Back to News
EnterpriseAI Understanding briefing

DeepSeek reports $1 billion ARR, adds CFO and prepares STAR Market IPO

DeepSeek announced annualized recurring revenue above $1 billion, appointed a new CFO, and moved into due‑diligence for a Shanghai STAR Market listing while rolling out a new V4.1 Flash model and a two‑tier pricing strategy.

5 min readRead the linked source
Source-page capture accompanying DeepSeek reports $1 billion ARR, adds CFO and prepares STAR Market IPO
Source referenceSource recorded
Publisher
eu.36kr.com
Source link
eu.36kr.comhttps://eu.36kr.com/en/p/3998096784562053
Source type
Linked source — primary-source status has not been established.
ContextUnderstand this in 60 seconds

Start here

Key terms

API (Application Programming Interface)
A structured way for one software system to send requests to and receive responses from another system.
Mixture of Experts (MoE)
An architecture with specialized subnetworks where only selected experts run per input.
Benchmark
A standardized test or dataset used to measure and compare model performance.
Test yourselfAI Models Explained Quiz

What happened

DeepSeek’s annualized recurring revenue (ARR) has topped $1 billion, more than doubling from under $500 million a few months earlier, according to The Information. The company raised API prices in August, yet reported no customer churn, and its gross profit margin on API services reached 82.9% in the first seven months of 2026. At the same time, DeepSeek hired Yan Wentao as chief financial officer, ending a three‑year vacancy. The firm is in the due‑diligence stage with CITIC Securities for a STAR Market IPO and is targeting a $75 billion financing round by October. On September 10, DeepSeek released the V4.1 Flash model—a 552‑billion‑parameter MoE model with lower activation costs—and cut its API price by up to 60% for this entry‑level offering. The company also announced plans to deploy 160,000 Huawei Ascend 950DT chips for inference in a new data centre in Inner Mongolia and is recruiting 150 engineers focused on server‑side and agent‑elastic computing.

The Information, citing sources familiar with DeepSeek, reported that the company’s ARR exceeded $1 billion, more than doubling from under $500 million a few months earlier. The revenue surge is attributed to a mid‑August API price increase that raised model invocation fees by 2.3‑4.5×, with cache‑hit prices for V4 Pro climbing roughly 1,100 % during peak hours. Despite the hikes, DeepSeek’s CEO Liang Wenfeng said demand remained strong and churn was negligible.

DeepSeek’s first‑seven‑months‑of‑2026 revenue reached about $475 million, with an 82.9 % gross profit margin on its API business and a 44.6 % overall gross margin, indicating a profitable model rather than a loss‑leader. The company’s revenue is still derived almost entirely from API charges, as its conversational bot app remains free and ad‑free.

In parallel, DeepSeek appointed Yan Wentao, a former GL Ventures partner, as CFO on September 21, signaling a move toward stronger financial governance. The firm is in the due‑diligence phase with CITIC Securities for a STAR Market IPO, though a formal listing counseling agreement has not yet been signed.

On September 10, DeepSeek launched the V4.1 Flash model, a 552‑billion‑parameter mixture‑of‑experts (MoE) system with a novel causal‑encoder‑decoder architecture. Benchmarks show it scoring 90.6 on Terminal‑Bench 2.1, 74.2 on DeepSWE v1.1, and 3,471 on Codeforces Rating, outperforming the flagship V4 Pro on several metrics. The API price for this model was cut by up to 60 %, creating a two‑tier pricing scheme that raises flagship model fees while discounting entry‑level offerings.

DeepSeek disclosed plans to install 160,000 Huawei Ascend 950DT chips in a new Inner Mongolia data centre for inference, aiming to build one of the largest domestic AI chip clusters. The company also announced a recruitment drive for about 150 engineers focused on server‑side development and agent‑elastic computing, with no AI research positions listed.

Source details: eu.36kr.com ↗

Why it matters

The revenue milestone shows DeepSeek’s ability to monetize its large‑language‑model APIs at scale, positioning it as one of the few Chinese AI firms with a $1 billion ARR without relying on consumer‑facing monetization. The CFO appointment and imminent STAR Market IPO suggest the company is moving from a growth‑stage startup to a publicly listed enterprise, which could unlock significant capital for further R&D and chip integration. The two‑tier pricing strategy—raising prices for flagship models while discounting lighter models—aims to broaden market reach and lock in diverse customer segments. Deploying a massive domestic chip cluster signals a strategic bet on Chinese hardware, potentially reducing reliance on NVIDIA and reshaping supply‑chain dynamics in the AI training and inference market. Together, these moves could accelerate DeepSeek’s competitive stance against global rivals such as OpenAI and Anthropic, especially in the Chinese market where regulatory and cost considerations are paramount.

The $1 billion ARR milestone demonstrates that DeepSeek can generate substantial revenue from API services alone, a rare achievement among Chinese AI firms that often rely on consumer‑facing products. This profitability reduces the need for heavy subsidies and suggests a sustainable business model.

The CFO hire and STAR Market IPO preparation indicate DeepSeek’s transition to a mature, publicly‑listed entity, which could attract institutional investors and provide the capital needed for large‑scale model training and chip procurement.

The two‑tier pricing strategy may set a precedent for differentiated pricing across model tiers, encouraging broader adoption while preserving high‑margin revenue from premium services.

Deploying a massive domestic chip fleet could lessen dependence on foreign hardware, lower inference costs, and align with Chinese policy goals for technology self‑sufficiency, potentially influencing the broader AI hardware market.

The release of V4.1 Flash, with strong performance and lower activation costs, expands DeepSeek’s product portfolio and may attract developers seeking cost‑effective, high‑performance models, thereby strengthening its ecosystem.

Interactive Mechanism

Interactive Mechanism: How It Actually Works

Explore the underlying technology behind this development interactively.

Agent Lifecycle Stage:
1
User Intent & Planning: "Audit customer refund request #4092 and settle payment."
2
Tool Calling: Emits structured JSON call crm_get_transaction(id='4092').
3
Guardrail & Verification:🛡️ Paused: High-value action requires human operator sign-off.
4
Final Settlement: Refund recorded, email receipt dispatched, and audit log stored.
Core takeaway: An AI agent is not just a language model—it is a closed loop of planning, tool invocation, and environment feedback. Production systems require self-healing retries and strict human approval guardrails.
Interactive Concept Check+10 Points
AI Models Explained Quiz

In AI, what are a model's "parameters"?

What to watch next

Investors and analysts will watch the progress of DeepSeek’s STAR Market IPO filing and the size and terms of the upcoming $75 billion financing round. The performance and adoption of the V4.1 Flash model, especially its pricing impact and results, will indicate whether the two‑tier pricing approach gains traction. Additionally, the rollout of Huawei Ascend chips for inference will be monitored for its effect on cost efficiency and potential supply‑chain independence. Finally, the company’s hiring drive for engineering talent may reveal the pace of its upcoming agent‑layer development and broader product roadmap.

The outcome of DeepSeek’s STAR Market IPO filing, including pricing, valuation, and investor appetite, will be a key indicator of market confidence in Chinese AI firms.

The size, timing, and terms of the targeted $75 billion financing round will reveal how much capital DeepSeek can secure to fund its ambitious R&D and chip deployment plans.

Adoption rates and developer feedback on the V4.1 Flash model, especially regarding its pricing and performance relative to competitors like OpenAI’s GPT‑6 Luna, will show whether the two‑tier pricing approach gains traction.

The operational performance of the Huawei Ascend chip cluster, including cost savings and inference latency, will be closely watched as a test of China’s domestic AI hardware strategy.

Progress on the recruitment of engineering talent and the development of the DeepSeek Harness agent framework will indicate how quickly DeepSeek can move from model‑centric offerings to full‑stack AI solutions.

Related guides & quizzes

AI Models ExplainedAI TrainingFuture of AITest what you know — try a free AI quizLook up an AI term in our glossaryFollow the AI funding tracker
Found this useful?