Zhipu AI's GLM5.2 costs 75% less than Anthropic, sparking tech rout

Zhipu AI's GLM5.2 model matches Anthropic's performance at one-quarter the cost, triggering a selloff that erased billions from US tech stocks and challenged the premium pricing model underpinning the AI boom.

Zhipu AI's GLM5.2 model delivers performance comparable to Anthropic's frontier offering at one-quarter the cost, triggering a selloff that erased billions from US tech stocks on Friday and reigniting concerns about the sustainability of premium AI pricing.

"This new model is almost equal to Anthropic as a competitor for the corporate market and is just one quarter of the cost in terms of cost per token," Christopher Wood, a strategist at Jefferies, said in a note. He described the past week to clients as "another DeepSeek moment."

Micron Technology fell 7%, while Advanced Micro Devices and Intel each dropped more than 4%. Oracle extended its five-day decline to 19%, hitting a fresh low. Morgan Stanley traders said the model from Z.ai, formerly known as Zhipu AI, has "very impressive coding capabilities" and noted that "the demand mix is clearly shifting towards lower-cost models."

The emergence of cost-effective Chinese AI models threatens to upend the pricing power of premium US providers just as OpenAI and Anthropic approach potential initial public offerings. If enterprise customers migrate to self-hosted, lower-cost alternatives, the "infinite AI infrastructure spending" thesis that has driven Nvidia Corp. and cloud hyperscalers faces a fundamental recalibration.

The cost gap is widening

Deutsche Bank's Jim Reid highlighted the scale of the disparity in a June 18 note, writing that DeepSeek's V4-Pro model "does much the same job at roughly 1.5% of the cost" of Anthropic's leading-edge Claude Fable 5 for roughly 90% of everyday tasks. The compounding effect of multiple Chinese entrants offering near-frontier performance at a fraction of the price is compressing the addressable market for premium AI providers.

The pricing pressure is already reshaping enterprise behavior. Multiple companies have blown through their AI budgets in recent months, and several have begun pulling back on spending tied to high-cost frontier models. Jefferies analysts noted that GLM5.2's privacy protections are comparable to those of leading US models, giving enterprises an additional reason to move AI workloads from cloud service providers back onto their own servers — a shift that would alter the investment case for data center buildouts.

OpenAI's IPO calculus shifts

The selloff comes as OpenAI is reportedly reconsidering its IPO timeline. The New York Times reported Thursday that the company is weighing a delay, citing the lackluster performance of SpaceX shares after its own public listing and recent volatility in tech stocks. A prolonged IPO delay, combined with mounting competitive pressure from cheaper models, could compress the valuation premium that OpenAI and Anthropic have commanded in private markets.

Morgan Stanley traders characterized the shift toward cheaper models as "a recalibration in willingness to pay for AI, not a deterioration in demand for AI." But for investors betting on a straight-line growth trajectory for AI infrastructure spending — a thesis that has supported elevated valuations across Nvidia's supply chain, cloud platforms, and data center REITs — the distinction may offer little comfort.

Nvidia shares face the most direct exposure to any slowdown in enterprise AI spending. The company's data center revenue of $30.8 billion in its most recent quarter reflected demand driven largely by US hyperscalers and frontier AI labs. If a meaningful portion of that demand shifts to cheaper, self-hosted models, the compound annual growth rate baked into Nvidia's current valuation would need to be revised lower.

This article is for informational purposes only and does not constitute investment advice.