February 13 marked an unprecedented surge in the demand for high-performance computing units, particularly L20 GPUs, as a result of two significant Chinese tech companies aligning with the needs of the open-source model DeepSeekReports indicate that these firms had to urgently procure computational resources from distributors, highlighting a drastic shift in the market dynamicsThis moment stands in stark contrast to the previous year, when decision-making around tech orders was slow and marked by analysis and comparisonToday, supply levels of even restricted inventory items have rapidly depleted, signaling a potentially transformative moment in the technology sector.

The advent of DeepSeek has challenged the long-cherished belief that high-performance GPUs serve as the backbone of AI technologyThis shift has led to an initial bearish sentiment surrounding Nvidia, the dominant player in the GPU marketYet, as industry observers have noted, a remarkable reversal has taken placeHigh-end GPUs, particularly from Nvidia's H series and RTX 40 series, are once again witnessing a spike in demand, indicating the swirling complexities of the market ecosystem.

The previously mentioned Nvidia distributor has ramped up efforts to meet this burgeoning demand

Advertisements

The crunch for Nvidia chips has become pervasive across the Chinese market, with previously banned H800 series units now nearly impossible to secureThe rising tide of partnerships with DeepSeek has enveloped hundreds of domestic and international businesses, spanning from cloud services and internet companies to telecommunications, financial institutions, healthcare, and automotive industries.

These rapidly increasing user statistics have translated into significant financial benefits for several companies involved; cloud service providers are now able to market computing resources, chip manufacturers leverage this opportunity to showcase their products, automotive firms are enhancing software at lower costs, and publicly traded companies are experiencing upward valuation pressuresThe transformative impact of DeepSeek on industry revenues is nothing short of remarkable.

Indeed, one IT firm recounted that their utilization of the DeepSeek platform resulted in an influx of tens of thousands of new users on its inaugural day, leading to server overloads and a desperate need for additional Nvidia GPUsMany enterprises realize that while they can activate DeepSeek, their existing infrastructure simply cannot keep pace with the astonishingly rapid growth of user demand.

Parallel Technology, a company specializing in supercomputing cloud services for financial, research, and aerospace industries, highlighted that a model's lifecycle typically consists of training, deployment, and inferenceCurrently, there is a discernible surge in demand for computational resources during the inference phase of models associated with DeepSeek.

As reported by industry insiders, legitimate Nvidia products available for sale in China include H20, L20, T4, and a range of RTX gaming graphics cards, which span from the 40 series (excluding RTX 4090) to preceding series, along with the newly released limited edition RTX 5090D.

While numerous domestic GPU brands have emerged in recent years, they often fall short in terms of performance and ecosystem stability compared to Nvidia

Advertisements

Consequently, Nvidia cards remain the preferred option for many enterprises seeking reliable computing power.

Following the launch of DeepSeek's R1 model, Nvidia's stock experienced a staggering 17% drop—its largest decline in a year—yet as of February 3, a rebound of 13% has been observed, indicating the market’s volatility and the ensemble factors influencing investor sentiment.

An individual from a domestic model company shared insights, noting DeepSeek's emergence has sparked critical discussions surrounding the potential 'computational bubble' and whether AI training needs to rely on massive and costly GPU clustersHowever, the burgeoning number of individuals and businesses both providing and utilizing models reinstates the substantial need for Nvidia products.

According to Chen Jian, the chairman of Parallel Technology, the disruptive impact of DeepSeek signifies a pronounced reduction in the computational power consumption associated with AIGC and large model training and inferenceIn his perspective, while the demand for inference-driven workloads is surging, the immediate need for model training resources has not experienced a similar explosion.

DeepSeek showcases powerful inference capabilities with impressive outputs; however, the cost of delivering this service is not negligibleAs specified by industry experts, the hardware requirements for running an optimal DeepSeek experience are quite demanding, necessitating at least two H800 GPUs, whereas many existing domestic foundational models could function with just one RTX 4090 card.

Typically, model performance is closely linked to the number of parameters and their accuracy; the parameter count dictates the breadth of knowledge an AI can acquire, while accuracy influences its precision and reasoning capabilities

Advertisements

Advertisements

Advertisements