The Enterprise Re-Architecture, and OpenAI Claws Back Business Share

This week's news split cleanly into two buckets: enterprises quietly admitting their AI rollouts need to be rebuilt from the ground up, and the ongoing tug-of-war between OpenAI and Anthropic for corporate customers.

OpenAI gains ground with businesses

New data from Ramp, the corporate card and expense-management company, shows OpenAI clawing back share among U.S. business users. Anthropic took the lead among Ramp's paying business customers back in May, hitting 41% share versus OpenAI's 39%.

As of July, Anthropic had grown that lead to nearly 44% versus OpenAI's roughly 40% — but Ramp's economist Ara Kharazian noted OpenAI is currently growing faster among this segment quarter-to-date. The data covers more than 70,000 American businesses and excludes large enterprises that use other expense platforms, so it's a signal rather than the full picture — but it shows Anthropic's enterprise lead isn't locked in.

The "Great AI Re-Architecture"

A wave of enterprise IT audits reported this week that roughly 95% of enterprises have paused AI projects over data governance friction. The core problem: legacy data architectures built for static analytics simply can't handle dynamic, agentic AI workloads, forcing companies into a costly infrastructure reset just to bring compute in line with governance-ready data stores.

Separately, benchmarks on agentic compute workloads found that multi-agent workflows can consume up to 1,000 times more tokens than a simple chat interface — a gap large enough that some teams are now migrating toward smaller, quantized open-weight models just to keep costs sane, rather than routing everything through public cloud APIs.

Where the ROI actually shows up

New adoption reports suggest AI's highest measured return isn't in flashy creative generation — it's in unglamorous, highly governed operational domains like security monitoring and treasury operations, where clear telemetry and deterministic guardrails already existed before AI arrived. It's a useful corrective to a lot of the more speculative "AI will replace creative work" narrative that's dominated headlines this year.

Publishers vs. AI summaries, again

France's communications regulator estimated this week that AI-generated summaries have cut traffic to publisher sites by 33–38%, reviving an old fight over how much of the open web AI companies can consume without compensating the sites they consume it from. It's the same fight that once produced a €500 million fine and payment requirements for Google in France — and it's clearly not over.

Legal AI gets proprietary

Legal AI startup Harvey is reportedly turning open models into proprietary legal intelligence through new partnerships, part of a broader trend of vertical AI companies building defensible moats on top of commodity open-weight models rather than training everything from scratch.

The bigger picture

The "just plug in an API" era of enterprise AI is ending. What's replacing it looks a lot more like traditional infrastructure work — governance, sandboxing, cost control — just with agents instead of pipelines.