"I was early, but I wasn't wrong."
The protagonist in the movie "The Big Short" would surely resonate deeply with these words, much like the youngest hedge fund manager on Wall Street today.
In 2024, a notice from OpenAI sent 23-year-old Leopold Aschenbrenner packing. Soon after, he turned around and founded a hedge fund, positioning himself against the prevailing consensus—he didn't go all-in on the then-sizzling hot AI chips and semiconductors, but instead went long on electricity, data centers, computing power infrastructure, and energy infrastructure. These were seen as slow, clunky "old-school" assets at the time.
His early calculation soon proved prophetic. Today, tech giants are pouring huge investments into AI infrastructure, and the capital markets have enthusiastically crowned new AI kings in areas like storage and optical modules. Leopold won. His fund's portfolio value had already reached $13.7 billion (approx. RMB 90 billion) by the end of Q1 this year, and his personal wealth has skyrocketed.
And he's only 25. In the AI era, the genius narrative is simply too compelling.
Graduated from University at 19
Fired by OpenAI Two Years Ago
His youth shouldn't be too surprising, considering he started university at 15.
This German teenager from a family of doctors showed exceptional academic talent. In 2021, at age 19, Leopold earned three degrees in Mathematics, Statistics, and Economics, graduating as the valedictorian from Columbia University, and subsequently worked at two funds.
Not long after, he joined OpenAI's Superalignment team. This was a star-studded team, co-led by OpenAI co-founder Ilya Sutskever, with the goal of solving the superintelligence alignment problem within four years—essentially ensuring highly intelligent AI remains under human control.
The drama came when Leopold was very publicly fired by OpenAI.
The trigger was an internal memo written by the OpenAI board warning of insufficient safety measures at the company. This memo, however, sparked tensions between management and the board. In April 2024, OpenAI dismissed Leopold, citing information leaks.
Experiences shape choices and hone foresight.
Shortly after being fired, Leopold published a profound, lengthy essay that almost predicted the current direction of AI development and investment trends. In it, he mentioned that by 2027, large models would be capable of doing the work of AI researchers or engineers.
To achieve this, the key constraints wouldn't be at the algorithmic level, but in electricity, chip manufacturing capacity, and physical space. The power consumption of a single training cluster would leap from megawatt scale to gigawatt scale, approaching the output of a large nuclear power plant.
Based on this prediction, at the end of 2024, Leopold chose to start a business—founding the hedge fund Situational Awareness LP, dedicating himself to going long on the energy and computing power infrastructure needed for AI development, while avoiding the crowded bubbles in chips and the application layer.
Shorting Nvidia
But Buying Into the High-Flier SanDisk
Thus, Wall Street's new genius trader was born.
In May 2026, as Leopold's hedge fund disclosed its latest U.S. stock holdings (13F filing) for the first quarter of the year, the astonishing expansion of this 00s-generation manager's wealth came into view:
The total market value of his portfolio had skyrocketed from $5.52 billion at the end of 2025 to $13.7 billion. At the end of 2024, the fund's size was merely $255 million. This speed is nothing short of meteoric.
Beyond his resume and genius stories, global onlookers were more interested in what he bought.
Looking at his latest holdings, Leopold maintained his long positions in AI infrastructure while establishing new short option positions worth $8.45 billion as a hedge against tech and semiconductors. As of the end of Q1, his top five holdings were all put options. These included put options on the VanEck Semiconductor ETF valued at around $2 billion, put options on Nvidia valued at around $1.6 billion, as well as put options on Oracle, Broadcom, and Advanced Micro Devices (AMD).
This portfolio clearly signaled wariness towards overheated chip stocks. An exception was that by the end of Q1, he had exclusively increased his holdings of SanDisk by 86,000 shares and established call options on SanDisk worth $390 million. SanDisk's subsequent performance undoubtedly became the envy of many; since the start of Q2 alone, SanDisk has risen approximately 160%.
The main action was on the long side, where Leopold heavily invested in critical infrastructure assets for the AGI era.
His top long holding was the fuel cell company Bloom Energy. Leopold holds nearly 6.5 million shares of the company, with a position value of about $879 million. More precisely, Bloom Energy makes fuel cells, which can efficiently convert natural gas directly into electricity.
Simultaneously, in Q1, Leopold also increased his holdings in companies like CleanSpark, Riot Platforms, Applied Digital, and IREN—firms involved in data centers or crypto mining, possessing land, power resources, data center capabilities, or grid permits.
"The speed of AI development is determined by physical bottlenecks, so you should invest in the bottlenecks themselves." Looking across these trades, they precisely correspond to the underlying logic Leopold had when founding the fund.
Of course, for ordinary investors, copying these moves comes a bit late. Holdings reports typically have a 45-day delay, and by the time the public sees what the big players have bought, the most lucrative part of the rally has often already passed.
The End Point of the AI World
"The whole world is starting to value AI infrastructure assets."
In just the past few months of this year, the sectors Leopold bet on—power supply, data center computing power, semiconductor optics—have fully demonstrated their potential and immense demand.
Take electricity, for example. IEA data shows that in 2025, global data centers consumed 485 TWh of electricity, with AI accounting for 170 TWh (35%). It's projected that by 2030, total global data center electricity consumption will reach 950 TWh, with AI consuming 510 TWh (54%), exceeding Japan's total national electricity consumption. China's figures are equally staggering: in 2025, AI electricity consumption reached 450 billion kWh (3.8% of total societal electricity use), and it's expected to reach 600 billion kWh (5%) in 2026, nearly matching the annual electricity consumption of China's entire steel industry.
Then look at "light." As AI competition rapidly evolves from a battle of computing power to one of connectivity, traditional copper wire connections are long overburdened, leading to explosive growth in demand for "optical links."
According to data from UK-based commodity research firm CRU, global data center fiber optic cable usage reached 69.6 million core kilometers in 2025 and is expected to exceed 100 million core kilometers in 2026. Their estimates suggest that by 2027, AI-driven fiber demand will account for 35% of total data center fiber demand. For optical modules, Goldman Sachs has significantly raised its 2026 sales forecast for 800G optical modules from an initial 25 million units to 33.5 million units, a 58% increase.
Unsurprisingly, tech giants have already begun building their AI infrastructure moats.
In 2026, capital expenditure plans from Amazon, Alphabet (Google's parent), Meta, and others have grown substantially, with massive funds directed towards building new data centers and a long list of equipment including AI chips, network cables, and backup generators. Domestically, the latest development is rumors that ByteDance is discussing a spending plan of up to $70 billion (approx. RMB 474.7 billion) this year, primarily for building data centers and other AI infrastructure.
This wave of AI infrastructure frenzy is even more directly and violently reflected in the capital markets.
Looking domestically, the "Yi-Zhong-Tian" trio of optical module giants have seen their stock prices double since last year, especially InnoLight Technology, which soared from a low of 66 yuan per share in April last year to over 1,000 yuan, with a market cap now nearing 1.3 trillion yuan.
Similarly, since the memory cycle rally began in the second half of last year, stocks like Longsys, Deming Li, and Biwin Storage have surged in sync. Dapu Micro, which just went public in April, multiplied its IPO price nearly 20-fold in just one month.
In contrast, one can't help but reflect: the internet was once imagined as a weightless world, yet it gave rise to server rooms, fiber optic cables, and undersea cables. Now AI might seem even lighter, but when it comes to actual implementation, it equally relies on electricity, land, chips, networks, and cooling systems.
A grand future always grows from those silent assets.
This article is from the WeChat public account "Investment Community" (ID: pedaily2012), author: Feng Yuchen
