Application Software: Citi’s Inference Ahead - Open v. Closed

Research evidence excerpt

Application Software: Citi’s Inference Ahead - Open v. Closed

Viewpoint |

19 Jun 2026 01:21:58 ET │ 18 pages

Application Software

Citi’s Inference Ahead - Open v. Closed

CITI'S TAKE

Heath Terry AC

The AI world’s focus is on the revelation that model access is subject to +1-212-723-4624

government fiat and the need to expand on-premise, open source, and non- heath.terry@citi.com

US model investment accordingly. Along with the emerging routing

technologies, enterprises are deploying to optimize token spend, this Shelby Spencer

revelation will accelerate open-source adoption. The pricing gap for output +1-212-816-0416

tokens between US+EU and Chinese frontier models widened from $5.93 shelby.spencer@citi.com

to $10.05 in four weeks (Figure 12), as US+EU average output pricing rose

28% MoM while it dropped 19% for APAC competitors following DeepSeek's Ashley Kim

permanent 75% price cut on V4-Pro. Meanwhile, the gap in Intelligence +1-212-816-6689

Index scores between the top open and the top closed model is widening ashley.kim@citi.com

(Figure 1). Despite this, our checks with enterprise leaders and vendors

continue to show accelerating enterprise usage relative to more constrained Janna Withrow

supply growth is pushing pricing as much as 2x higher in multiyear +1-212-723-0439

hyperscaler and neocloud contract renewals. janna.withrow@citi.com

Abstraction. AI adoption is ramping, with two-month gains of roughly 1%, 3%, 4%,

and 5% in technology, finance, healthcare, and manufacturing, respectively (Figure

7). Proprietary providers still hold an edge in the abstraction layer, but the bar for

value-add is rising as enterprises grow more capable of building comparable

capabilities with intelligent routing, usage-profiling data, and a growing bench of

open models.