DeepSeek's Fundraising Round Is Back On at $74 Billion, and Prices Are Going Up

DeepSeek's on-again, off-again fundraising round is on again. And the company just told its customers the free lunch is over.

The Chinese AI lab is targeting roughly 50 billion yuan, about $8 billion, at a valuation near 500 billion yuan, or $74 billion, according to Caijing as reported by The Standard and confirmed by Briefs Finance. That's a sharp jump from the $52 billion valuation DeepSeek fetched in June, when it raised $7.4 billion from Tencent, CATL, Beijing's National AI Industry Investment Fund, NetEase, JD.com, Monolith Management and IDG Capital, according to ZeroHedge.

Monolith Management, a backer of Moonshot AI's Kimi model, is negotiating to take part in the new round, Briefs Finance reported. Some money is earmarked for data center buildout, including a multibillion-dollar facility planned in Inner Mongolia.

Nothing is signed yet. Briefs Finance reported the amount, timing and investor list could all still change, citing unnamed people familiar with the talks. Neither DeepSeek nor Monolith responded to requests for comment, per that report.

Why the Round Stalled in the First Place

This isn't DeepSeek's first attempt at this raise. The round was shelved roughly two weeks ago after a transcript of founder Liang Wenfeng's May 20 closed-door meeting with investors leaked onto GitHub, according to ZeroHedge.

In that meeting, Liang reportedly admitted China still trails the U.S. in AI, a gap he attributed to compute and capital rather than talent. He also conceded DeepSeek remains heavily dependent on Nvidia hardware, said Huawei's chip alternatives fall well short, and acknowledged the company could obtain processors he called "noncompliant," per ZeroHedge's account of the leaked transcript.

Liang was reportedly furious about the leak and froze the deal in response. Federal investigators have been probing whether Singapore-based intermediaries funneled restricted Nvidia chips to DeepSeek as far back as early 2025, and the House Select Committee on the CCP had reportedly reached its own conclusions on the matter by April, according to ZeroHedge. No charges or formal findings from that probe have been reported as resolved in these sources. If DeepSeek is in fact using export-controlled chips through workarounds, that represents a compliance question for U.S. regulators.

The round is now back on, and the process is reportedly being kept deliberately quiet, with some participating institutions not yet formally told the deal has restarted, according to ZeroHedge.

The Price Hike Nobody Saw Coming

A day after the fundraising news broke, DeepSeek posted a notice to its developer platform on Thursday, August 6, warning that API rates will rise across the board, and that the increase will be "significant," according to ZeroHedge and Briefs Finance.

Ground News reported the company hasn't released a specific new pricing schedule or figures yet. DeepSeek has also floated peak-hour surge pricing, potentially doubling rates during busy periods, though those adjustments haven't been activated, per Ground News.

This is a real reversal for a company that built its entire reputation on undercutting everyone else. DeepSeek's low prices triggered a massive user surge that's now straining the compute resources needed to keep the service running, according to Ground News. Every query costs real money in chips and electricity, and ultra-cheap pricing at that scale apparently isn't sustainable.

The timing is awkward. The price hike warning came barely a week after DeepSeek launched V4-Flash, a lightweight model that independent testing had crowned the cheapest capable option on the market, per Ground News. Briefs Finance noted the V4-Flash model reset cost expectations across the industry and created what some are calling a "death zone" for rivals who can't match DeepSeek's price-to-performance ratio, echoing the disruption DeepSeek caused in early 2025.

Analyst Michael Guo, cited by Ground News, questioned the timing given that Meta's Muse Spark and OpenAI's newest models now match DeepSeek on both capability and price. That competitive parity has eroded the core advantage that got DeepSeek into this market in the first place.

What's Actually Going On Here

DeepSeek needs more cash to build out compute infrastructure, including that Inner Mongolia data center, and it's charging investors a 42% higher valuation than eight weeks ago to get it. At the same time, it's telling its own customers to expect to pay more for a product that was supposed to be the cheap disruptor.

Whether investors bite at $74 billion after a leaked transcript exposed the company's own founder admitting dependence on Nvidia chips and a lag behind U.S. AI capability is the open question. So is whether regulators in Washington take a closer look at how DeepSeek keeps getting its hands on hardware it isn't supposed to have. Neither has been resolved in the reporting so far, and DeepSeek hasn't publicly addressed the chip-sourcing allegations.