Bending Spoons self-hosts open-weight models for 99% of AI requests
Summary
Bending Spoons, a Milan-based tech company led by CEO Luca Ferrari, revealed that approximately 99% of its AI requests and tokens are handled by self-hosted open-weight models, with only about 1% allocated to frontier models for the most complex tasks or to verify outputs from the open models. This strategic approach aligns with the company's commitment to Sovereign AI, as it enables them to maintain full control and vendor neutrality while keeping token costs low. Ferrari emphasized that this shift allows the company to minimize cloud expenses by relying predominantly on their self-hosted models.
Analysis
Luca Ferrari: Luca Ferrari serves as CEO and co-founder of Bending Spoons. He has publicly detailed the company's shift toward sovereign AI infrastructure, highlighting the use of internally hosted open models as the default for routine requests. Ferrari described frontier models as supplementary tools applied selectively for advanced reasoning or oversight, similar to senior engineers reviewing junior work. Bending Spoons: Bending Spoons is a Milan-based Italian tech company focused on building consumer software products. The company has adopted a sovereign AI strategy that prioritizes self-hosting open-weight models for the vast majority of its AI workloads. This approach enables full vendor neutrality while reserving frontier closed models for only the most complex tasks or quality checks. Sovereign AI: Bending Spoons prioritizes self-hosted open-weight models to maintain full control and vendor neutrality in its AI operations. Hybrid Model Strategy: The company reserves frontier closed-weight models for the most complex tasks or automated verification of outputs from open models.
Categories
techaiai_agents