Morgan Stanley's analysis shows a typical enterprise AI use case can generate ~$55 in value for just a few dollars in token costs. This massive return on investment suggests that widespread concerns about enterprises aggressively curtailing AI token spending are likely overstated, as the value proposition remains overwhelmingly positive.
For critical enterprise uses like coding, the cost to remediate a single error from a cheaper AI model far outweighs any savings. This high cost of failure ensures businesses will continue paying a premium for more reliable, high-end proprietary models for crucial tasks, while using open-source options for lower-stakes work.
Contrary to fears that cheaper AI models will hurt the market, the opposite is likely true. As the cost of AI tokens and compute drops, it unlocks more use cases and spurs greater demand. This phenomenon, known as Jevon's paradox, suggests total capital expenditure on AI infrastructure will continue to rise despite falling unit costs.
Despite growing local, bipartisan opposition to data center construction, the US government views AI leadership over China as a critical national security issue. This federal priority makes a nationwide ban unlikely. Instead, expect a "conditional buildout" where developers must offer community benefits like grid modernization to proceed.
To circumvent growing local opposition and permit denials related to water and power usage, data center developers are pivoting to off-grid strategies. By building self-sufficient facilities powered by natural gas turbines and fuel cells, they aim to eliminate community impacts and sidestep political and regulatory hurdles slowing traditional projects.