57. The Butterfly Effect: DeepSeek Ripple Through the AI Industry

How a single innovation is changing the economics of artificial intelligence

Dealing with change-driven effects on business models feels a lot like dealing with the “butterfly effect”: a flap of wings could indeed cause a storm elsewhere. The unprecedented level of interconnectedness in today’s world allows any change to ripple through business ecosystems in ways not seen before. Let’s navigate the storm that DeepSeek stirred up in the AI industry.

We saw that DeepSeek changed the AI industry, yet the effects on companies’ financial valuations were not homogenous. Why? The reason behind these changes lies in companies’ business models. A business model offers a peek into the companies’ recipe to generate profit. The sum of all business models within any given industry provides the so-called industry value chain—the industry’s business model, if you will.

Value chains explain how products and services come to market. Think of it as a fancier, business-level version of the famous TV Show “How It’s Made”.

The Value Chain

Value chains are usually divided into three sections: the upstream, the midstream, and the downstream section. Respectively, how things are built from the ground up, how things are combined, and how things are brought to the end customer.

Upstream, the AI industry is populated by companies that deal with supercomputers’ hardware. From companies manufacturing semiconductors, like the Taiwanese company TSMC, to fabless companies, like NVIDIA, AI caused a massive shift in demand for high-performing hardware components. The mismatch between demand and supply of hardware components allowed upstream companies to jack up their prices, recording sky-high margins. NVIDIA posted a 90% margin on some of its most advanced products.

Midstream, companies build the AI infrastructure from the ground up and rent it to firms that train AI models.

The AI infrastructure consisted of huge data centers filled with supercomputers. The massive investments and their long-term horizon of building the AI infrastructure scared away most players—of course, not Big Tech. Not only did they have the resources, but they also had the financial stability to take such an endeavor head-on. However, only Microsoft, Amazon, and Google took part in the AI investment frenzy. This made sense from a business perspective as they already had top-tier cloud infrastructures in place.

All three companies raced each other to secure the most advanced chips, competing to offer the best AI infrastructure for training models. In fact, part of the business model of Big Tech was to rent out the AI infrastructure to companies like OpenAI—companies whose core business was to train AI models.

AI model-makers, next in line after Big Tech, relied on tech giants for financial resources and physical capital. Big Tech, in turn, took this opportunity to invest in these companies’ equity: Microsoft invested in OpenAI, and Amazon in Anthropic, while Google created its model, i.e. Gemini.

So, Big Tech made sure to secure the best-in-class equipment to maximize the success of their invested companies. Their investment frenzy earned them the name “hyperscalers”, quoting Michael Symbolist.

Downstream, the AI industry connects the high-tech side of AI to the operational, practical needs of businesses. Players here understand enough of both businesses to bridge the gap between them. Here operates companies like Salesforce and SAP—providers of Gen AI products and services destined to end customers, either businesses or private customers.

DeepSeek’s effects on AI value chain players

DeepSeek’s blast hit businesses differently, reflecting differences in business models.

Companies manufacturing hardware components, i.e. upstream players, had it rough. These businesses’ profitability was tied up in hardware components’ demand. DeepSeek innovation in AI model training revealed that the industry was less dependent on supercomputers than originally thought. The original “AI pie”, representing the potential market for advanced chips, was split into high- and low-end AI products. As a result, the market for advanced chips shrank, reducing prospective high-margin chip sales and financial performance.

Downstream companies, on the other hand, benefited from DeepSeek’s rise. Less need for supercomputers lowered barriers to entry, leading to greater competition among AI model trainers. As advocates of static framework love to say, greater competition drives prices down. In the economics of downstream players, that leads to lower costs in their profits-and-loss statement and, all else equal, higher profits.

For midstream players, DeepSeek’s impact remains ambiguous due to opposite forces at play. The existing AI infrastructure seemed at risk due to the reduced demand for computing power. However, The Economist argued that the increased number of players in the industry would drive demand back up.

“Where did Big Tech end up in the spectrum of badly hurt to mint condition?” is something that we will discuss further in the next article.