Microsoft (MSFT) Office chief Lamanna pushed to use Anthropic models for Copilot despite concerns over higher costs, and is weighing Anthropic and OpenAI models to improve capabilities and margins, The Information reports

Diversifying the model supply behind a flagship AI product is a pattern that has recurred wherever a large platform has depended on a single external partner whose incentives do not fully align with its own. The distinction worth drawing is between capability and margin: shopping across model providers addresses both, since competitive tension among suppliers compresses inference costs while best-of-breed routing can lift quality, but it also signals that the incumbent relationship no longer covers the full product surface. For the supplier being diversified away from, episodes of this kind have tended to reprice the assumed durability of exclusivity rather than near-term revenue, with the read-through concentrated in the private-market valuation and compute-contract narrative rather than the customer's equity. The higher-cost concern flagged here is the tell: if the more expensive alternative is still being weighed, the constraint is quality, not price, which historically precedes multi-model architectures rather than wholesale switching. Worth watching is whether this surfaces in disclosure around AI gross margins, any commentary on model routing in product updates, and whether the secondary supplier's other enterprise relationships follow the same template. Attribution to a single outlet also matters; such reports have a mixed record of being confirmed on the record.