An anonymous user named Serenity on the X platform achieved a 45-fold return on investment in 2026. The 25 stocks he publicly invested in all saw gains ranging from 1 to 10 times. Among them: Swedish semiconductor company SIVE has seen its market capitalization rise more than 20 times from less than $150 million; US substrate material manufacturer AXTI, starting at $12, is approaching its target price of $150 set two years ago; British single-board computer manufacturer RPI saw a 44% surge on earnings day. Serenity, having joined X in less than a year, has over 400,000 followers and over 37,000 paid subscribers. In the financial investment field, 30,000+ subscribers is considered a top-tier account, second only to Elon Musk. In Chinese online communities, people call him "White-Haired Female Stock Goddess" because of his white-haired, anime-style avatar. Wall Street has also begun to pay close attention to him, listing his account as a daily must-read. Bloomberg and Reuters have cited his research, and hedge funds have started tracking his posts.

Who is Serenity?
Serenity's self-described resume includes being a former AI research scientist, a Nature paper author, a member of the RISC-V Foundation, and rejecting an offer to lead Nvidia's AI team when Nvidia's stock price was around $6 in 2018.
Serenity's self-described resume includes being a former AI research scientist, a Nature paper author, a member of the RISC-V Foundation, and rejecting an offer to lead Nvidia's AI team in 2018 when Nvidia's stock price was around $6.

Serenity's self-described resume includes being a former AI research scientist, a Nature paper author, a member of the RISC-V Foundation, and rejecting an offer to lead Nvidia's AI team in 2018 when Nvidia's stock price was around $6.

The Logic of Continuously Deconstructing
Serenity's investment methodology is known in the Chinese community as the Perilla Leaf Theory.Serenity's investment methodology is known in the Chinese community as the Perilla Leaf Theory.
... In top-tier sushi restaurants, diners chase after tuna belly, but what they truly cannot afford to run out of is the perilla leaf from a small farm in Izu. Without tuna, the menu can be reduced by a few items; without perilla leaves, the entire restaurant would shut down. The same applies to the AI industry chain. GPUs and large-scale models are like tuna, while InP substrates, CPO lasers, and high-purity phosphorus raw materials are like perilla leaves. This is also Serenity's core investment logic. Many people see AI companies frantically buying GPUs, but Serenity will continue to break it down: How do GPUs communicate? How is data transmitted? When tens of thousands of GPUs work simultaneously, what will be the first to break down? Their answer is copper wire interconnects. In the past, data centers mainly relied on copper wires to transmit electrical signals, but as AI clusters expand, traditional copper wire interconnect solutions will hit physical limits in terms of power consumption and bandwidth. When data centers begin running collaborative training with tens of thousands of GPUs, high-frequency electrical signals attenuate severely in the copper wires, creating a huge heat dissipation burden. Ultimately, the bottleneck is no longer GPU computing power, but the inability to transmit information. This is also the reason for the emergence of the CPO (Co-packaged Optoelectronics) approach, which packages optical devices and computing chips onto the same substrate, compressing data transmission distances from several meters or even tens of meters to the millimeter level. Following the CPO line upstream layer by layer, he identified five nodes he considered truly critical bottlenecks: nanoscale alignment elements between optical fibers and chips, the external continuous-wave laser source essential for the CPO architecture, MBE equipment for growing compound semiconductor epitaxial layers, high-purity phosphorus raw materials requiring a purity of 6N or higher, and SOI substrates for silicon photonic chips. For each node, Serenity found a corresponding globally scarce supplier. He bypassed financial reports, starting with hyperscaler capital expenditures to deduce the pace of data center expansion, then the physical boundaries of bandwidth/power consumption, and finally pinpointed the narrowest bottleneck. He devoured materials science papers, analyzed patents, outlined capacity expansion plans, tracked supplier certification cycles, and monitored export control dynamics in various countries. After completing the research draft, he would input the entire logic into multiple AI models, allowing them to specifically search for vulnerabilities, threats from alternative solutions, and potential valuation discrepancies. Only after passing this round of adversarial stress testing would he publicly release the data. The Most Representative Targets Over the past few years, Serenity's publicly discussed core targets have mostly revolved around the same main theme: the physical bottlenecks in AI infrastructure. From InP substrates and CPO laser sources to optical transceivers, silicon photonics, and edge computing hardware, all his firmly held positions point to the same question: If AI continues to expand, what will become scarce first? The most representative examples are $AXTI, $SIVE, and $AAOI. $AXTI was established because it noticed something two years in advance that almost no one paid attention to at the time. In early 2022, before ChatGPT was released, market attention was focused on GPUs, Nvidia, and AMD, on who could create a better training chip. InP substrates were an extremely obscure topic; a material supplied to a nascent niche market was monopolized by a virtually unknown small company. He used the Strait of Hormuz as an analogy: about 20% of the world's oil passes through the Strait of Hormuz; whoever controls it controls everyone else. AXTI is doing exactly that in the InP substrate field. 
$SIVE(Sivers Semiconductors) is his most firmly held position.
This Swedish company provides external continuous-wave laser sources required for CPO architecture, and is one of the scarcest upstream assets in the optoelectronic co-packaging industry chain.
This Swedish company provides external continuous-wave laser sources required for CPO architecture, and is one of the scarcest upstream assets in the optoelectronic co-packaging industry chain.
He began accumulating shares when the company's market capitalization was around $150 million, believing that AVGO or MRVL would likely acquire it for $200-300 million, thus securing a direct position in the CPO laser supply chain. $AAOI (Applied Optoelectronics) is a complete player in the optical transceiver industry chain, integrating laser design, assembly, and sales. He believed its $660 million market capitalization was severely undervalued when he bought in, and its value has since increased more than sevenfold. The Raspberry Pi (RPI) is perhaps the best example of his judgment. Raspberry Pi was seen by most as a manufacturer of inexpensive single-board computers for educational purposes, and when it listed on the London Stock Exchange, institutional interest was minimal. He noticed, however, that many AI startups were using $RPI as a physical isolation base for deploying intelligent group control systems. This real demand was disruptive for a company with a market capitalization of only about £500 million. He predicted two months in advance that the company's full-year revenue growth would be 58%, while the market consensus was 14%. The actual figure was 58%. After the earnings report was released, the stock price rose 44% in a single day and another 27% the following day. Of course, he also made mistakes in his judgments. In 2026, he made a mistake betting on the financial report of Japanese packaging equipment manufacturer TOWA, causing its stock price to plummet by over 20% that day. He posted that day: "I've learned something; I sometimes make mistakes in short-term timing. The main reason was that one-off accounting factors and early-stage costs from new customers compressed profits, but I believe the logic for the second half of the year remains intact." His latest investment target is $XFAB, the only high-volume SiC foundry in Europe and the US, which also has silicon photonics manufacturing capabilities. Its market capitalization is approximately $1.28 billion, with a price-to-book ratio of only 1.29. The EU CHIPS Act 2 is expected to be implemented soon, and Nvidia and Nokia's silicon photonics evaluations are also underway. From an external perspective, $XFAB shares a highly consistent profile with all its past holdings: small market capitalization, no coverage, stuck at the physical bottleneck of AI hardware, and awaiting institutional validation. The truly scarce resource is the physical world bottleneck. Most of Serenity's core holdings actually share the same main theme: AI data centers will ultimately rely increasingly on optical interconnects and CPO architectures. While the portfolio is diversified across different companies, the underlying logic is highly consistent. If future data center interconnection routes change, or CPO progress falls short of expectations, these companies could also be affected. On the other hand, most of these targets are small-cap companies with limited liquidity. For large institutional funds, building positions, exiting, and managing positions are inherently difficult, which is one reason why Serenity was able to identify them early. Most people discuss AI by focusing on parameter scale, model architecture, and application scenarios. Serenity discusses lasers, substrates, heat loss, and the molecular purity of phosphorus. While the market is still debating which large-scale model's benchmark looks better, he's already asking: At what scale will the copper wires hit a wall first when data center clusters expand? What is that wall physically? Where must we go to get around it? Who's there? Who controls the scarcest nodes in the physical chain of AI infrastructure? Whoever controls those nodes has the power to collect tolls from the entire industry. This logic doesn't rely on any single company's product decisions, nor on market sentiment; it relies solely on the laws of physics. From an ordinary retail investor permanently banned by the WSB to an anonymous supply chain detective with 400,000 followers who read his articles daily, Serenity has been constantly dismantling the system. From GPUs to optical modules, from optical modules to substrates, and from substrates to materials and production capacity. In this age of information overload, what's truly scarce may not be the viewpoints themselves, but rather who can see those insurmountable physical bottlenecks earlier.