DeepSeek’s Reported $500 Million Run Rate Faces a $74 Billion Valuation Test

• DeepSeek’s annualized revenue reportedly reached $400 million to $500 million, mostly from cloud API access; that is a run rate, not revenue recorded over a completed year.
• Access to its V4 model is reported to carry a 70% to 80% gross margin, but no company-wide operating expenses or net income were disclosed.
• The company may seek as much as 50 billion yuan ($7.4 billion) at a valuation of about 500 billion yuan ($74 billion), only weeks after raising roughly the same amount.
• Its mainland IPO preparations are preliminary, and the fundraising terms and timetable could change.

DeepSeek’s reported API economics give investors a commercial result to examine after a year of attention to its low-cost models. They do not yet show whether the company can fund frontier research, data centers, hiring and chip development without repeated infusions of capital.

Two reported figures anchor the IPO pitch

DeepSeek recently reached $400 million to $500 million in annualized revenue, driven largely by cloud-based sales of model access through application programming interfaces, according to a published account summarizing the original reporting. That original reporting cited three unidentified people familiar with the matter.

The attribution chain matters. The figure is attributed to anonymous sources rather than a company disclosure, and annualized revenue extrapolates a recent pace rather than measuring sales booked over a completed year. The account does not disclose the period used for the calculation, company-wide costs or net income.

The same secondhand account said access to V4, described as DeepSeek’s flagship model, produces a gross margin of 70% to 80%. It attributed the margin to infrastructure improvements that let the company handle more queries with fewer chips, even while charging less than leading US competitors.

That comparison has no current price schedule or named rival attached to it. The account’s only concrete price comparison comes from February 2025, when it said DeepSeek charged $2.19 per million output tokens while OpenAI charged $60 for its o1 model. Those were different models at an earlier point in the market, so the 27-fold gap reported then is not a like-for-like measure of V4’s present price advantage.

Product delivery margin is not company profit

The reported V4 margin is more commercially relevant than a much larger number DeepSeek publicized in March 2025, but it remains narrow. Neither figure establishes the profitability of the company as a whole.

In the earlier exercise, DeepSeek applied R1 pricing to one day of usage across its V3 and R1 services. It calculated theoretical daily revenue of $562,027 against $87,072 to lease the necessary graphics processors, producing what it called a 545% “cost profit margin,” according to an analysis of the company’s infrastructure post.

DeepSeek also said actual revenue was “substantially lower.” Nighttime discounts, cheaper V3 pricing and free web and app access meant only a subset of the measured usage was monetized. The exercise therefore compared hypothetical billings with GPU rental costs; it was not an income statement.

The V4 figure avoids that particular mismatch because it is reported as a gross margin on model access. But the available account supplies no operating-cost base, research spending or profit figure. A strong margin on delivering one product can coexist with heavy spending elsewhere in the business.

Low price is not the only competitive test, either. DeepSeek faces ByteDance and Alibaba in China, along with well-funded developers Z.ai, Moonshot and MiniMax. Its lower-cost models have encouraged business adoption, while data-privacy and national-security concerns can push companies toward self-hosting rather than DeepSeek’s consumer application. That alternative preserves use of the models but does not necessarily produce API revenue for DeepSeek.

A second giant raise reframes the efficiency claim

DeepSeek is planning a new funding round at a valuation of about 500 billion yuan ($74 billion) and could seek as much as 50 billion yuan ($7.4 billion), according to reporting based on people with knowledge of the plans. At the top of the reported revenue range, the proposed valuation would equal about 148 times annualized revenue.

Both numbers are prospective. The people cited said the financing was at an early stage and its terms and timing could change. The same report noted that separate reporting put the contemplated valuation at no less than 480 billion yuan, underscoring that there is not yet a settled price.

The scale is significant because DeepSeek had raised about $7.4 billion only weeks earlier, at a reported post-money valuation of roughly 450 billion yuan. Founder Liang Wenfeng committed 20 billion yuan to that June round. Tencent invested 10 billion yuan and battery maker CATL invested 5 billion yuan; other participants included China’s national AI fund, NetEase, JD.com and several investment firms.

That first external financing already marked a change in how DeepSeek funds itself. Liang had largely bankrolled the company through his quantitative hedge fund, High-Flyer, and DeepSeek had previously rejected outside capital. The investor list now adds large corporate and state-linked sources of money, although the available reporting does not disclose governance rights or show how control changed.

The uses of capital extend beyond serving today’s API traffic. DeepSeek said after the June round that it planned to double staff across departments including data centers and AI agents. It is also reported to be exploring an inference chip and hiring chip-design engineers. Those projects could reduce some technology dependencies over time, but the reporting establishes only that they require capital now—not that they will lower costs or succeed.

The IPO needs audited evidence, not a run rate

DeepSeek has begun preparing for a mainland IPO and may file this year for a possible 2027 debut. It has started talks with accounting and banking advisers, according to people familiar with confidential discussions. Separate reporting places the contemplated listing on Shanghai’s STAR Market.

The company did not respond to a request for comment on the fundraising and IPO plans. The evidence that would resolve the central question is still missing:

• completed-period revenue showing whether the $400 million to $500 million pace persisted and how much came from APIs;
• V4 pricing, utilization and direct delivery costs, alongside company-wide operating expenses and net income;
• final financing terms, including the amount raised, valuation and any governance rights attached to the new capital; and
• infrastructure spending sufficient to judge whether query efficiency offsets the cost of data centers, staff and chip development.

The available reporting supports a narrower conclusion than the proposed valuation might suggest. DeepSeek appears to have built a meaningful API business with attractive reported delivery economics. An IPO filing would need to show whether those economics survive scrutiny at company scale—and why another multibillion-dollar round is needed so soon after the first.