New Research: Startup ARR Security Is at an All-Time Low

Enterprise AI spending is surging, with IDC predicting that companies will allocate $4.25 trillion to technology in 2026, a figure driven largely by artificial intelligence adoption. Despite this massive influx of capital, a new report from venture capital firm Madrona reveals a precarious reality for startups: startup ARR security is at an all-time low. While 74% of surveyed enterprise IT professionals plan to increase their AI budgets over the next year, fewer than half of their initial AI pilots successfully transition into full production.

This retention rate marks a slight improvement over previous years. Last year, MIT reported that 95% of enterprise AI projects failed to deliver on their return on investment expectations. While a success rate below 50% remains concerning, it represents a significant shift from the near-total failure rates seen previously. However, the more critical issue identified by Madrona is not just the failure to launch, but the lack of long-term commitment even after successful deployment.

The “Fast In, Fast Out” Dynamic

The traditional enterprise software model relied on multi-year contracts that created a protective moat around recurring revenue. That stability has evaporated in the AI sector. According to the Madrona report, 77% of enterprises reevaluate their AI vendors every six months or on a rolling basis. This behavior creates a volatile environment where switching costs are minimal and evaluation cycles are relentless.

“This creates a ‘fast in, fast out’ dynamic that is fundamentally different from traditional enterprise SaaS, where multi-year contracts provided a moat of inertia,” Madrona writes in the report. “In enterprise AI, switching costs are lower and the re-evaluation cadence is relentless.”

This instability directly impacts the growth narratives of high-flying AI startups. The initial boom in 2025 was fueled by enterprise trial budgets, leading to dramatic claims of rapid revenue scaling, such as companies growing from zero to $10 million in just three months. These figures rely heavily on the assumption that trials convert into stable, long-term contracts. For the first time, however, revenue remains insecure even after a product moves past the pilot stage.

Pricing Models Under Scrutiny

A significant barrier to long-term contract stability is the struggle many startups face in establishing effective pricing structures. New research from Andreessen Horowitz, which surveyed 50 technical AI buyers, indicates that more than half of respondents prefer fees tied to outcomes or work produced rather than usage metrics like token consumption.

Traditional Software-as-a-Service (SaaS) pricing often scales based on inputs, such as the number of users or data volume. This model works well for established utilities like email or cloud storage. AI, however, requires a value-based approach to demonstrate tangible worth to the buyer. Charging based on outputs, such as the number of reports processed or tickets resolved, aligns the startup’s success with the customer’s business goals.

When fees revolve around specific results, the product becomes “economically valuable to both sides,” according to a16z partners Tugce Erten and Sarah Wang. Without this alignment, enterprises may view AI tools as optional experiments rather than essential infrastructure.

A New Era of Experimentation

The current landscape suggests that AI has ushered in a period of heightened enterprise experimentation. While this openness provides startups with easier access to potential clients, it also means that securing an enterprise contract no longer guarantees future revenue streams. The market is shifting away from the inertia of legacy software toward a fluid environment where loyalty must be earned repeatedly. Whether enterprises will eventually revert to long-term buying habits or continue this cycle of constant reevaluation remains an open question for the industry.