Amid a Wave of Price Hikes, DeepSeek Chooses to Clear the Field

A Wave of Price Hikes Sweeps In, Changing the Industry Landscape

In 2025, the large language model industry reached a subtle turning point.

After the brutal "Hundred Models Price War" of 2024, major domestic LLM vendors seemed to have reached a tacit agreement — raise prices. Starting from the beginning of the year, multiple leading vendors successively increased their API pricing, citing largely identical reasons: soaring compute costs, upgraded model capabilities, and business sustainability. Almost overnight, "price cuts are unsustainable" and "returning to value-based pricing" became the dominant industry narrative.

Yet amid this wave of price hikes, DeepSeek chose a radically different path — not only refusing to follow the increases but continuing to offer services at extremely low prices while maintaining its open-source strategy. This is not simple price competition; it is a carefully designed operation to "clear the field."

The Logic and Hidden Risks Behind Price Hikes

Vendors are not entirely without reason for raising prices.

The costs of training and running inference on large models remain steep, especially in the race toward larger parameter counts and stronger reasoning capabilities, where GPU cluster investments easily run into the billions. The 2024 price war left many vendors trapped in a "selling at a loss for market share" dilemma — API call volumes looked impressive, but revenue fell far short of covering costs. For startups that need to answer to investors and large-company business units that need profit statements, raising prices is a pragmatic choice for survival.

But the risks of price hikes are equally apparent. The application ecosystem for large models is still in its early stages, with vast numbers of developers and small-to-medium enterprises just beginning to embed AI capabilities into their products and workflows. Higher prices mean higher barriers to entry, and cost-sensitive long-tail users may be forced to exit or turn to alternatives. More critically, when everyone is raising prices, whoever can offer a "high-quality yet affordable" option stands to capture the entire incremental market.

DeepSeek's Logic for Clearing the Field

DeepSeek's strategy is called "clearing the field" rather than simple "low-price competition" because it is backed by a complete set of structural cost advantages and strategic intent.

First, a technology-driven cost revolution. DeepSeek has carved out its own path in architectural innovation. From the MLA (Multi-head Latent Attention) mechanism introduced in DeepSeek-V2, to the deep optimizations on Mixture of Experts (MoE) architecture in the V3 and R1 series, DeepSeek has achieved dramatically lower inference costs at comparable performance levels. According to publicly available data, DeepSeek's inference costs are only a fraction of those of competitors at the same tier. This means that while others maintain low prices at a loss, DeepSeek may already be approaching breakeven or even turning a profit.

Second, building a moat through open source. DeepSeek's insistence on open-sourcing its core models may appear to be "cutting off its own revenue stream," but it is actually constructing a deeper competitive barrier. Open-source models have attracted a massive number of developers and enterprises to fine-tune and deploy on top of them, creating enormous ecosystem stickiness. Once developers have built entire workflows around DeepSeek's architecture, migration costs become prohibitively high. This ecosystem lock-in effect carries far more long-term value than API pricing alone.

Third, squeezing competitors' survival space. By maintaining low prices amid a wave of hikes, DeepSeek is effectively applying selection pressure on the entire industry. Second- and third-tier LLM companies that rely on venture funding to subsidize operations can neither match the low prices (lacking cost advantages) nor retain users through price increases (lacking sufficiently clear performance advantages). They will be caught in the middle, trapped in a no-win situation. This is the essence of "clearing the field" — not killing competitors, but making them exit on their own.

The Transfer of Pricing Power

The deeper significance of this shakeout lies in the transfer of pricing power in the large model market.

Over the past year, pricing power was effectively held by the leading cloud vendors. They possessed the largest customer bases, the most complete infrastructure, and the strongest brand endorsements. Smaller players could only follow passively — either bleeding alongside price cuts or losing customers alongside price hikes.

But DeepSeek's emergence has shattered this dynamic. With its extreme cost-effectiveness and the rallying power of its open-source ecosystem, DeepSeek is becoming the market's "price anchor." When developers evaluate LLM solutions, DeepSeek's price and performance have already become the default benchmark. This means that even leading vendors must now factor in DeepSeek's presence when setting prices.

Even more noteworthy is that DeepSeek's pricing strategy is reshaping the industry's perception of "reasonable profit." When DeepSeek demonstrates that it can sustain technological iteration and commercial operations at extremely low prices, the narrative that attributes high pricing to "unavoidable costs" loses its persuasiveness. The market will begin to ask: for the same level of performance, why is your price five or even ten times that of DeepSeek?

Who Will Be Eliminated in the Shakeout?

Based on the current landscape, the following types of players will feel the pressure first:

Followers lacking technological differentiation. Vendors whose model performance sits in the industry's middle tier and who compete primarily on price and service will face the greatest existential threat. They can neither open up a technical gap with DeepSeek nor compete with it on cost.

Pure model companies overly reliant on API revenue. If a company's primary business model is selling API calls, DeepSeek's low-price strategy will directly undermine its revenue foundation. Pivoting to vertical applications or solutions may be their only way out.

Hesitant peripheral businesses within large enterprises. Some large companies have treated LLMs as a strategic investment, but internal disagreements persist about their commercialization prospects. DeepSeek's clearing effect may become the last straw that breaks the camel's back, accelerating the contraction or shutdown of these peripheral business units.

What Will the Landscape Look Like After the Shakeout?

If DeepSeek's clearing strategy continues to prove effective, the future LLM market may take on a new competitive structure:

At the top will be a handful of "super players" with cutting-edge model capabilities and complete ecosystems, including OpenAI, Google, and domestically, DeepSeek along with a few other leading vendors. Competition among them will focus more on breakthroughs at the frontier rather than on pricing.

The middle layer will consist of numerous companies building vertical applications and industry solutions on top of open-source models — most likely DeepSeek's models. They will no longer pursue developing their own foundation models but will instead focus on scenario understanding and engineering capabilities.

The bottom layer will be fully commoditized general-purpose inference services, with prices converging toward the cost of compute itself and extremely limited profit margins.

The formation of this structure means the large model industry is transitioning from the chaotic "Hundred Models War" phase to a mature period where "a few define the rules, and the majority compete based on those rules." And DeepSeek is attempting to become the one who defines those rules.

Conclusion

Maintaining low prices against the tide of hikes may appear to be an aggressive business strategy on the surface, but underneath it reflects DeepSeek's deep confidence in its own technological advantages and its clear-eyed judgment about the industry's endgame. While most players are still agonizing over "should we raise prices or not," DeepSeek is already answering a more fundamental question through its actions: who should ultimately set the prices in this market?

The answer is becoming clear.