Apollo’s John Zito, speaking at the Morgan Stanley US Financials Conference, argued that token-based measures misprice artificial intelligence economics by tracking throughput rather than business value. His assessment reframes tokens as by-products rather than indicators of productivity, shifting the focus to cost per unit of “useful intelligence”. On that basis, the price of AI is described as falling, even as headline narratives remain centred on token counts.

Lower unit costs, however, do not automatically translate into lower spending: surging usage can still expand overall invoices and pull CFOs into tighter governance. Enterprises are moving away from the “free buffet” phase towards caps, alerts, model routing, and stricter cost controls. Premium pricing is likely to persist for frontier models where ROI is clear, while lower-value tasks migrate to cheaper options. The broader AI trade may continue, but valuation multiples face scrutiny as markets separate genuine intelligence leverage from gains driven by unchecked compute consumption.

Efficiency Over Growth: The New AI Investment Narrative

We see the market is beginning to grasp that the AI story is changing from a simple growth narrative to one of efficiency. The idea that AI spending is a runaway train is being challenged, as corporations are now focused on the return on investment from that spending. This means the indiscriminate buying of anything AI-related is likely coming to an end.

This shift is already showing up in the data. Recent Q1 2026 earnings from major cloud providers indicated that while overall AI consumption is up, growth in the most expensive, frontier-model APIs has slowed from 60% to 45% year-over-year. Meanwhile, the use of smaller, open-source models on their platforms has more than tripled, confirming that customers are actively seeking cheaper alternatives for less critical tasks.

For us, this points to a significant rise in volatility for the big AI infrastructure players. The upcoming July 2026 earnings for a name like NVIDIA will be a critical test, and we believe option markets are underpricing the risk of a guidance revision based on this corporate efficiency push. Buying straddles or strangles heading into that report could be a prudent way to play the potential for a large price move in either direction.

We have seen this pattern before, particularly in the early 2000s when the market shifted from valuing “eyeballs” to valuing actual earnings. The cost per useful unit of intelligence is collapsing, a trend similar to Moore’s Law, but the market is still rewarding companies based on the raw volume of compute they sell. A recent analysis shows the cost for a complex coding query has fallen over 90% in the last 18 months, a powerful deflationary headwind for companies selling raw processing power at a premium.

Trading Strategies and Market Implications

This situation creates clear opportunities for pair trades. We are looking at going long on companies that provide the picks and shovels for AI cost management and efficiency, while taking short positions in pure-play compute providers whose multiples don’t reflect the coming margin compression. For example, a company like ServiceNow announcing a new AI cost-control platform last month is exactly the kind of name that benefits from this new phase of corporate scrutiny.

The broader market seems complacent, with the VIX hovering near a low of 14 despite these fundamental shifts under the surface. This suggests that hedging strategies are relatively cheap right now. We believe buying puts on baskets of high-flying AI software companies that simply embed expensive, third-party models is a cost-effective way to position for the coming “audit” of AI multiples.