Microsoft is evaluating the integration of a fine-tuned version of the Chinese open-source model DeepSeek V4 into its enterprise AI tool, Copilot Cowork, as a lower-cost alternative to models from OpenAI and Anthropic. According to reports, Microsoft is expected to finalize and announce its decision within the coming weeks.
This development aligns closely with a core question recently highlighted by Goldman Sachs Delta-One trading desk head Rich Privorotsky, who termed the current AI industry's pricing dilemma a "trillion-dollar question": does lowering the cost of intelligence create more demand, or does it destroy pricing power?
On this issue, Microsoft appears to have already voted with its actions. Supporting data shows the Silicon Data Token index, which tracks AI token prices, has fallen in 12 of the last 13 trading sessions, approaching recent lows.
What's Driving Microsoft's Model Switch? Mounting Cost Pressures
Microsoft's Copilot Cowork was previously offered to enterprise users with unlimited usage, but this approach is no longer sustainable.
Charles Lamanna, Microsoft Executive Vice President in charge of the Copilot business, stated plainly: "Some users complete hundreds of tasks per week with high efficiency—but the cost can skyrocket."
Consequently, Microsoft has announced it will switch Copilot Cowork to a usage-based billing model tied to compute consumption. Simultaneously, the company is exploring the introduction of a fine-tuned DeepSeek V4 version or other open-source models to significantly reduce model inference costs.
The underlying logic is straightforward: Chinese models are cheaper, while American models are more expensive. Token pricing data reveals a significant gap in input/output pricing between Chinese and U.S. models.
Matching Top Models at Half the Price – The Trillion-Dollar Question
Recent experimental results cited by Goldman Sachs show that a multi-model ensemble composed of Gemini 3 Flash, Kimi K2.6, and DeepSeek V4 Pro comprehensively outperformed individually run GPT-5.5 and Opus 4.8 in benchmark tests. This ensemble narrowed the performance gap to within 1% of Fable 5, at approximately half the cost.
Privorotsky characterized this result as a "direction the market has consistently underestimated."
This trend has dual market implications.
The bullish case: Falling costs and lower barriers to entry should ultimately drive increased AI usage and a corresponding expansion in computing demand.
The bearish case: This directly accelerates token deflation, undermining the sustainability of the existing model economics.
Privorotsky distilled the core conflict into one question: "Does lower intelligence cost create more demand, or destroy more pricing power?" He calls this the "trillion-dollar question."
Is Cheap AI Ultimately a Positive or Negative?
The Real Impact: Does the Trillion-Dollar Capex Thesis Still Hold?
Microsoft's potential choice has implications far beyond a simple supplier switch for one company.
A core assumption behind the massive capital expenditures committed by tech giants over the past two years has been that enterprise customers would continue purchasing expensive, top-tier U.S. AI models, generating revenue to justify these investments.
However, if even Microsoft—OpenAI's largest investor and partner—is finding OpenAI too expensive and considering Chinese open-source alternatives, the foundation of that assumption begins to weaken.
Reports also note that since controversy over "token prices hitting all-time highs" emerged, the Silicon Data Token index has fallen in 12 of the last 13 trading sessions, a clear signal of the market voting with its feet.
Microsoft's multi-model strategy also reflects a broader industry shift—moving away from betting on a single supplier toward flexibly allocating models based on task complexity and cost. For OpenAI and Anthropic, this signifies a dilution of their bargaining power.
Comments