DeepSeek "scared" Zhipu and MiniMax, but JPMorgan believes: V4 breaks the constraints of computing power, which is positive for the industry.

On April 24, DeepSeek released the V4 preview, causing a sharp decline in the domestic AI large model sector—was this a concentrated release of market fears about a "potential threat", or the real reshaping of the competitive landscape?

According to Wind Chasing Trading Desk, JPMorgan Securities (China) issued a research report on April 26, offering a completely different assessment of the market reaction above: The launch of V4 is beneficial to the industry rather than a zero-sum shock.

What did the market misunderstand behind the 9% plunge in share prices?

On the day DeepSeek V4 preview was launched, shares of Zhipu and MiniMax both fell 9%, while the Hang Seng Index rose 0.2% on the same day.

The market logic was straightforward: DeepSeek released another new model, competition intensified, and the survival space for pure large language model (LLM) companies is squeezed.

But the analysts above believe this is an "overreaction".

Their core assessment: The release of V4 actually strengthens three key pillars supporting China's LLM industry—computing power supply release, pricing discipline, structural cost curve compression. The only dimension where competition intensifies is relative positioning, but this is “fierce competition” rather than “rebalancing of the landscape”.

More importantly, V4 had always been the “biggest single negative competitive catalyst” tracked by analysts for the April-May event calendar. With V4 landing and being digested by the market, this uncertainty has been officially resolved.

Computing Power Bottleneck Eased: The Real Industrial Significance of V4

Analysts pointed out that the biggest “price and interpretation gap” between the market and their research framework lies precisely in computing power.

DeepSeek V4 is compatible with Huawei Ascend chips and successfully supports the inference of a cutting-edge model with 1.6 trillion parameters. This is a key validation: domestic chips are not only technologically feasible, but compared to international chips, may also have cost competitiveness.

In the past, computing constraints had always been the invisible ceiling for China’s LLM companies—they had token demand but found it hard to convert into confirmed revenue (ARR). DeepSeek has clearly said that as the Ascend 950 supernode rolls out in bulk in the second half of 2026, the inference cost for V4 Pro will further decrease.

Analysts believe this directly benefits Zhipu and MiniMax: expansion in computing power supply means these companies can provide services at lower cost, thus unlocking the conversion path from token demand to confirmed revenue.

We believe this is a key benefit for the overall Chinese LLM industry, including Zhipu and MiniMax, as it alleviates the constraint on computing power supply—one of the main barriers hindering these companies from scaling operations fully. DeepSeek states that as Ascend chips are more widely deployed and optimized, the inference cost curve will improve, opening new paths for Zhipu and MiniMax to convert token demand into confirmed revenue.

12x Price Gap: DeepSeek Itself Undermines the “Commoditization” Narrative

Another market concern: DeepSeek's low pricing strategy will turn the whole industry towards commoditization, destroying everyone’s pricing power.

Data does not support this view.

According to pricing data cited in the report (as of April 24, 2026):

• DeepSeek V4 Pro: Input/output price is $1.74/$3.48 per million tokens
• DeepSeek V4 Flash: Input/output price is $0.14/$0.28 per million tokens

The output price of the two differs by about 12 times. DeepSeek’s own product line has already established a tiered pricing system based on task complexity: high-end coding and agent workflows are priced high, simple high-throughput tasks are priced low.

This is fully aligned with the monetization logic Zhipu and MiniMax have always promoted. Analysts point out that by comparison, GLM-5.1’s output price is $3.50 per million tokens, almost the same as V4 Pro’s $3.48—so the pessimistic narrative of “DeepSeek pricing crushes everything” is inconsistent with actual data.

V4 Leads Technologically, But Does Not Suppress Rivals in B2B Core Scenarios

V4 indeed advances technically. The report cites DeepSeek’s official data showing that compared to V3.2, V4’s inference FLOPS dropped 3.7 to 9.8 times, KV cache size shrank 9.5 to 13.7 times, and long-context processing efficiency significantly improved.

But in the B2B core scenarios investors care most about—coding and agent workflows—V4 did not dominate.

According to LMArena Code Arena rankings (as of April 24, 2026):

• GLM-5.1: Ranked 5th, score 1534
• Kimi K2.6: Ranked 6th, score 1529
• DeepSeek V4 Pro: Ranked 14th, score 1456

Analysts note this ranking reflects industry normality: Each new version released prompts rankings to reshuffle. Zhipu and MiniMax are expected to release GLM-5.5 and MiniMax M3 around June, which may again refresh the rankings.

Furthermore, V4’s token compression and DeepSeek Sparse Attention (DSA) architecture are open source releases, not moats. Historically, V3’s MoE routing mechanism was absorbed by Qianwen, GLM, and Kimi in about 4-6 months. These technical innovations ultimately become industry-standard configurations, lowering the overall cost curve and benefiting all players.

Next Two Months: Negative Catalysts Cleared, Positive Catalysts Clustered

Analysts define the current time window as an “asymmetric positive landscape”.

With V4 landed, the biggest short-term competitive uncertainty is eliminated. Before the lockup shares are unlocked in July, the catalyst path for the next two months is:

• May: Index inclusion review (effective in June), bringing passive capital inflows
• Early June: Zhipu joins Stock Connect, opening the channel for mainland investors
• Around June: GLM-5.5 and MiniMax M3 released, potential step change in ARR

Risk points to watch: July, Zhipu’s 6-month lockup period expires, 5.8% of shares unlocked; MiniMax’s unlock proportion is higher, reaching 39.0%. January 2027, Zhipu’s 12-month lockup expires, then 90.3% of shares will unlock.

Analysts maintain a “overweight” rating for both companies:

• Zhipu (2513.HK): Target price HK$950 (as of December 2026), based on 30x 2030E P/E, WACC 15%. 2030E forecast revenue RMB 98.832 billion, adjusted net profit RMB 20.36 billion
• MiniMax (0100.HK): Target price HK$1100 (as of December 2026), also based on 30x 2030E P/E, WACC 15%. 2030E forecast revenue USD 9.136 billion, adjusted net profit USD 2.322 billion

Both companies’ expected revenue CAGR for 2026-2030 exceeds 100%, which is the fundamental basis for the valuation premium.

~~~~~~~~~~~~~~~~~~~~~~~~

The above excellent content comes from Wind Chasing Trading Desk.

For more detailed insights, including real-time interpretation and frontline research, please join [Wind Chasing Trading Desk▪Annual Membership]

Risk Warning and DisclaimerThe market has risks, investment should be cautious. This article does not constitute personal investment advice, nor does it take into account individual users’ special investment objectives, financial situations, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article are suitable for their particular circumstances. Investment made based on this is at your own risk.