CEO Column | It’s Not That PLTR Has a Problem — the Valuation Method of the Entire AI Investing World Has Begun to Distort
On the surface, this fall in PLTR looks like a pullback in a heavily run-up AI-concept stock. But if you truly follow international capital markets, technology roadmaps and enterprise purchasing behavior, you know it isn’t that simple. Palantir fell from its November 2025 high of US$207.52 to a close of US$128.06 on April 10, 2026 — a drop of about 38%. And even when the market rebounded, it was still sold off against the trend. That kind of move is usually not purely technical; it is money beginning to re-assess one thing: whether the profit logic of future AI is still yesterday’s logic.
Many people read this as a provocative line: those selling platforms can’t beat those selling workers. That line is seven parts true and three parts dangerous. Anthropic’s growth is indeed fast enough to rewrite market perception. Reuters’ latest report notes that Anthropic said in April 2026 that its annualized revenue has exceeded US$30 billion, up from only around US$1 billion at the start of the year — an explosive rise in a short time, driven not by ordinary chatbots but by enterprise tools like Claude Code with high token consumption and high task density. This means companies no longer just want to “adopt AI”; they want to buy AI that can work directly. But the other side must also be made clear: if you therefore say Palantir has been eliminated, that is too amateurish. Palantir’s latest results show Q4 2025 revenue up 70% year over year, US commercial revenue up 137% year over year, and the company’s guidance for full-year 2026 revenue growth is still about 61%. This is not a company in collapse; it is a company with very strong fundamentals. Its real core is not just model access but tying together data, workflows, decisions, permissions and organizational governance — especially in high-trust settings like government, defense and critical infrastructure. So PLTR’s problem now is not “whether there is business,” but whether the market is still willing to look at it through the mythical valuation it once enjoyed. That is the point.
Palantir’s price-to-sales ratio is still around 68x. That valuation does not reflect an ordinary software company, nor a steady defense-tech supplier; it reflects the market’s former belief that it could capture most of the dividends of the future AI infrastructure platform. But once the market starts to doubt — will companies buy fewer platforms and more agents in the future? Will they shift from “large system builds” to “pay per task”? Will budgets move from heavy-platform CAPEX/OPEX to buying outcomes directly? — the valuation model loosens.
This is also what I have been saying: the speed of technological change in the AI era is already too fast to be viewed through the traditional linear valuation of five or ten years.
In the past, when we looked at a company we would ask where it would be in five years, and whether its moat would still exist in ten. That logic mostly held in the industrial era, the internet era and even the SaaS era, because although products iterated fast, business models, infrastructure and customer habits did not flip the table within three months. Today is different: a model’s advantage can be caught up within three months; a platform narrative that once looked beautiful can have a large chunk eaten by agent products in half a year; a company that used to charge high prices for integration capability may suddenly find that customers are no longer asking “can you do more” but “why shouldn’t I just buy the outcome?” This is not theory; it is what is happening on the industry floor right now. Anthropic’s explosion, OpenAI’s re-narrowing toward the enterprise route, and even the model-supply-chain replacement issue involved in Palantir’s own Maven system all show that AI is not a steady long-distance race but a battlefield of high-frequency reorganization.
From PSF’s experience in recent years with international AI projects, compute, enterprise adoption and sovereign-level applications, I would give investors an unpopular but very real reminder: you cannot look only at technological leadership, and you cannot look only at rapid revenue growth; you also have to look at whether that technological leadership can survive the next round of architecture switching.
Many people investing in AI are actually not looking at the company but at sentiment. As long as the market believes a company is riding the wave, it is willing to give an outrageous multiple. But the problem is that when the underlying rules of an industry are still changing, giving too full a multiple is itself a risk. Technological leadership does not equal a secure business position; a secure business position does not equal a reasonable valuation; and a reasonable valuation does not mean the market will not first knock you down in the short term. These three things are often conflated.
Palantir is a classic case. It is not without strength, nor without a moat. On the contrary, its position in the US government and defense sector has become more secure as Maven has been folded into a more central system. Reuters reports that Palantir’s Maven AI system has been advanced by the Pentagon into a more core long-term system, with more stable funding going forward. This kind of moat cannot be taken by ordinary model companies in the short term. But the market now fears something else: government business is stable, yet may not support the imagination built into a market capitalization of more than US$300 billion. What truly decides the premium is still whether the commercial side can keep growing at high speed. So PLTR was sold off not only because of short sellers, nor because the fundamentals suddenly deteriorated; it was sold off because everyone began to ask the same question again: if the form in which AI’s value is delivered shifts from “platform” to “agent,” who deserves the highest valuation? There is no standard answer today, but as long as the answer is not settled, the market will not be too lenient. Put more bluntly, in this era the most dangerous thing in investing is not misjudging a company but measuring a new world that is changing shape with an old-era ruler.
In the past we could assume that as long as a company proved it could build products, create cash flow and expand its market, it had a chance to gradually deliver on a five- or ten-year story. Now it is different. The market does not wait five years for you; every quarter it asks: has your model been replaced? Has your interface been bypassed? Has your customers’ procurement logic changed? Could the most valuable segment of your value chain suddenly be given away free by someone else? This is the cruelest part of AI investing: it’s not that you are bad, it’s that the world changes too fast. So my view is simple: what should really be discounted in the AI era is not only cash flow but “the credibility of the future narrative.” The future can still be looked at, but we can no longer take it for granted to look five or ten years ahead and extrapolate optimism in a straight line, because in this era things can reverse in two or three months: yesterday’s king may today be just a link in the supply chain; yesterday’s sexiest business model may tomorrow be penetrated by a lighter, cheaper, more direct product route. Therefore, for highly valued AI stocks, I would in fact be more conservative. Not because I don’t believe in AI — quite the opposite: because I believe too much in AI’s destructive power. I know it will not only replace traditional industries; it will destroy even the valuation logic of the previous generation of AI companies. That is what is truly to be feared.
Conclusion
This fall in PLTR looks like a single-stock event on the surface but is in fact a stress test of the entire AI capital market. It reminds all investors: when the pace of technological change reaches the quarterly level, when enterprise procurement shifts from buying systems to buying outcomes, and when market sentiment is willing to push any “AI story” to outrageous highs, the most dangerous companies are often not those with poor fundamentals but those with very good fundamentals that the market has imagined too perfectly. What investment markets fear most is never bad news; it is that at the most optimistic moments everyone treats risk as growth and valuation as faith. And the real homework of this round of AI is not learning to be more excited, but learning to stay calm when the whole world is excited.
⸻CEO Yang
