Agents aren’t the only AI-related threat Omdia discussed during this year’s Canalys Forum. Open weight LLMs came up as a danger too, both to potential victims of cybercrime and to Anthropic.
The dangers for cybercrime victims include supply chain vulnerabilities and added attack surface. The danger for Anthropic, which is mere days away from one of history’s most anticipated IPOs, is to its ARR.
“We think that 86% of that revenue, which is going to be over $100 billion, isn’t durable,” said Omdia chief analyst Jay McBain during a keynote. After all, he added, why pay premium rates for premium models when open weight and open source models, not to mention older generations of Claude itself, can handle most of the tasks most people use AI for just as effectively and far less expensively?
Indeed, per data presented by Ishan Dutt, Omdia’s director of PC and tablet research, open models are less than six months behind the frontier in capabilities and way, way cheaper.
Recent figures from Ramp Economics Lab showing token spend barely moving even as token consumption continues to soar make the financial repercussions plain.
“What kind of discounting is going to have to happen to get more people to go to the frontier?” McBain asks.
Enough in Omdia’s view, apparently, to erase 86% of Anthropic’s ARR.