The Risky Economics Behind Today’s AI Giants

I‘ve recently been wondering about the business model of highly-hyped AI companies, such as OpenAI. In my opinion, it‘s quite risky:

These companies have made — and continue to make — massive infrastructure investments, leading to significant upfront CAPEX and resulting in a lock-in effect to existing cloud setups. Working in the automotive industry, I know that investing in fixed capacities creates substantial pressure for utilization. Without sufficient use, these assets quickly become financial liabilities. This lock-in further restricts flexibility, making it difficult to abandon the expensive infrastructure you‘ve built. Automotive OEMs face a similar dilemma, as they‘re heavily reliant on specific platforms and factories they’ve constructed.

Revenue for companies like OpenAI primarily comes from API sales rather than direct end-user products. This dependency means they have limited control over the actual products generating customer revenue. If the products don’t fly, no API calls and therefore no utilization. Moreover, the recent emergence of competitive products like Google Gemini and DeepSeek R1 has triggered a price war, driving down prices and reducing margins. Customers, especially startups, are typically price-sensitive, and as improvements in Large Language Model (LLM) performance slows down, cheaper alternatives become increasingly attractive.

Another challenge is the narrow market scope focused primarily on text and image generation via remote services. This model does not effectively address less sexy but potentially broader and lucrative markets—particularly those requiring immediate processing, heavy individualization or customization, or sensitive data handling (e.g., financial data, industrial or agricultural sensor information, individual health data streams).

Furthermore, the actual ROI for solutions built upon these APIs remains uncertain and largely unproven. While ChatGPT and similar products have achieved widespread consumer adoption, robust enterprise-scale solutions or major consumer products capable of justifying the significant CAPEX and ongoing OPEX have yet to materialize at scale.

The conclusion? Obviously, the technological achievements are undeniably impressive, but the economic viability remains uncertain. It remains to be seen whether these companies can develop sustainable, profitable business models - or if challengers emerge that provide one. In that case at least, we might see the market flooded with cheap second-hand graphics cards :)