Artificial Intelligence: Citi’s Inference Ahead - Pricing, Financing, Routing, and Securing

研报英文原文证据摘录

Artificial Intelligence: Citi’s Inference Ahead - Pricing, Financing, Routing, and Securing

Viewpoint |

31 Jul 2026 17:19:51 ET │ 22 pages

Artificial Intelligence

Citi’s Inference Ahead - Pricing, Financing, Routing, and Securing

CITI'S TAKE

Heath Terry AC

AI lab strategies are increasingly shifting from maximizing benchmark +1-212-723-4624

scores toward developing a spectrum of models designed to address the full heath.terry@citi.com

stack of enterprise needs across price, function, and speed. This is

amplifying enterprises’ need for an orchestration layer for model routing, Shelby Spencer

governance, and security, a need we expect to be met by a combination of +1-212-816-0416

homebrewed solutions and professional offerings. With major next- shelby.spencer@citi.com

generation frontier model releases coming soon (Figure 5), we expect the

gap between frontier and open models will re-widen considerably, reversing Ashley Kim

the negative sentiment that has developed in recent weeks. Security is +1-212-816-6689

emerging as a differentiator alongside intelligence, with the Aug. 1 deadline ashley.kim@citi.com

for federal agencies to adopt a prerelease framework for frontier models

likely to impact cadence (Figure 4). Janna Withrow

+1-212-723-0439

Pricing. Price competition accelerated this week as frontier labs lowered top-model janna.withrow@citi.com

costs, with blended frontier pricing falling 16.2% WoW to $1.34 even as the closed-

open gap widened to 4 points (Figure 1). Upcoming releases could move more

workloads to cheaper models while concentrating willingness to pay around high-

value enterprise use cases like cybersecurity, finance, and legal. Public benchmarks

are becoming increasingly difficult to rely on, as a UK AISI / US CAISI evaluation