Part of: this research — view all 4 rows
The Measurement Baseline The denominator for this ratio is 11%, the share of enterprises that changed their primary model vendor in the 12 months prior to Menlo Ventures' mid-2025 survey 2025 Mid-Year LLM Market Update: Foundation Model ... 2025 Mid-Year LLM Market Update: Foundation Model .... It is critical to distinguish this annual rate from cumulative "ever switched" metrics, such as the 55% figure reported in a 2026 Dataiku/Harris poll When AI model choices become a switching problem Best LLMs for Businesses in 2026: Ranked and Compared. Because Menlo's late-2026 edition omitted a follow-up switching measure 2026: The State of Enterprise AI | Menlo Ventures 2026: The State of Enterprise AI | Menlo Ventures, resolving this question will likely depend on a different survey with alternate sampling, wording, and definitions of a "primary" vendor. This introduces substantial measurement noise, driving the wide uncertainty in the outer percentiles.
Continual Learning and Lock-In The structural lock-in thesis depends heavily on models learning directly from deployment sessions. However, as of mid-November 2026, no frontier model updates its weights from live production traffic; all active frontier systems rely strictly on retrieval or context injection. Because weight-level continual learning is not yet operating at the frontier, this specific lock-in mechanism is unlikely to meaningfully restrict enterprise movement during calendar 2027. Instead, the switching rate will be determined by conventional market forces and architectural friction.
Upward Pressures on Switching Several forces push the ratio above 1.0. Multi-homing is expanding rapidly: 37% of enterprises now run five or more models How 100 Enterprise CIOs Are Building and Buying Gen AI ... How 100 Enterprise CIOs Are Building and Buying Gen AI ..., and 16% pay both major providers, up from 8% 2025: The State of Generative AI in the Enterprise. When secondary vendors are already integrated, elevating an alternative to the "primary" position is largely a matter of volume-shifting rather than a cold migration. Price competition remains aggressive, and the capability frontier frequently changes hands. Additionally, early 2027 will see forced migrations as OpenAI shuts down self-serve fine-tuning job creation and retires the GPT-4o and GPT-4.1 APIs Deprecations | OpenAI API, pushing some affected customers to reconsider their vendor alignment.
Downward Pressures on Switching Conversely, agentic workflows and complex integrations are increasing the engineering cost of switching. Prompts, guardrails, and reliability testing are often model-specific; a Stacklok survey of advanced agent adopters found that 81% had delayed or scaled back provider migrations due to these switching costs State of AI Architecture 2026. Furthermore, enterprise contracts remain annual and cloud-scoped, and closed-model fine-tunes cannot be exported. Market consolidation also limits mobility: Anthropic and OpenAI now command 86% of enterprise API spend 2026: The State of Enterprise AI | Menlo Ventures 2026: The State of Enterprise AI | Menlo Ventures, leaving fewer credible alternative providers for a primary migration. Potential costs of dedicated deployment instances could also make maintaining readily usable alternatives less economical, tempering the extent to which multi-homing translates into primary-vendor switching.
Synthesis The median estimate of roughly 1.16 anticipates a modest increase in the primary-vendor switching rate. While agentic integrations create genuine architectural friction, the practical ease of reweighting traffic in multi-homed environments, combined with aggressive price and capability churn, outpaces these constraints. Enterprise API spending trends and the potential costs of dedicated deployment instances give somewhat less weight to extreme outcomes: concentrated spending constrains mobility, while multi-homing preserves a practical route to switching without requiring a cold migration. However, the distribution remains exceptionally wide and right-skewed—spanning from roughly 0.58 at the 10th percentile to 2.35 at the 90th, with the 25th and 75th percentiles at 0.85 and 1.63—accommodating the high likelihood of survey measurement variance and the competing extremes of duopoly consolidation versus fierce price wars.
Evaluating this estimate alongside enterprise API spending trends and the potential costs of dedicated deployment instances led to a slight narrowing of the uncertainty interval, while preserving the median near 1.16 to balance rising multi-homing against architectural switching friction.
Ask a followup
Sign in to run · $20 free credit, no card · every claim cited