How DeepSeek Redefines the Rules of AI and Semiconductors

The world of artificial intelligence (AI) is buzzing with a new name: DeepSeek

This new player in the AI industry has brought a fresh perspective and a captivating model to a market sector that's poised to win over $1 trillion in market share in the next few years. Thanks to DeepSeek-R1's low development costs and groundbreaking capabilities, the semiconductor industry that powers these AI giants is set to flourish.

Beyond its accuracy and cheaper price, why is the tech industry in such an uproar over DeepSeek’s new AI? Its disruptive price tag and stunning intelligence have flipped the current market narrative for AI on its head.  

AI Development in 2025 Comes at a High Price

In 2022, OpenAI’s ChatGPT generative capabilities and quality content generation caught the world’s attention. Chinese tech giants were quick to try and catch up, but the country’s first ChatGPT equivalent, made by Baidu, was considered a disappointment.

Since then, the U.S. has safeguarded its AI technology, with the Biden Administration passing several orders that restricted the export of AI chips and advanced semiconductor manufacturing equipment. This included limiting the sale of graphics processing units (GPUs) to other countries, even those it considers allies.  

Nvidia, the golden goose of AI-capable GPUs, faced the brunt of these export controls, with most of its products facing sales limitations to many countries, including China. Lacking sufficient market competition, despite attempts by AMD to capture a portion of Nvidia’s market share, AI became an expensive sector.

Depending on its specifications, a single Nvidia Blackwell GPU costs between $30,000 and $70,000. This price tag is earned, as Nvidia’s GPUs that have trained most, if not all, current AI models. There is also an additional cost on AI companies for training and “inference”—or run–a large language model (LLM) like ChatGPT.

These models' calculations also require specialized hardware, hence the need for massive GPUs that simultaneously perform simple calculations, inference, and training. These requirements have made Nvidia the go-to for most AI organizations, and the need for a lot of them.

In March 2023, Clement Delangue, the CEO of AI startup Hugging Face, told CNBC that training the company’s Bloom large language model took more than two and a half months and required access to a supercomputer with “something like the equivalent of 500 GPUs.”

“We are actually doing a training right now for the version two of Bloom, and it’s gonna cost no more than $10 million to retrain,” Delangue said. “So that’s the kind of thing that we don’t want to do every week.”

Business Insider reported that ChatGPT could cost OpenAI $700,000 daily due to expensive servers. These numbers have only increased in the almost two years since LLMs exploded in use.  

Then came DeepSeek, which offered incredible quality and extreme efficiency at a fraction of the cost and GPUs.

DeepSeek-R1 Has Sparked an Industry Revolution

At just $5 million, the model’s training expenses pale in comparison to the billions industry giants like OpenAI spent on models such as GPT-4.  

Despite its modest budget, DeepSeek has demonstrated stunning accuracy in natural language processing (NLP) tasks, matching or even surpassing the performance of some of its pricier counterparts, like OpenAI’s o1.  

The AI model is also entirely open source, making it available for anyone to view, modify, and distribute. DeepSeek has even encouraged its users to modify and commercialize its R1 model, which allows for rapid innovation and greater transparency within the industry.

Historically, open source AI models like OpenAI’s GPT-2, have catalyzed advancements by providing researchers and developers with powerful tools to experiment and innovate. However, in recent years, major players have shifted towards more proprietary approaches. DeepSeek’s decision to be open source greatly challenges this trend.

DeepSeek’s two models, V3 and R1, have earned praise from Silicon Valley executives and U.S. tech engineers. DeepSeek-V3 and DeepSeek-R1 are quickly joining the ranks alongside OpenAI and Meta’s advanced models.  

“They are also cheaper to use,” Reuters reports based on DeepSeek’s announcement. “DeepSeek-R1 is 20 to 50 times cheaper to use than OpenAI‘s o1 model, depending on the task.”

Part of this low cost is attributed to the small amount of GPUs used in training. High-Flyer’s AI unit, the funding company behind DeepSeek, said that it owned only 10,000 A100 Nvidia chips in 2022. Nvidia’s A100 chips were only briefly made available in China before being placed under an export ban that same year.

DeepSeek-R1’s capabilities have caused a ripple among analysts, who are scrutinizing companies like OpenAI and Meta more closely and asking if the large number of GPUs and billions of dollars are necessary. This has led to a fiery debate within the AI sector, as critiques over DeepSeek’s transparency and security concerns temper its success.

DeepSeek Geopolitical Challenges are Reflected the Semiconductor Industry

While DeepSeek-R1’s achievements are laudable, it is not without scrutiny. One central point of contention lies in its claims regarding GPU utilization. The model’s developers assert that they achieved their results using a fraction of the GPUs typically required for such training efforts. Skeptics argue that these claims lack sufficient transparency and have called for independent verification of the training process.

Alexander Wang, CEO of Scale AI, stated that China likely has more GPUs than it claims to have, but due to export controls, won’t say.

In comparison, Emad Mostaque, founder of Stability AI, wrote on X that “DeepSeek [is] not faking the cost of the run.”

“It’s pretty much in line with what you’d expect given the data, structure, active parameters, and other elements and other models trained by other people. You can run it independently at the same cost.”

After reviewing the math, Mostaque said that their chip efficiency looks a little lower than expected for 8-bit training and stated that with optimized H100s, DeepSeek could train the V3 and R1 models for under $2.5 million.  

Furthermore, if DeepSeek conceals the accurate number of GPUs utilized for its R1, other AI companies using it, such as Hugging Face, will quickly find out.  

Additionally, some critics state that DeepSeek’s low cost is the most significant threat to U.S. equity markets, as it calls into question the utility of billions of capex being poured into the U.S. AI industry.  

Others contend that while it is possible to train a model as cheaply as DeepSeek, the expenses usually come from the inference of the model. Almost all tech developments have experienced something called the Jevons Paradox, where “technological advancements make a resource more efficient to use (thereby reducing the amount needed for a single application); however, as the cost of using the resource drops, overall demand increases, causing total resource consumption to rise.”

Should this occur, DeepSeek’s AI models could eventually become as expensive as OpenAI’s  o1 model. Other tech experts have also weighed in on the matter, explaining that if training models do get faster and easier for cheaper, the demand for inference will accelerate, thereby raising costs.  

Most experts agree that in the end it doesn’t come down to cost but to who has the smartest AI.

DeepSeek’s Introduction Will Push for Greater Innovation in AI Tech

Despite the controversies, DeepSeek-R1 represents a significant step forward for the AI industry. It demonstrates that world-class models can be developed at a fraction of the usual cost, which could be even lower if components, such as Nvidia’s H100 chips, were better optimized. The growing competition has the possibility to lead to more efficient technologies and a broader distribution of AI capabilities worldwide.

“The United States is going to need a huge amount of computational capacity, a huge amount of infrastructure,” Wang told CNBC, “We need to unleash U.S. energy to enable this AI boom.”

Similarly, Wang notes, each AI company possesses its own strengths, meaning that if more source DeepSeek models debut, this will be an excellent opportunity for others to learn.

If major government-backed pushes like Stargate are any indication, artificial intelligence is expected to be a focus for companies and countries over the next several years. The greater use of AI will also have far-reaching effects on many industries, such as the semiconductor market, which provides the components necessary to power these groundbreaking models. However, this could push popular components, like Nvidia’s, under greater strain.

Organizations must collaborate with a global electronic component distributor that leverages its premier market intelligence tool to inform customers of upcoming shifts or risks. This will help them cope with shortage-like challenges caused by overwhelming demand.

Sourceability and its digital tool, Datalynq, can help give users more visibility into the complex semiconductor supply chain. They can highlight risk areas so companies can find alternative sources with greater multi-source availability or establish case management guidelines for risky components that cannot be ignored.

To prepare for the upcoming shortages and general component unavailability, Sourceability’s experts can help you secure hard-to-find stock, including Nvidia’s GPUs. For those who want to start monitoring their existing supply chains after DeepSeek’s AI debut, Datalynq can help get you started. It’s easy to get started using the power of AI to leverage data-driven insights at your fingertips with Datalynq.