Tencent, Alibaba in Talks to Back DeepSeek at $20 Billion Valuation, Report Says

Chinese technology companies Tencent and Alibaba are in discussions to invest in DeepSeek at a valuation above $20 billion, The Information reported Wednesday, citing four people familiar with the talks. The figure has doubled in five days after DeepSeek's first outside pitch drew heavy domestic investor demand. That is up from the $10 billion target tied to a $300 million round the publication described Friday. Alibaba, Tencent, and DeepSeek did not respond to requests for comment. A close near that level would put the two-year-old Hangzhou lab alongside Moonshot AI among China's top-tier frontier labs. Alibaba's U.S.-listed shares rose about 1.6% in premarket trading.

That rise is small. The story is big.

What Liang changed his mind about

DeepSeek founder Liang Wenfeng spent two years turning down Chinese venture firms and the same two companies now at the table. He built the lab out of profits from High-Flyer Capital Management, the hedge fund that holds 99% of DeepSeek. High-Flyer had a banner 2025. The fund cleared a 56.6% return on more than 5 billion yuan in revenue. Those profits bankrolled the $5.6 million training run that produced R1 that January, and they made outside capital look like surrender. "VCs manage money for LPs. They all need returns," Liang said in one interview. "So we can never see eye to eye."

Fifteen months later, he is taking calls from those LPs. The reversal is not a liquidity story. High-Flyer has not run out of money, and DeepSeek still runs on roughly 300 people. What changed is what the money buys. You do not raise at $20 billion because you need $300 million. You raise at $20 billion because a price tag does work cash cannot.

The $20 billion isn't capital. It's a coordination device

A valuation is a market signal. It tells employees what their options are worth. It tells partners the company is real. It tells Beijing the lab is strategic enough to protect.

Consider what happened to DeepSeek's bench. Luo Fuli, a core developer of the V3 model, left for Xiaomi in late 2025. Guo Daya, lead author of R1's GRPO algorithm, went to ByteDance's Seed team. Wang Bingxuan left for Tencent during the same stretch. Headhunter accounts describe offers for top DeepSeek talent at two to three times their old salary, with immediately priced stock options attached. DeepSeek could not match on the equity line because DeepSeek's equity had no price. Without a formal round, options are abstract. With one, they are currency.

That is what $20 billion buys first. Retention. The engineers finishing V4 need a reason to stay that does not depend on Liang's ascetic speeches about research purity. Options they can value in a spreadsheet will do.

The Huawei gamble needs cover

DeepSeek's next model, V4, is being trained to run on Huawei's Ascend chips, and Huawei engineers are reportedly embedded with the team to troubleshoot training stability. Alibaba, Tencent, and ByteDance have reportedly placed bulk orders for hundreds of thousands of Huawei AI units, betting that Chinese silicon can substitute for Nvidia inside China's AI perimeter.

That bet needs a validator. If V4 ships on Ascend at frontier performance, DeepSeek becomes the adapter between China's best research and China's best domestic silicon. That role is worth more than a model lab. It is exactly the role Beijing wants funded.

An investment from Tencent and Alibaba gives DeepSeek more than a war chest. It buys inference-scale compute for V4, and it buys political cover against the regulatory friction that finds any Chinese AI firm crossing $10 billion in value. None of that shows up on a term sheet.

The efficiency story was never the whole story

Remember the $5.6 million number. It has shaped every conversation about DeepSeek since January 2025, the peg on which investors, chipmakers, and U.S. policymakers built their arguments. Reuters later clarified the figure referred only to the final V3 training run, not total R&D. Everyone ignored the clarification. The myth was more useful than the footnote.

The $20 billion valuation drags the myth back to earth. Cheap AI never erased the bill. It moved the bill downstream, into inference, reliability, chip migration, talent, and ecosystem work. DeepSeek's chatbot went down for 7 hours and 13 minutes in March, its longest outage since the viral launch. Usage kept climbing anyway. The lab started recruiting data center operations staff in Ulanqab, the Chinese compute hub in Inner Mongolia. That is not a research org. That is infrastructure.

You can read Liang's concession in the valuation math. At $20 billion, DeepSeek would sit below Moonshot AI's reported $18 billion mark and far above its own estimated $3.4 billion tag from mid-2025. It would still trail Anthropic at $380 billion and OpenAI at $852 billion by orders of magnitude. The number is not a vindication of efficiency. It is a declaration that the company has entered a category where efficiency alone stops being the point.

Who wins, who loses

Tencent and Alibaba get a hedge, and they look emboldened for taking it. Both spent 2025 and early 2026 watching their proprietary model businesses erode against free-to-download Chinese open-source rivals, a category DeepSeek helped create. Owning a piece of the most credible name in that category is cheaper than trying to out-compete it. Alibaba's Qwen family already commands over half of global open-source downloads. DeepSeek adds a second lane.

Huawei wins too, quietly. A frontier model trained on its chips is a recruiting poster for the whole domestic silicon stack.

The loser is the narrative. Nvidia's case for keeping China access rested partly on the claim that Chinese labs could not match frontier performance without its chips. A $20 billion valuation attached to a lab running Huawei hardware weakens that argument before V4 even ships, and it leaves American chip strategists looking a little cornered.

What to watch

The real test is the V4 release, expected by the end of April. If the Ascend run holds, $20 billion is cheap. If it slips again, the round closes anyway, because the money is buying the option, not the product. You will know which it was by whether Liang keeps his hand on the architecture, or whether Tencent's voice starts showing up in the release notes.

The lab that turned down Tencent and Alibaba for two years is about to become their portfolio company. That is not a pivot. That is institutionalization, priced.

TSMC's AI Bet Grows as War Tests Chip Supplies

TSMC came into Thursday's earnings call with two messages that do not sit neatly together. Customers still want more AI chips. The war in Iran is now close enough to the fab floor that investors are a

The Implicator

Jensen Huang Defends China Chip Sales. Nvidia's Real Bet Is Bigger.

Jensen Huang said Nvidia can keep winning the AI chip race even if Google TPUs gain ground, China builds its own accelerators, and the supply chain strains under trillion-dollar demand. In a wide-rang

The Implicator

Washington’s Intel stake would rewrite the chip playbook

💡 TL;DR - The 30 Seconds Version 👉 Trump administration discusses taking equity stake in Intel to accelerate domestic chip manufacturing and Ohio fab construction. 📊 Intel shares jumped 7% on

The Implicator

Maria Garcia

Los Angeles

Tech culture and generative AI reporter covering the intersection of AI with digital culture, consumer behavior, and content creation platforms. Focusing on technology's beneficiaries and those left behind by AI adoption. Based in California.