High costs force a shift—Microsoft considers replacing expensive American AI models with DeepSeek.

```

Microsoft is considering introducing a fine-tuned version of the Chinese open-source model DeepSeek V4 into its enterprise AI tool Copilot Cowork as a low-cost alternative to OpenAI and Anthropic models. According to Axios, Microsoft is expected to decide and announce its final selection in the coming weeks.

This news closely echoes the core issue recently raised by Rich Privorotsky, head of Goldman Sachs' Delta-One trading desk—he refers to the current pricing dilemma in the AI industry as the “trillion-dollar question”: does a lower cost of intelligence create demand, or destroy pricing power?

On this question, Microsoft seems to have already “voted with its feet.” Data likewise corroborates this: the Silicon Data Token Index, which tracks AI Token prices, has fallen on 12 out of the last 13 trading days, heading straight for recent lows.

Why does Microsoft want to change models? Cost pressures are becoming unbearable

Microsoft previously offered unlimited use of Copilot Cowork to enterprise users, but this path is no longer viable.

Charles Lamanna, Executive Vice President in charge of Copilot, was direct: "Some users complete hundreds of tasks each week, which is very efficient—but the cost can skyrocket."

As a result, Microsoft announced Copilot Cowork will switch to a usage model charged by computing volume. At the same time, Microsoft is exploring the introduction of a fine-tuned DeepSeek V4 or other open-source models to significantly reduce model invocation costs.

The logic behind this is simple: Chinese models are cheap, American models are expensive. According to Token pricing data, there is a significant gap in input/output pricing between Chinese and American models.

“Matching the top models at half price”—Goldman raises the “trillion-dollar question”

Goldman cites the latest OpenRouter experiment results: a multimodel set comprised of Gemini 3 Flash, Kimi K2.6, and DeepSeek V4 Pro comprehensively outperformed GPT-5.5 and Opus 4.8 running alone in benchmark tests, narrowing the performance gap with Fable 5 to less than 1%, all at about half the cost.

Goldman’s Privorotsky characterizes this result as "a direction that the market has consistently underestimated."

This trend has two-sided implications for the market.

Bullish logic: lower costs and reduced entry barriers should ultimately drive synchronous expansion in AI usage and computing power demand.

Bearish logic: this directly accelerates token deflation, shaking the sustainability of existing model economics.

Privorotsky distills the central contradiction into the question: "Does a lower cost of intelligence create more demand, or destroy more pricing power?" He calls this the "trillion-dollar question."

Is cheap AI ultimately a good thing or a bad thing?

The real shock: does the logic behind trillion-dollar capital expenditures still hold?

Microsoft’s choice impacts the market far beyond a single company’s supplier switch.

Over the past two years, tech giants have spent or pledged massive capital expenditures. One of the core assumptions behind this is: enterprise customers will continue to purchase high-priced top American AI models, and revenues will sustain these investments.

But if even Microsoft—the largest investor and partner of OpenAI—starts to consider OpenAI too expensive, and turns to Chinese open-source models, the foundation of this assumption begins to shift.

Axios also points out, since the debate over "token prices at historic highs" began, Silicon Valley’s data Token Index has fallen on 12 out of the last 13 trading days, signaling a clear market vote.

Microsoft's multi-model strategy also reflects a larger industry shift—no longer betting on a single supplier, but flexibly adjusting models based on task complexity and cost. For OpenAI and Anthropic, this means bargaining power is being diluted.

Risk warning and disclaimerThe market involves risks. Investment requires caution. This article does not constitute personal investment advice, nor does it account for the specific investment objectives, financial circumstances, or needs of individual users. Users should consider whether any opinions, views, or conclusions herein are appropriate for their particular situation. Investing based on this article is at your own responsibility. ```