Author: Silicon Valley Wang Chuan, investguru
1/ Human anxiety is often an amygdala-driven stress response. To alleviate anxiety, simply offering platitudes like "Don't be anxious, it's useless" is ineffective, as it fails to address the underlying mechanisms triggering the stress response. The truly effective method is to have the anxious person slowly write down the specific things causing anxiety and their reasons on paper. The act of writing transfers control from the amygdala, which only knows tension and alertness, to the rational prefrontal cortex. The more detailed the writing and the deeper the analysis, the more complete this transfer of control becomes, naturally reducing anxiety. Even if you can't finish in one sitting, just starting to write and having a rough idea of how long it might take to logically sort things out can begin to lower anxiety. Relying solely on feel-good "inspirational soup" without forcing oneself to slowly write it down prevents this genuine transfer of control and thus fails to fundamentally change anxiety.
2/ One anxiety-inducing event in May 2026 was the comprehensive surge in semiconductor industry stocks. The famous Philadelphia Semiconductor Index was up 150% compared to a year prior. Various semiconductor storage company stocks rose even more sharply, with SNDK closing at $1,562 on May 8th—38 times its price from a year ago. While the author holds no investments in semiconductor companies, this article attempts to deconstruct the underlying logic behind these stock rallies to share and discuss with readers.
3/ This story begins with the rapidly climbing valuations of various AI startups. Many speculative investors on the periphery, who only care about stock price fluctuations, lack some fundamental knowledge:
- First, the vast majority of AI startups incur heavy losses, but many people completely confuse the two basic concepts of revenue and profit (some even intentionally), and this misunderstanding persists and spreads.
- Second, when a company says its ARR (Annual Recurring Revenue) has reached $100 million, it does not mean its total revenue over the past year was $100 million. At most, it implies its revenue in the most recent month was about $8.33 million. If you misinterpret this out of wishful thinking, that's your own problem.
- Third, what a company does *not* say often reveals more than what it does say. Reporting ARR without mentioning profitability implies it is incurring heavy losses. If it said six months ago its ARR reached $100 million but has been silent since, it most likely means its revenue has shrunk and losses have worsened.
4/ Taking the currently red-hot large model company Anthropic (referred to as Anth below) as an example, some media disclosures at the end of April indicated the company's ARR was approaching $40 billion. In other words, the company's single-month revenue in April was close to $3.3 billion ($40B / 12). In March, the company's ARR was below $30 billion, meaning March revenue was below $2.5 billion. Anth's CFO, in a public document in March, disclosed cumulative revenue exceeding $5 billion from its founding in 2021 to March 2026. In other words, Anth's cumulative revenue (emphasizing again, not profit) over the past five years since founding is approximately $10.8 billion.
5/ There is also an issue regarding the definition of revenue here. A significant portion of Anth's revenue comes from Google Cloud Services and AWS. Cloud service providers take at least a 20% cut of the money paid by customers, with the remainder going to Anth. Therefore, the ARR Anth publicly reports needs an additional 20% discount. But we won't dwell on that here because Anth is still losing a lot of money every month.
6/ How much is the loss? Asking several different AIs and based on limited public information, assuming its single-month revenue reached $3.3 billion, guesses for the single-month loss range from $1.1 billion to $1.7 billion. This is also the fundamental reason why Anth needs to continuously raise external funding. As of the end of April, Anth has raised a cumulative $72.3 billion over five years, but cumulative revenue is only $10.8 billion, and it still needs to continue fundraising. Those who only glance at headlines mistakenly believe Anth is already making a fortune but don't understand the immense pressure on management to constantly seek new external capital just to survive.
7/ Anth is not without competition. Its president recently admitted in a public speech that AI models from other US companies are only one to three months behind Anth, while Chinese companies' AI models lag six to twelve months behind. The features and strengths of each company's products ebb and flow, changing very rapidly, making it difficult to judge the ultimate winner based on one or two years of performance. For example, Cursor was the most popular programming assistant tool among AI developers from 2023 to early 2025, and the company was once highly sought-after. But Anth's launch of Claude Code in the first half of 2025, which could automatically generate large amounts of code, allowing many non-programmers to participate in programming, has been the core driver of Anth's sudden rise and rapid growth in revenue and valuation over the past year. Recently, some programmers have complained that Claude Code's experience and cost-performance in some areas are inferior to OpenAI's Codex, leading them to switch to the latter. For many users, the switching cost is not high. Besides OpenAI, SpaceX's X.AI also began close collaboration with Cursor last month to develop a competitor to Claude Code.
8/ Investors bullish on Anth might cite specific scenarios, claiming Claude Code achieves profit margins as high as 70% for inference work. However, this argument first ignores the enormous ongoing cost Anth incurs to continuously train new models, and second assumes competitors are completely absent with no pricing pressure. This clearly does not align with reality. When Anth raised funds in February, its valuation was $380 billion. Just three months later, it began seeking new funding with an asking valuation of over $900 billion. Recently, in a secondary market for private equity, a valuation of reportedly $1.2 trillion even appeared (some podcasts even mentioned a long-term valuation of $5 trillion). This is a company with only a five-year history, cumulative revenue (not profit!) of $10.5 billion, yet still losing over a billion dollars per month. Simply because of high growth over the past year, some believe it can last forever. Raising $72, losing money for five years to do $10.5 worth of business, yet being able to cash out at a market valuation of $1 trillion—such a wonderful thing, no wonder so many entrepreneurs aspire to it! 😊
9/ Berkshire Hathaway (referred to as BRK below), under Warren Buffett, had revenue of $370 billion in 2025, operating cash flow of $45.9 billion, after-tax profit of $66.9 billion, nearly $400 billion in cash on hand, and a market capitalization of only about $1 trillion. Looking at it another way, BRK's revenue in ten days equals Anth's total revenue throughout its entire history. Yet some investors are willing to believe that Anth, which still highly depends on external investment to survive, deserves a higher valuation than BRK. The immense courage required to construct such a belief is truly awe-inspiring.
10/ The valuation growth of Anth and OpenAI is crucial for the explosive increase in capital expenditure (capex) across the entire AI industry and the subsequent prosperity of the storage sector. Looking back at the developments over the past few years, this logical chain is clear. For the rest of the story, please listen to the next installment.
