DeepSeek Welds V4-Pro's 75% Discount Permanent: A High-Stakes Bet to Reshape Global AI API Pricing Logic

When a "limited-time offer" is declared permanent, it ceases to be a promotion and becomes a rewriting of the pricing system. According to confirmed information, DeepSeek has made its 75% discount policy for the V4-Pro model permanent, significantly reducing the cost of AI inference and usage. The developer community widely views this as a further escalation in the AI price war.

Why "Permanent Discount" Is Not Just a Price Cut

In the history of SaaS and cloud services, discounts are temporary levers used to acquire customers, while pricing is the true expression of the business model. What makes DeepSeek's move so noteworthy is not simply that it is "cheaper," but that it has welded a promotional variable—originally revocable at any time—into the long-term price list.

This implies two things: First, enterprise customers and developers can now base long-term assumptions on this new price when making technology selections, budgeting, and architectural design—the scarcest form of "certainty" in the API market. Second, DeepSeek is effectively sending a signal to competitors that is not easily replicated: it is willing to treat the current price level as a floor, not a ceiling.

Temporary discounts are marketing moves; permanent discounts are business model statements. The former tests cash flow, the latter tests unit economics.

An Unusual Signal: Why "Permanent" Instead of "Extended for Another Quarter"?

This is the most pressing question in the event. A rationally run AI company would not easily make a discount permanent without sufficient confidence in its inference cost structure—because doing so voluntarily forfeits future pricing power. Several possible explanations exist, but all point to the same underlying trend:

• The inference cost curve has sunk: Engineering dividends from model architecture optimization, MoE sparse activation, KV Cache reuse, and improved inference chip utilization may have pushed the marginal service cost of V4-Pro to a level where the 75% discount still allows for a viable business.
• A bet on volume over margin: Permanently low prices can lock in developer migration, converting market share into inference call volume, which then dilutes fixed costs. This is a classic "internet playbook" applied to the AI infrastructure layer.
• A judgment on competitors' cost structures: If DeepSeek believes that under the closed-source model approach, OpenAI and Anthropic cannot match per-token inference costs in the short term, then "welding" the price in place becomes a structural attack—forcing them to choose between margin and market share.

It must be clearly noted that the above is derived from industry logic, not from DeepSeek's disclosed motivations. The unit economics and long-term sustainability behind the permanent discount remain a blank area in public information.

What It Means for OpenAI and Anthropic

The real destructive power of a price war has never been "the competitor lowered its price," but rather "the competitor turned the price cut into a new baseline." When the developer community begins using V4-Pro's discounted price as the mental anchor for what constitutes a "reasonable API price," OpenAI and Anthropic face not just a response to a single promotion, but a downward shift in the entire market's pricing benchmark.

This poses a sharp question to the two leading closed-source vendors: Can a high premium still be independently justified by "capability leadership"? In the past year, the capability advantage window of closed-source models has been rapidly narrowing against open-source and low-cost competitors. And when the price gap widens from "2–3x" to "4x or more," enterprise procurement committees will recalculate ROI—especially in mid-to-long-tail scenarios where latency and compliance requirements are not extremely stringent.

A Necessary Skepticism

A professional perspective cannot be one-sided. The credibility of the permanent discount promise ultimately depends on three questions that have not been publicly verified by data:

• Has DeepSeek disclosed the gross margin structure of V4-Pro inference services? Not yet.
• What does "permanent" mean at the corporate governance level? Is it written into service terms, or is it marketing rhetoric? Public information remains unclear.
• If upstream compute prices experience drastic fluctuations, does this commitment include a force majeure clause? Unknown at this point.

In other words, while developers welcome the price reduction, they should still base critical business on the robust assumption that "prices may adjust"—a baseline discipline any engineering team relying on external APIs should maintain.

Independent Judgment

The true strategic significance of DeepSeek's "permanent discount" lies not in low prices themselves, but in its attempt to redefine the price anchor of the global AI API market. Once the developer mindset shifts, even if closed-source vendors do not follow with price cuts, they will have to answer a harder question: How long can your premium capability moat hold? This is a multi-front game about cost curves, market narratives, and time windows. In the short term, it appears as a price war; in the long term, it is a redistribution of pricing power at the AI infrastructure layer—and this redistribution is far from over.