Anthropic Gains, AI QA Fails

Anthropic is closing in on OpenAI in business spending, even as a separate company’s all-AI quality check reportedly blew a $6 million hole in sales. (ramp.com) (economictimes.indiatimes.com) Ramp said on April 11 that 50.4% of U.S. businesses on its platform paid for at least one artificial intelligence product in March, up from 35% a year earlier. Anthropic’s penetration rose to 30.6% of businesses in March from 24.4% in February, while OpenAI was flat at 35%. (ramp.com) (pymnts.com) Ramp said its index is based on anonymized card and bill-pay data from more than 50,000 businesses and about $100 billion in annualized purchasing volume. The Financial Times report cited by PYMNTS said the March data showed Anthropic gaining more than 6 percentage points in a single month. (ramp.com) (pymnts.com) This is not the first sign that business buyers are spreading work across more than one model maker. Menlo Ventures said on July 31, 2025 that Anthropic held 32% of enterprise large language model usage, ahead of OpenAI at 25% and Google at 20%. (menlovc.com) (techcrunch.com) The quality assurance story points in the other direction: buying more artificial intelligence tools is not the same as safely handing them production control. Economic Times reported on April 12 that a company replaced its quality assurance team with an artificial intelligence testing system, which then issued a faulty discount that priced products at zero and led to a reported $6 million loss. (economictimes.indiatimes.com) (newsbytesapp.com) Economic Times said the account came from a software professional describing a 12-person quality assurance team that had been cut to save about $1.2 million a year. After the failure, the chief executive reportedly asked a laid-off senior quality assurance lead to help fix the system without pay. (economictimes.indiatimes.com) (letsdatascience.com) OpenAI still has the larger business footprint in Ramp’s March data, and it remains far larger in consumer use. PYMNTS, citing the Financial Times, said OpenAI has about 900 million weekly active users, with a little more than 5% paying. (pymnts.com) (cnbc.com) OpenAI has also been telling investors that computing capacity will decide the next stretch of the race. CNBC and Bloomberg reported last week that OpenAI argued Anthropic is constrained by a smaller infrastructure build-out, even as Anthropic gains ground with enterprise customers. (cnbc.com) (bloomberg.com) Taken together, the April numbers show two separate decisions inside companies: which model vendor gets the budget, and whether a human still signs off before software touches revenue. March favored Anthropic on the first question; the reported $6 million mistake is a costly answer to the second. (ramp.com) (economictimes.indiatimes.com)