For a time, Silicon Valley encouraged employees to use AI as much as possible, even using internal leaderboards to track token consumption and treating the scale of calls as a signal of efficiency improvement. But this approach is starting to cool down, with more and more companies turning their attention back to a more direct question: what results are being achieved with the money spent?
Amazon shuts down internal usage rankings
According to a previous report by the Financial Times, Amazon has shut down an internal dashboard that tracked employee AI usage because some employees started using it for the sake of using it, rather than deploying the tool around actual business problems, in order to boost their rankings.
Amazon Senior Vice President Dave Treadwell told employees that AI should not be used for its own sake, but rather to solve customer problems, business problems, and drive innovation. An Amazon spokesperson later told Business Insider that the unofficial dashboard was originally intended to raise employee awareness of AI's efficiency-enhancing capabilities, not to encourage simply pursuing high usage rates.
Uber says returns are still not obvious.
Uber has also signaled a similar sentiment. In an interview released at the end of May, the company's Chief Operating Officer, Andrew Macdonald, stated that he has not yet seen a direct correlation between increased AI spending and business improvements.
These statements have sparked discussions about the return on investment cycle in AI. Some skeptics believe that if more companies give similar feedback, the market's high expectations for AI may need to be revised. However, some investors believe that this is more like the industry shifting from extensive expansion to intensive management, rather than a sudden disappearance of demand.
Billing model changed to pay-as-you-go
Enterprises' sensitivity to costs is also changing the business models of AI products. This week, Microsoft's GitHub Copilot began switching from a fixed monthly fee to a usage-based billing model. In a blog post in April, GitHub stated that the company had borne a significant portion of the rising technology costs over the past period, making the fixed pricing model unsustainable.
Companies like Anthropic and OpenAI are also gradually de-emphasizing uniform seat fees in the enterprise customer market, shifting towards a model that emphasizes actual usage volume. For developers, this means costs will be more transparent, but potentially higher; for service providers, it means that subsidized expansion is contracting.
Businesses place more emphasis on unit cost
As model capabilities continue to improve, the focus of competition in the AI industry is also shifting. In the past, vendors emphasized whether the model was more powerful; now, they are increasingly emphasizing performance per unit cost. Companies like OpenAI and Anthropic are launching smaller, cheaper, and more efficient models to lower the barrier to enterprise deployment.
Oded Tahori, founder and CEO of AI startup Jeen.ai, stated that early-stage companies' pursuit of token usage largely stemmed from anxiety about missing out and a lack of understanding of the costs and complexities of implementing AI. Now, companies are more concerned with whether their spending aligns with business results.
Overall, tech companies haven't slowed down their AI initiatives, but the metrics are changing. Compared to the previous emphasis on reach and usage popularity, companies are now placing more importance on cost control, actual output, and whether AI can generate verifiable returns at the business level.
