DeepSeek Posts 10x Revenue Surge, 82.9% API Margin as Valuation Nears $70B

DeepSeek is demonstrating that China's most closely watched AI startup can generate Silicon Valley-grade unit economics at a fraction of Western rivals' scale, with new financial disclosures showing an 82.9% gross margin on its API business even as it closes in on a second fundraising round that would value the company at RMB 500 billion (US$69.4 billion).

The figures, cited by The Information and Bloomberg on August 26, 2026, mark the first granular look at DeepSeek's commercial trajectory since it upended global AI benchmarks in early 2025. The data lands at a strategically sensitive moment: the company is simultaneously negotiating its second funding round—expected to close by end of August—while investment banks are already on retainer to prepare a STAR Market IPO filing targeted for late 2026, with a listing expected in 2027.

Market participants tracking China's AI sector will note that the combination of explosive revenue growth, narrowing losses, and a near-term public listing timeline transforms DeepSeek from a research-led disruptor into a capital markets story with hard numbers to anchor valuation debates.

Revenue Rockets 10x, Exposing the Gap Between Hype and Monetization

DeepSeek generated approximately RMB 475 million (US$65.9 million) in revenue during the first seven months of 2026, roughly ten times the company's total revenue for full-year 2025, according to two people with knowledge of the financials cited by The Information.

The pace of acceleration is the critical signal. The Information had previously reported an annualized revenue run rate of US$400 million to US$500 million, and the latest seven-month figure corroborates that trajectory. Annualizing the January–July 2026 data implies a full-year 2026 revenue figure approaching RMB 800 million to RMB 850 million (approximately US$111 million to US$118 million)—still a fraction of OpenAI's US$5.7 billion in Q1 2026 alone, but a meaningful proof point for a company that was barely commercialized 18 months ago.

Net losses narrowed to approximately RMB 715 million (US$99.3 million) for the same seven-month period, compared with a full-year 2025 net loss of RMB 935 million (US$129.9 million). The directional improvement in loss magnitude—despite a near-tenfold jump in infrastructure spending—suggests that revenue is beginning to outpace the marginal cost of growth, a structural inflection that early-stage AI investors have been waiting to see.

API Margin of 82.9% Reframes the Competitive Benchmarking Conversation

The headline metric that will draw the most scrutiny from institutional investors is DeepSeek's API gross margin of 82.9%—a figure that significantly outpaces its best-capitalized Western counterparts.

For context: OpenAI reported a 39% gross margin on US$5.7 billion in Q1 2026 revenue. Anthropic's gross margin is projected to rise from approximately 40% in 2025 to 63% in full-year 2026 on Q2 2026 revenue of US$1.15 billion. DeepSeek's API margin of 82.9% already exceeds Anthropic's forward target by nearly 20 percentage points—achieved on a revenue base that is orders of magnitude smaller, which limits the absolute dollar significance, but validates the underlying cost architecture.

The overall blended gross margin for the January–July 2026 period stood at 44.6%, according to two sources cited by media reports. The gap between the blended 44.6% and the API-specific 82.9% reflects the drag from non-API revenue streams—likely consumer-facing products and enterprise deployments that carry higher service delivery costs.

DeepSeek's margin advantage is structurally rooted in its inference efficiency. The company has built its commercial model around executing more computational tasks per chip, reducing the per-token cost of running its models. This approach—widely analyzed after the release of its R1 and V4 model series—means that each incremental API call contributes disproportionately to gross profit relative to peers who rely on more conventional, compute-intensive inference stacks.

Pricing Hike Signals Monetization Maturity, While Remaining Far Below Peers

DeepSeek raised API prices this month, with output costs for its flagship V4-Pro model now set at US$3.96 per million tokens during peak hours and US$1.98 per million tokens off-peak, up sharply from a prior rate of US$0.87 per million tokens.

The increase is significant in percentage terms but leaves DeepSeek's pricing well below industry benchmarks. Moonshot AI's Kimi K3 charges US$15 per million output tokens; Anthropic's Claude Opus 5 commands US$25 per million tokens. DeepSeek's post-hike peak rate remains roughly 84% below Kimi K3 and 84% below Claude Opus 5.

The pricing strategy reflects a deliberate sequencing: build global developer adoption through aggressive underpricing, then raise rates incrementally as switching costs accumulate. The rapid uptake of DeepSeek's V4-Flash model—a lightweight, low-cost variant of the V4 series—among international developers suggests that this strategy has generated meaningful lock-in. The pricing revision is the first signal that DeepSeek believes its user base is now sticky enough to absorb cost increases without significant churn.

This dynamic mirrors the broader shift across China's AI industry. Zhipu AI, Alibaba, Tencent, and Baidu have all implemented API price increases in 2026, marking an industry-wide pivot from user acquisition subsidies toward margin recovery.

Infrastructure Spending Surges Nearly 10x, Telegraphing Scale Ambitions

The same period that produced DeepSeek's revenue surge also saw infrastructure expenditure climb to approximately RMB 11 billion (US$1.53 billion) for the first seven months of 2026—a near-tenfold increase from approximately RMB 1.2 billion (US$166.7 million) for full-year 2025.

Spending at this scale—covering server rentals equipped with AI accelerators and direct chip procurement—is consistent with reports that DeepSeek is targeting an incremental 1 gigawatt of compute capacity to support model iteration and user growth. The capital intensity of this build-out explains why the company remains in a net loss position despite strong gross margins: the gap between API-level profitability and company-level profitability is being filled by deliberate infrastructure investment, not structural inefficiency.

For investors evaluating the pending funding round, the infrastructure ramp is a double-edged signal. It demonstrates conviction in long-term demand but also underscores the cash consumption that makes the RMB 500 billion (US$69.4 billion) pre-money valuation fundraise a near-term necessity rather than an opportunistic capital raise.

Second Funding Round Nears Close, Elevating Valuation 43% Above Series A

DeepSeek's second funding round—targeting RMB 50 billion (US$6.94 billion) in new capital at a pre-money valuation of RMB 500 billion (US$69.4 billion)—is expected to close before the end of August 2026, according to the South China Morning Post.

The pre-money valuation represents a 43% premium over the post-money valuation of approximately RMB 350 billion (US$48.6 billion) established in the company's first round, which closed in June 2026 after launching in April. That first round, also sized at RMB 50 billion, was the largest Series A in Chinese AI large-model history. If the second round closes as expected, DeepSeek will have raised over RMB 100 billion (US$13.9 billion) across two rounds in less than five months.

Returning investors in the second round include Monolith Capital, GL Ventures, and Contemporary Amperex Technology (CATL), which committed approximately RMB 5 billion in the first round. New investors in advanced discussions include CPE Yuanfeng and Legend Capital.

The first-round investor roster—which includes the National AI Industry Investment Fund, Tencent, CATL, NetEase, JD.com, IDG Capital, and CATL Capital — reflects a cross-sector coalition that spans state capital, consumer internet incumbents, and industrial conglomerates, providing DeepSeek with both financial resources and distribution infrastructure.

IPO Timeline Crystallizes, Putting STAR Market Listing on a 2027 Track

DeepSeek has retained investment banks and accounting firms to begin financial due diligence, with a target of completing the audit process by December 2026, according to two people familiar with the matter. The company is expected to submit an IPO application to the Shanghai Stock Exchange's STAR Market by end-2026, with a formal listing targeted for 2027.

The STAR Market designation is significant. Designed for high-growth technology companies, the STAR Market has accommodated pre-profitability listings under specific criteria—a regulatory framework that could accommodate DeepSeek's current loss-making status given its growth trajectory and strategic importance to China's AI ambitions.

The IPO preparation timeline is aggressive by any standard. Moving from a first institutional funding round to a public listing application within approximately 20 months would require DeepSeek to simultaneously scale revenue, manage infrastructure costs, and satisfy disclosure requirements—a parallel workload that explains the urgency behind the current fundraising push.

The financial disclosures now in circulation—however limited—serve a dual purpose: they support the second funding round negotiations and begin the process of familiarizing institutional investors with DeepSeek's financial profile ahead of a prospectus filing.

Impact Assessment: What the Numbers Mean for the Broader AI Investment Landscape

DeepSeek's data points arrive at a moment when global AI investors are grappling with a central question: can AI companies generate returns commensurate with their valuations? DeepSeek's 82.9% API margin provides partial evidence that the answer can be yes—under specific conditions of infrastructure efficiency and pricing discipline.

However, the comparison with OpenAI and Anthropic requires calibration. DeepSeek's absolute revenue remains roughly 1/100th of OpenAI's quarterly run rate. At scale, maintaining 82.9% API margins while absorbing the fixed costs of a 1-gigawatt compute infrastructure and the variable costs of model development will be materially harder. The company's blended margin of 44.6%—not 82.9%—is the more relevant near-term profitability indicator.

For China's AI supply chain, DeepSeek's infrastructure spending trajectory—from RMB 1.2 billion in 2025 to RMB 11 billion in the first seven months of 2026 alone—represents a meaningful demand signal for domestic chip suppliers and data center operators, particularly as U.S. export controls continue to constrain access to Nvidia H100 and B200 series accelerators.

Related Coverage:

DeepSeek Open-Sources Harness Agent Runtime, Targeting the AI Execution Layer