Citi: restrictions on frontier AI are accelerating open-source adoption – and the performance gap is closing fast
Category: AI Models
What it is
Citi analysts say demand for open-weight and open-source AI models has risen sharply
as enterprises face higher costs and tightening access to frontier proprietary models.
The performance gap has narrowed following new releases from Z.ai, DeepSeek,
MiniMax, and Moonshot AI. OpenRouter’s share of open-source tokens rose from 34%
in January to 65% in June 2026. Cohere tripled its 2027 ARR projection. Fireworks’
open-source token volume roughly doubled between April and June. Citi also noted
300+ US data-center moratoriums since 2023 as an infrastructure headwind.
Why it Matters for Enterprises
Open-source AI is no longer the lower-quality fallback - it's becoming the strategic
choice for cost-sensitive, compliance-conscious, or compute-constrained
enterprises. With frontier access shrinking and open models closing the performance
gap, the case for a hybrid model strategy (proprietary for cutting-edge, open for scale)
is stronger than ever.
