In a pointed analysis published October 28, 2026, Frank Flight of Citadel Securities Global Macro Strategy draws a careful distinction between where the AI revolution generates demand and where it generates durable profit. Tokenomics 2.01 finds that falling token prices, stalling enterprise spend, and rising compute costs are converging in ways that challenge the assumption that frontier labs will be the primary beneficiaries of the intelligence era they helped create.
The price declines at the frontier have been swift and structural. Flight notes that "consumption-weighted token prices have declined approximately 50% over six months and average token prices have declined 62% at the frontier, reflecting both the growing share of cheaper open-weight models and price cuts by proprietary providers." The mechanism is competitive displacement: open-weight models now "account for around 65-70% of smart routing token volume, potentially putting pressure on excess rents at the frontier as cheaper models become 'good enough' for a growing range of enterprise tasks." Proprietary labs are no longer competing only with each other.
The enterprise spending picture adds texture. Ramp's monthly data shows total AI spend essentially flat across August and September, the product of lower prices meeting higher volume. Beneath that flatline lies a striking concentration: Flight finds that "80% of spend on frontier labs is driven by 1% of firms, and that the top 1% spent $9,568 per employee per month in Sept vs the median firm that spent $15.7 per employee per month." The AI economy currently rests on a narrow base of heavy enterprise users, a vulnerability rarely surfaced in the broader AI adoption narrative. Flight flags the October release as consequential, noting that "more timely token volume data from October imply a further acceleration of inference demand, but the implications for total spend remain unclear given shifts in token prices and composition of spend."
Token volumes, meanwhile, are surging, up approximately 5x over six months and 50% over the last four weeks. The complication is that this surge is bidding up compute, the primary input cost for inference. H100 GPU rental rates are up approximately 14% over six months and B200 rentals have risen 27.5%, with the forward curve pointing to continued strength. Flight observes that "continued strength in older H100s is particularly striking against the depreciation one would ordinarily expect from technological obsolescence, implying sellers of scarce compute are well placed to extract rents."
For the frontier labs, the margin arithmetic is the crux. Falling output prices and rising input costs compress margins unless inference efficiency improves sufficiently. Flight illustrates the requirement precisely: given the 27.5% rise in B200 GPU rental rates combined with the 50% decline in average token prices, throughput would need to increase 2.55-fold "merely to preserve the gross-margin." He acknowledges that "inference efficiency gains in the last year have been substantial, and compute costs can be locked in via multi-year deals," but whether successive rounds of open-source competition will repeatedly reset that efficiency bar is the forward risk that markets may not be fully pricing.
The structural analogy Flight reaches for is clarifying. "The economics may resemble those of airlines and airports," he writes, with frontier labs as the established airlines providing an essential but fiercely contested service, open-weight models as low-cost carriers "expanding the market through cheaper fares whilst constraining what incumbents can charge," and hyperscaler infrastructure operations as the airports: "with stickier demand and greater scope for higher, more durable margins from supplying scarce compute, and a commercial architecture that is model agnostic in the long run."
The investment implication follows directly. Flight identifies hyperscalers as "potentially the most durable beneficiaries of growing inference demand, with diversified cash flows and strong balance sheets allowing them to fund additional capacity and withstand a reassessment of frontier profitability more comfortably than leveraged neoclouds." Markets, he cautions, "move on gaps between expectations and reality," and the question is whether expectations for frontier profitability "fully accommodate sustained competition that passes efficiency gains through to customers whilst infrastructure suppliers capture the benefits of scarcity in the form of excess rents." The airport earns whether the airline is profitable or not.
5 Key Takeaways for Advisors and Investors
- Token prices are structurally lower. A 50-62% decline at the frontier is not a temporary discount. Open-weight competition has reset pricing expectations, and the direction of travel is unlikely to reverse.
- AI revenue is more concentrated than it appears. With 80% of frontier lab spend driven by 1% of firms, the revenue base is narrow and fragile. Broad enterprise adoption remains incomplete and uneven.
- Infrastructure beats intelligence, currently. Rising GPU rental rates and model-agnostic demand structures favour hyperscaler infrastructure over frontier labs. Compute scarcity rewards the platforms supplying it.
- The efficiency requirement is demanding and ongoing. A 2.55x throughput improvement just to hold gross margins flat is not a one-time adjustment. The pace of inference efficiency gains must be stress-tested against an accelerating open-source competitive environment.
- Leveraged neoclouds carry asymmetric downside. In a frontier profitability reassessment scenario, players with concentrated AI infrastructure exposure and limited financial flexibility carry substantially more risk than hyperscalers with diversified earnings and strong balance sheets.
Footnote:
1 Flight, Frank. "Tokenomics 2.0." Citadel Securities Global Macro Strategy, Citadel Enterprise Americas LLC, 28 Oct. 2026, https://www.citadelsecurities.com/news-and-insights/global-macro-strategy/tokenomics-2-0/.
