Has DeepSeek Disrupted AI Economics and Efficiency?

The pace of innovation and adoption in AI is relentless and, for technology leaders, one lesson is clear: advances can come from unexpected places. 

Take DeepSeek as an example. The Chinese AI startup, founded in 2023 and funded by Chinese hedge fund High-Flyer, caused hype and uncertainty when it released its DeepSeek-R1 model in January. 

So much so that, on Monday 27 January the Nasdaq Composite dropped by 3.4% at market opening, with Nvidia declining by 17% and losing approximately US$600bn in market capitalisation. 

Such extreme attention was, in some ways, unavoidable given the fever pitch of interest in AI. 

However, behind the Chinese firm’s (unverified) claims of outperforming other models, there are real reasons for the level of disruption R1’s launch caused: cost and efficiency. 

How DeepSeek lowered cost and improved efficiency

DeepSeek claims to have developed its R1 AI model, which uses machine learning and natural language processing to solve complex reasoning tasks, for under US$6m. 

This is a significantly lower cost compared to other leading AI models developed by US companies, which can cost hundreds of millions of dollars to train – OpenAI’s top model, GPT-4 is believed to have cost US$78m to train, while over US$190m was reportedly spent on training Google’s Gemini Ultra.

Training also involved less time and fewer AI accelerators. DeepSeek claims to have trained its model using 2,000 Nvidia H800 graphics processing units compared to the 16,000 H100 GPUs needed to develop Meta’s LLaMA 3. 

Performance-enhancing innovations

According to a report by Bain & Company the AI model’s “performance appears to be based on a series of engineering innovations that significantly reduce inference costs while also improving training cost.

“Its mixture-of-experts (MoE) architecture activates only 37 billion out of 671 billion parameters for processing each token, reducing computational overhead without sacrificing performance,” the report adds.

The company has also employed several other innovative approaches to DeepSeek-R1’s development that aid cost effective performance. 

These include reinforcement learning techniques that enhance performance without requiring extensive supervised fine tuning, sparsity techniques that allow the model to predict which parameters are necessary for specific input thus improving speed and efficiency, and memory compression and load balancing to maximise efficiency.

Impact on users and the market

DeepSeek’s open-source and cost efficient model, combined with its perceived high capabilities, generated instant appeal.  

The company rapidly rose to the top of app store charts after release and maintained its position for several days, effectively displacing more established chatbots like ChatGPT. 

Subsequently, the app faced several bans and restrictions in countries worldwide as a result of concerns around data privacy and security including in South Korea, Italy, Australia and Taiwan. In the US, several agencies and organisations have banned workers from using DeepSeek.

Researchers from security firm Feroot Security claimed to have found links between DeepSeek and a Chinese state-controlled telecoms company previously flagged for security risks.

Despite this attention, the company’s rise means boardrooms and leadership teams are playing closer attention to how AI efficiency and improvements could impact long-term investment plans and strategy. 

According to Bain & Company, the AI model’s impact could unfold in several ways. In a more bullish market scenario, for example, continuing efficiency improvements would lead to greater AI adoption as cost reductions increase demand. This would see spending on cutting-edge AI staying strong. 

More moderately, it predicts a scenario in which spending on AI inference infrastructure decreases by 30% to 50%. In a bearish market, Bain & Co suggests AI training budgets shrink and spending on inference infrastructure declines significantly. 

As this market develops, executives should have three key areas of focus, says Bain & Co. They should avoid overreaction but prepare for cost disruption – DeepSeek highlights the rapid decline in AI costs, creating an environment in which organisations should plan for cheaper technology and wider adoption. 

Market analysis is critical to understand adoption, spend, development and infrastructure demand and to ensure organisations have the ability to quickly integrate AI. 

Lastly, Bain & Company says businesses should view AI through the lens of it being a business model catalyst rather than a productivity tool. “The real winners in AI will be those that use it to redefine their core offerings, not just cut costs,” it says. 

“CEOs should push their organisations beyond automation and into AI-driven innovation—whether in product development, customer personalisation or entirely new services.”

Explore the latest edition of AI Magazine and be part of the conversation at our global conference series, Tech & AI LIVE. 

Discover all our upcoming events and secure your tickets today.

AI Magazine is a BizClik brand