Chinese AI company Zhipu’s GLM 5.2 is closing the gap with American frontier models on key agentic benchmarks, and it is free, open-source and adopting faster than DeepSeek. That is the basic shape of the latest scramble in the AI hierarchy: one company’s model is moving closer to the U.S. frontier while being distributed as free, open-source software, forcing enterprise buyers and incumbents to stare down a market where control is slipping away from the usual gatekeepers.
CNBC’s Deirdre Bosa explored what that means for enterprises and vertical AI in a segment published one day ago. The discussion centered on who gets to choose the tools, who gets locked into them, and how the race for inference costs is pressuring the firms that sit at the top of the AI stack.
Who Gets to Set the Terms
Bosa spoke with Box CEO Aaron Levie on model selection, Harvey’s Gabe Pereyra on building atop open-source, and Bernstein’s Stacey Rasgon on OpenAI’s new Jalapeño chip and the inference-cost race hitting Nvidia and Broadcom. The lineup itself shows the power map: enterprise software leaders, startup builders, and market analysts all trying to navigate a field shaped by a handful of dominant AI firms and chip suppliers.
Zhipu’s GLM 5.2 is described as closing the gap with American frontier models on key agentic benchmarks. The article does not say ordinary users are being asked what they want; instead, it frames the issue through enterprise adoption, model selection, and the economics of infrastructure. In other words, the people at the bottom of the stack are left to live with whatever the platform owners and chip vendors decide is viable.
The model is also described as free and open-source, and as adopting faster than DeepSeek. That matters because the usual tollbooths of proprietary AI are not the only route anymore. Open-source tools can weaken the chokehold of closed systems, even as the larger corporate race keeps trying to reassert control through chips, infrastructure, and pricing.
The Cost Race Beneath the Hype
Bernstein’s Stacey Rasgon discussed OpenAI’s new Jalapeño chip and the inference-cost race hitting Nvidia and Broadcom. That is the real machinery under the glossy AI marketing: a fight over who captures the value when models are run at scale, and who gets squeezed when costs shift.
The article does not provide figures on those costs, but it does make clear that the race is affecting Nvidia and Broadcom. Those names sit near the top of the hardware and infrastructure pile, where the decisions made in boardrooms and product labs ripple outward into the systems everyone else is expected to use.
OpenAI’s new Jalapeño chip is mentioned as part of that contest. The chip race is not presented as a public good or a democratic process. It is a competition among powerful firms over control of the infrastructure that underwrites the next layer of automation.
Open Source as a Crack in the Wall
Harvey’s Gabe Pereyra spoke about building atop open-source, which points to a different mode of production than the closed, proprietary model favored by the biggest players. The article does not spell out a mutual-aid network or a grassroots collective, but it does show a practical alternative to locked-down systems: building on open-source rather than waiting for permission from the bosses of the platform economy.
Box CEO Aaron Levie’s comments on model selection fit into the same terrain. Enterprises are being forced to choose among models as the frontier shifts, and those choices are shaped by cost, access, and control rather than any meaningful public say. The result is a market where the language of innovation masks a familiar hierarchy: a few firms set the terms, everyone else adapts.
The CNBC segment, published one day ago, presents Zhipu’s progress as a competitive threat and an enterprise opportunity. But the deeper story is the same one that keeps repeating across the tech economy: the apparatus of AI is built by companies fighting over dominance, while the rest are left to absorb the consequences, chase lower costs, and work around whatever walls the industry puts up next.