DeepSeek continues to stir more controversy over its ultra-cost-effective AI model, surpassing proprietary models like OpenAI's o1 reasoning model across a range of benchmarks, including coding, science, and math, at a fraction of the development cost.
More specifically, DeepSeek's R1 V3-powered model was reportedly trained using $6 million, and the reinforcement learning technique amid claims AI progression has hit a wall due to a lack of high-quality knowledge for training. However, a new report by SemiAnalysis reveals more intricate details of DeepSeek's AI model (via Tom's Hardware).
As it turns out, the Chinese AI startup dug a little deeper into its pockets than $6 million and 2,048 AI GPUs to train its R1 model. The report details that DeepSeek's R1 model set the company back a whopping $1.6 billion in hardware, including 50,000 NVIDIA Hopper GPUs. The Chinese AI startup reportedly spent up to $944 million on operating costs.
DeepSeek's buzz caused a dramatic shift in the AI industry, raising investor concern, which prompted NVIDIA to shed up to $600 billion in market valuation in a single day. Top executives in the AI industry, including Microsoft CEO Satya Nadella, have weighed in, touting DeepSeek's model as super impressive, urging key stakeholders to take developments from China very, very seriously.
Meta's lead AI scientist, Yann Lecun, claims there's a "major misunderstanding" about the billion dollars invested in AI and how the resources will be used. The executive claims that the money invested in US-based AI firms was dedicated to inference, not to train AI models.
"DeepSeek has some real innovations." added Nadella. "When token prices fall, inference computing prices fall, that means people can consume more, and there'll be more apps written."
DeepSeek hires from within
Despite the vast resources required to facilitate AI advances, the Chinese startup remains a self-sponsored entity, locking out outsider interference and hostile takeovers. The company leans more on efficiency and algorithmic enhancements over the rapid scaling of models.
Rival AI firm OpenAI is under pressure to evolve into a for-profit entity or risk refunding the money raised by investors during October's funding round, which pushed its market cap well beyond $157 billion and saved it from bankruptcy.
However, the move has received backlash from key players, including former OpenAI co-founder Elon Musk, who filed two lawsuits against the company, citing a stark betrayal of its founding mission and alleged involvement in racketeering activities. Musk claimed he was lured to invest in the AI firm with a fake humanitarian mission.
To this end, DeepSeek has reportedly invested up to $500 million in its AI advances.
Elsewhere, the emerging AI firm reportedly recruits its employees exclusively from China, with a keen focus on skills and problem-resolution capabilities. The hiring managers reportedly target potential employees from reputable institutions in China, including Peking University and Zhejiang University. Plus, DeepSeek doesn't shy away from handsomely compensating its employees, with some of its AI researchers earning up to $1.3 million.
