Compared to "open-source models," the premium on "frontier models" is similar to that of luxury handbags! Deutsche Bank: This could lead the market to reassess AI.

Compared to "open-source models," the premium on "frontier models" is similar to that of luxury handbags! Deutsche Bank: This could lead the market to reassess AI.

Deutsche Bank directly points to the most core yet most structurally underestimated contradiction in today's AI industry: **there is a staggering "cost gap" between cutting-edge proprietary AI models and open-source/open-weight models**. On June 20, according to Chase Trading Desk, Deutsche Bank stated in its latest research report that the data is clear: Anthropic’s top frontier model Claude Fable 5 scored 60 on the Artificial Analysis Intelligence Index, with a weighted average cost of about **$3.25 per task**; while DeepSeek V4-Pro scored 44, with a cost of about **$0.05 per task**—the former is about **65 times** that of the latter. However, for roughly 90% of ordinary daily tasks, cheap models perform almost as well as frontier models. Deutsche Bank's core judgment is: **the high premium of frontier proprietary models resembles the "status symbol" pricing of luxury handbags, rather than pure performance premium**. As leading AI companies prepare for IPOs and shift from subscription to token-based billing, enterprise users will be forced to re-examine whether this "frontier premium" is worth paying. This trend may trigger a market revaluation far deeper and more enduring than the "DeepSeek Moment" in early 2025. **This may indicate** that the "operating cost narrative" in the AI industry is quietly replacing the "compute demand narrative" as the new pricing anchor. If the true cost-efficiency of proprietary models is fully priced in by the market, the valuation pressure on AI-related stocks will be structural, not temporary. ## The Numbers: How Deep Is the Cost Gap Between Frontier and Open-Source Models? Deutsche Bank cited Artificial Analysis Intelligence Index data to visually compare mainstream AI models based on intelligence scores and per-task costs. **Frontier proprietary model camp:** > **Anthropic Claude Fable 5** (with fallback mechanism): Intelligence Index score **60**, weighted average cost per task about **$3.25** > > **OpenAI GPT-5.5** (ultra-high configuration): falls in the high-cost range > > **Google Gemini 3.1 Pro Preview**: also in the high-cost frontier camp **Open-source/open-weight model camp:** > **DeepSeek V4-Pro** (max configuration): Intelligence Index score **44**, per-task cost about **$0.05** > > **Meta Muse Spark**: low-cost range (price data currently unavailable) > > **Nvidia Nemotron 3 Ultra**: low-cost range > > **OpenAI gpt-oss-120b** (high configuration): low-cost range Deutsche Bank specifically points out that **the cheap camp is not composed solely of Chinese models**. Open-weight models from Meta, Nvidia, and OpenAI itself are also in the low-cost range. Thus, the essence of this competition is **not "US vs China", but "frontier proprietary vs open weight"**. Deutsche Bank uses a vivid analogy to describe this mismatch in performance and cost: **frontier models are roaring brand-new supercars, open-source models are reliable second-hand family station wagons**. The report admits that in the toughest reasoning tasks and agentic work, frontier models do have real and significant capability advantages. The gap between intelligence scores of 60 and 44 is substantive when handling the most complex tasks. However, the key point is: **for roughly 90% of everyday tasks, cheap open-source models can accomplish almost the same work, at only about 1.5% of the cost of frontier models**. This means the vast majority of enterprise users pay a premium for frontier models not out of actual business need, but more because of brand perception, habitual inertia, or even a sense of identity as "using the best AI"—which is highly similar to the psychological logic of consumers buying luxury handbags. ## The Premium Won’t Disappear, Only "Migrates"—But Every Layer Is Sliding Toward Zero Deutsche Bank raised an important structural observation: **the cost of AI capabilities is dropping at about tenfold per year**, but the frontier premium doesn’t disappear—it keeps "migrating". The logical chain is as follows: 1. Today's frontier models will become commoditized capabilities tomorrow; 2. Meanwhile, new generations of stronger frontier models will emerge with new high premiums; 3. Therefore, the price gap between "best available" and "good enough" is structurally long-lasting; 4. But every point along the pricing curve keeps sliding toward zero. This mechanism means: **the frontier AI premium is a perpetually moving target**, not a fixed moat. This requires ongoing, dynamic evaluation of AI companies’ pricing power and profit margins—not one-off judgments. ## Business Model Shift Under IPO Pressure: From "Subscription" to "Pay-as-You-Go" Deutsche Bank notes that the AI cost issue has become particularly urgent due to a key commercial catalyst: **leading proprietary AI labs are preparing for IPOs, shifting business models from fixed-rate subscription to usage-based token billing**. This shift will transmit cost pressure directly to enterprise users. The report cited a highly persuasive real case: > **Uber has already burned through its entire token budget in just four months, and has now capped every employee’s monthly AI usage at $1,500.** Deutsche Bank argues this case clearly illustrates: **when AI usage cost shifts from "implicit subscription" to "explicit pay-as-you-go", cost awareness among enterprises activates quickly**. Those who only need a "reliable workhorse" rather than a "supercar" will increasingly ask: **is the frontier premium really worth paying?** ## "Second DeepSeek Moment": Quieter but Possibly More Far-Reaching Deutsche Bank compared the current situation to the "DeepSeek Moment" of early 2025 and issued a forward-looking warning worth close attention. **Recap of the "DeepSeek Moment" in early 2025:** The market realized that near-frontier AI capabilities could be built at far lower than expected costs, causing AI stocks to be sharply hit. But as AI overall demand continued to rise, the stock market recovered. **Deutsche Bank’s judgment:** The brewing "operating cost narrative" is a "quieter, but more lasting sequel" to that shock. The core logic: If proprietary AI models were previously priced and traded partly as **"status goods"**—i.e., their high price was part of the appeal—then once their true cost-efficiency ratio is fully exposed and priced in, **AI valuation systems may undergo a second revaluation; this impact will be less dramatic but deeper and longer-lasting**. Deutsche Bank leaves a thought-provoking open-ended conclusion: **Unless, like luxury handbags, the high price itself is AI’s ultimate selling point.** The report also cites research data from Epoch AI, providing independent evidence for the above analysis: - **The gap between the US and China in frontier AI capabilities averages around seven months;** - Epoch AI also notes that this gap is **almost exactly equivalent to the capability gap between proprietary and open-weight models**. This finding further reinforces Deutsche Bank’s core argument: **The "US-China AI gap" in the geopolitical dimension and the "proprietary/open-source gap" in the commercial dimension are essentially two expressions of the same gulf.** This means AI geopolitical risk assessment should not be separated from commercial competition logic. Risk Disclosure and Disclaimer The market has risk, investments require caution. This article does not constitute personal investment advice, nor does it take into account the individual user's specific investment objectives, financial situation, or needs. Users should consider whether any opinion, viewpoint, or conclusion herein is suitable for their own circumstances. Invest accordingly at your own risk.