A recent tweet by Evan Conrad has sparked discussion regarding the long-term economic sustainability of the artificial intelligence industry, particularly concerning the massive investments in compute infrastructure and the business models of AI application developers. Conrad suggested that three prevailing trends in AI—trillion-dollar compute buildouts, market dominance by a few model companies, and sales to low-margin app customers—cannot simultaneously hold true.
The scale of investment in AI compute is indeed monumental. Projections indicate that companies across the compute power value chain will need to invest an estimated $5.2 trillion into data centers by 2030 to meet global AI demand, according to McKinsey. Other forecasts, such as Dell’Oro Group's, suggest over a trillion dollars in AI-related infrastructure spending within the next five years, highlighting the unprecedented capital commitment.
At the core of this buildout are a handful of dominant foundational model developers. Companies like OpenAI, Google, Anthropic, Microsoft, and Meta are consistently identified as leaders in this space, benefiting from significant venture funding and access to cutting-edge hardware. OpenAI, for instance, is reported to have sought substantial funding rounds to fuel its research and scale its compute infrastructure.
However, the third point of Conrad's tweet, concerning foundational model companies selling to "low-margin app customers," points to a critical challenge in the AI ecosystem. While foundational models are expensive to train and operate, the profitability of applications built on top of these models remains a subject of debate. Some analysts note that high inference costs, where AI tokens are expensive to serve, make it challenging for many AI applications to translate rapid user growth into sustainable profits.
Foundational model companies are exploring various revenue streams, including API access, premium subscriptions, and enterprise solutions, and are also developing their own "new products" or agents. Yet, the question persists whether these models can generate sufficient revenue to justify the immense upfront and ongoing infrastructure investments. The high cost of operating and competing, coupled with the emergence of high-quality free models, could further suppress prices and lengthen the path to profitability for many AI product providers.
Experts suggest that while infrastructure and ancillary services have seen significant financial gains, the application layer faces pressure to differentiate beyond being mere "wrappers" for underlying models. The long-term viability of the AI industry may depend on the emergence of truly transformative applications that can command higher margins and justify the colossal investments being poured into its foundational layers.