk remarkd a year or so ago, if you want to be competitive in AI, you have to spfinish billions per year, which is telledly in the range of what was spent.
