Mr. Duan has also invested in AI

Mr. Duan quietly invested in AI.

Duan Yongping's portfolio hasn't changed for quite some time. For a long period, he mainly made incremental adjustments or played with financial instruments, for example, call and put options.

According to disclosures, in Q4 2025, H&H International Investment, managed by Duan Yongping, opened positions in three AI IPOs at once. What's different from previous years is that these are not traditional platform giants, but three very granular vertical AI enterprises—AI computing power leasing provider CoreWeave, high-speed connectivity chip company Credo, and AI healthcare company Tempus AI.

The allocations in these three positions are very light, totaling only 0.28%, equivalent to about 340 million RMB (48.31 million USD). In the context of a total portfolio of 120 billion RMB, this is a drop in the bucket—a typical observation position. However, Duan Yongping’s philosophy has always been “don’t invest in what you don’t understand,” so even a small move signifies a lot.

Another noteworthy point is that Duan Yongping sharply increased his holdings in Nvidia, ramping up eleven-fold at once (an increase of 1,110%), and his position surged from 0.7% to 7.72%. Nvidia has now surpassed Pinduoduo and become his third highest holding, behind only Apple and Berkshire, with a market value of approximately 9.5 billion RMB (1.35 billion USD).

Considering that in Q4, Nvidia’s market cap once broke through 5 trillion USD, becoming the world’s first 5-trillion-dollar company, this “chasing high” move appears bold, even aggressive.

At the same time, Duan Yongping also significantly increased his holdings in TSMC (by 370.95%) and made smaller additions to Google. After these operations, the AI concentration of Duan Yongping's portfolio has become even higher.

To summarize, Duan Yongping is investing in AI in a very positive and advanced posture, leading among value investors, but his method remains unchanged. He sticks to the standard Duan Yongping template: start with observation, test the waters with a small position, and once he understands, make a heavy bet, making money consistently and safely time after time.

This model can be seen in any of Duan Yongping’s big holdings, whether it’s Apple, Kweichow Moutai, Tencent, Pinduoduo, or Nvidia—the philosophy “don’t invest in what you don’t understand” has never changed. His primary goal has never been to make quick money, but to win in the long run.

The only difference this time is that he’s investing in technology, in AI. Now that investing in AI has become a global consensus and even political correctness, Duan Yongping has made AI part of his “duty.”

Three AI companies that don’t make much profit

As the title says, this is a bit surprising.

The first is CoreWeave, an AI computing power leasing company founded in 2017. Initially, it was a crypto mining firm, but after the crypto market cooled, it pivoted to being a GPU cloud computing infrastructure provider, mainly supporting large model training and inference with computing power. By 2025, CoreWeave went public on NASDAQ.

CoreWeave has the nickname “Nvidia’s godson,” with its rapid rise closely linked to Nvidia’s strong support, especially in securing the latest GPUs, technology, and financing—all of which were prioritized thanks to Nvidia.

CoreWeave precisely addressed the pain point that AI giants were too slow at building their own data centers. It aggressively secured sites, electricity, procurement and deployment, rapidly leasing and constructing data centers to capture early massive market opportunities. However, data centers are heavy assets, so to expand quickly, CoreWeave started to play financial games—borrowing from financial institutions with long-term contracts from Nvidia and Microsoft, buying Nvidia GPUs with loans, and then repaying debt with data center rental incomes.

In addition, CoreWeave is tightly tied with several giants, and its revenues are highly concentrated. For example, in 2025, Microsoft alone accounted for 67% of its revenue, and after OpenAI signed a five-year, 12 billion USD mega-deal, it added another 4 billion USD. This helped CoreWeave lock in a first-mover advantage for several years. However, the disadvantage of such customer concentration is also clear: a single change can lead to passivity, weak bargaining power, etc.

In summary, CoreWeave’s characteristics are precise timing, bold moves, and lightning-fast expansion. The company is not yet profitable, but is growing rapidly. In 2025, CoreWeave’s revenue is over 5 billion USD with a net loss of 1.17 billion USD. Other significant pressures include capital leverage-induced debt, long payment cycles of big contracts, and hardware depreciation.

CoreWeave’s stock is equally aggressive. Last year, its market cap once hit six times its IPO valuation before halving, now at around 68 billion USD.

The second company, Credo (MaxLinear Technologies), was founded in 2008 in Shanghai Zhangjiang by three former Marvell Chinese engineers. Founder and CTO Lawrence Cheng was previously Director of Analog Design for storage products at Marvell. In 2022, Credo went public on NASDAQ.

Worth noting, the founder of Walden International and Intel CEO David Liwu Chen is a board member at Credo; Walden International was also among the first-round investors in 2015. According to their official WeChat account, Shanghai is Credo's global R&D center, with subsidiaries in Silicon Valley, Hong Kong, Taiwan, and Wuhan.

Credo focuses on high-speed connectivity solutions for AI data centers. Its core products include active electrical cables (AEC), optical DSP chips, and retimers. The AEC product is mainly used in data center scenarios, realizing high-speed, stable, low-latency, and high-performance data transmission, and accounts for 88% of the segment market share.

In July last year, Elon Musk posted a photo inside xAI’s data center on X, featuring thousands of purple cables—all of which were Credo's active electrical cables. Industry research estimates that a single purple AEC from Credo sells for 500–650 USD.

Unlike CoreWeave's high-profile launch, Credo went through years of technology accumulation and product iteration. The founder even once personally funded operations. Like CoreWeave, Credo benefited from the giants’ AI infrastructure race—Amazon, Microsoft, Meta, Google, xAI, and other major AI companies are all Credo’s clients.

Starting in 2024, Credo entered a period of rapid development. In 2025, Credo achieved revenues of 437 million USD, a year-on-year growth of 126%, and was profitable for the first time. Currently, Credo’s stock price has increased from just over 1 billion USD at IPO to more than 20 billion USD—a 20-fold increase.

Both CoreWeave and Credo are “shovel sellers” in the AI infrastructure field. The former uses full leverage and bold expansion; the latter positions accurately and reaps the rewards. Both benefited from the AI arms race among several giants, realizing explosive revenue and market value growth.

The third company is different. Tempus AI is an AI precision medicine company founded in 2015 by Eric Lefkofsky, a co-founder of Groupon. Groupon was the original group-buying platform in the US, which Meituan in China initially modeled itself after.

Tempus has enjoyed smooth sailing in the capital markets, raising a total of 1.3 billion USD in nine funding rounds before its IPO. In 2024, Tempus went public on NASDAQ with a first-day market cap exceeding 6 billion USD. The original fundraising target was 100 million USD, but the company ended up raising 400 million USD. Recently, Kathy Wood (“Sister Wood”) and “Congressman Stock Goddess” and former US House Speaker Nancy Pelosi also increased their holdings in Tempus AI stock.

Tempus’s business falls into three lines: genomic diagnostic testing, data services, and an AI application platform. These are not isolated; they collaborate with each other. Tempus uses genetic testing as both a traffic and data entry point, continuously rolling out testing products to expand tumor diagnosis coverage, and accumulating data. In 2022, through acquiring medical imaging AI company Arterys, Tempus entered medical imaging and AI.

When the underlying dataset is large enough, Tempus can, on one hand, train its own AI models, including disease and genetic testing models etc., to assist diagnoses and treatment. On the other hand, it can “multiply” the data's value, anonymize and package it into data products, and sell them to pharmaceutical companies. This is Tempus's fastest-growing and most profitable business, with gross margins of about 75%.

By 2025, 19 of the world’s top 20 pharmaceutical companies had become Tempus’s clients, including AstraZeneca, GlaxoSmithKline, Pfizer, Novartis, Eli Lilly, Sanofi, etc.

Like Credo, Tempus is in a stage of hyper-growth, just crossing into profitability. In 2025, Tempus posted revenues of 1.27 billion USD, up 83% year-on-year. Adjusted EBITDA was positive for the first time—about 12.9 million USD.

Just in December 2025, the quarter Duan sir opened positions, Tempus hit its historical high market cap of about 13.5 billion USD, doubling since its IPO, and has now retreated to around 9.3 billion USD.

Doing AI is just “duty”

It’s obvious that these three new stocks are completely different from past picks. They’re not industry giants with mature moats but AI rookies that have just validated their market, built some barriers and highlights, and are entering an explosive phase.

Duan Yongping hasn't publicly commented on opening these three positions. It's possible that next quarter’s 13F will see them gone, making them strictly observation positions. But it’s clear that Duan Yongping’s stance toward the AI wave is even more aggressive and cutting-edge than expected. He usually takes action rarely and has a low portfolio turnover, so the establishment of positions in three AI companies at different industry chain nodes and with distinct business models at least proves that Duan has made systematic and very granular research into the AI industry.

His approach to AI has been step-by-step, from watching, to testing, to eventually making large allocations. The first time he publicly mentioned AI was in 2023, stating on Snowball (Xueqiu), “AI is a major direction and worth paying attention to, but I don’t understand it yet, so I won’t jump in easily.” After a year, Duan concluded his AI study: “AI isn't simply a technological upgrade, but a revolution at the industrial level, possibly more impactful than the internet and the traditional industrial revolution—this trend is irreversible and not to be missed.”

A year later, Duan made a move. In Q3 2025, he first bought into Nvidia, taking a 0.7% position. He posted on Snowball, “Nvidia is truly impressive. I’ve watched many videos of Jensen Huang; what he talked about over a decade ago is the same as today. He saw the direction for AI long ago and always moved toward it. That kind of strategic resolve is important.”

That's finding someone who resonates. Recently, we talked with investor Qu Tian, one of J&T Express's earliest investors and someone very close to “Duan’s circle”. He recalled an interesting detail: whether at OPPO, vivo, or J&T Express, the business philosophy and corporate culture have never changed in thirty years—again and again, just these five words: duty, normal mind.

Reading through Duan Yongping's posts on Snowball and open interviews—“don’t invest in what you don’t understand,” “buying stocks is investing in companies”—you’ll see he’s repeated the same ideas over and over. Duan found a kindred spirit in Jensen Huang, whose philosophy hasn’t changed in a decade.

Later in Q4 2025 came his aggressive eleven-fold increase in Nvidia holdings—again putting values into action.

AI is a new thing, yet Duan has branded his approach to AI as an extension of his “duty.” “Duty” is a philosophy, a vessel—not at odds with novelty. That’s the brilliance of Mr. Duan’s approach.

Another representative of value investing, Li Lu, has taken an even more pragmatic, almost conservative stance on AI, holding Google as his only AI-related position.

Li Lu first bought Google in Q2 2020—a contrarian bet during a downturn. He kept adding while the market fell in 2022, and since then, Google has been his No. 1 holding, only being slightly reduced in Q1 2025. According to the latest data, Google now accounts for 44% of Li Lu’s portfolio.

When Li Lu opened his Google position, large language models didn't exist yet, so AI may not have been the main consideration—but it hasn’t prevented Google from being another outstanding investment for Li Lu.

Roughly estimated, Google’s unrealized gains for Li Lu are at least 1 billion USD; in just Q4 2025, his Google holdings grew by 338 million USD in market value. This quarter, Google released Gemini 3.0 and delivered a great financial report, quickly winning consensus among large funds. The three global asset management giants—BlackRock, Vanguard, State Street—plus Norway’s sovereign wealth fund (the largest in the world), as well as Fidelity, Goldman Sachs, Ark Invest, and top domestic hedge funds like Dan Bin and Jinglin, all made Google a top holding.

In Google, Li Lu is forward-looking; in terms of AI investing, he's conservative. He opted not to invest directly in AI industry companies, but in Google—a giant with a wide moat and high certainty, the “no-fail ticket” in the AI age. Of course, his position may relate to his identity and responsibilities: Li Lu is a professional investor, while Duan Yongping mainly manages his own money and has more freedom.

Another giant, Buffett, also bought Google. Berkshire opened its Google position in Q3 2025, investing over 4 billion USD—making Google 1.62% of the portfolio, among the top ten holdings. Google rose over 13% in Q3.

Back to Duan Yongping. He's now fully immersed himself in AI—not only by buying stocks but also in his daily life and work. He once said on Snowball that he uses AI for fitness—“AI is the best fitness coach”; when chatting with netizens about healthcare, he asserted that AI absolutely outperforms the (top-tier) average... “It may be faster, more convenient, even more accurate than an expensive private doctor.” He even reviews contracts with Gemini and ChatGPT, exclaiming “very impressive.”

In a New Year’s post, Duan said, “In the new year, I need to seriously learn how to use AI, hoping to understand it enough to make big bets.” No doubt, this year Duan Yongping will continue increasing his AI investments.

Of course, even though he values AI, he still defines it as a tool—as an amplifier. “Many questions can be asked of AI—just not about which stocks to buy to make money.”