Anthropic’s $200 Billion Compute Bet: Why Wall Street Isn’t Betting on Revenue, They’re Betting on Silicon Chokeholds
(SeaPRwire) – Let me cut straight to what every analyst note buried under those Buy ratings.
Two firms just covered Anthropic at $83 per share. Both say the stock doubles from here. Neither one is actually betting you will pay more for Claude. They are betting that someone out there is going to figure out how to turn a $35 billion revenue stream into a $10 billion profit machine, even if the math right now looks impossible.
Here is the raw number no one is framing correctly. Trailing revenue sits at $35.3 billion. Trailing loss per share is negative $59.74. Gross margin this year was 22 percent. Needham wants you to believe gross margins will hit 90 percent and operating margins will clear 35 percent. Melius says 30 percent operating margins are on track by 2029 or 2030. That gap between twenty-two percent and ninety percent is where every dollar of the $150 to $155 price targets lives.
I had lunch with a procurement lead last week who works at a mid-size cloud integrator. He told me something blunt about GPU allocation negotiations. He said large model companies are not competing on talent anymore. They are competing on compute booking rights. The person who locks in early access to H100 clusters at favorable rates wins the next two years of product delivery. Everyone else figures out whose leftovers they can use.
That is why Anthropic’s S-1 numbers on silicon choices matter more than anything else in these analyst reports. They are running workloads across Nvidia chips, AMD processors, Trainium instances, and TPUs. Few companies can say that honestly. It means they are not trapped behind a single vendor’s allocation decisions. It also means they have the engineering depth to migrate inference across architectures without breaking the product experience.
Melius calls this a flexible compute pool. I call it optionality under extreme scarcity. Baird flagged $200 billion in commitments tied to Nvidia-based compute. That is extraordinary capital already committed toward future capacity. New Street Research labeled this a generational IPO. Maybe it is. But look closer at what generated the money. Anthropic sold 980 million shares at $83, raising $81.34 billion, up from an original 825 million planned shares. Existing shareholders sold into this too. Not all that cash belongs to the operating company buying the chips.
Needham’s Scott Berg frames the addressable opportunity as ten percent of a fifty-six trillion dollar labor market. That sounds impressive until you remember that the fifty-six trillion figure covers global payroll, not AI displacement economics. Shifting even one percent of knowledge-worker hours into Claude-compatible workflows requires infrastructure that does not exist at scale today. Building it requires exactly the kind of compute flexibility Anthropic is describing. That is the real bet.
Revenue grew from forty-one million dollars in twentytwentythree to three hundred eighty-six million in twentytwentyfour. Fiftyeight billion in the first half of twentytwentyfix alone. Second quarter revenue jumped thirteen hundred sixtyeight percent yearoveryear. Growth is there. Margin transformation is the question nobody has answered publicly yet.
R&D as a percentage of revenue is shrinking. Cost per token is falling. Multi-product customers are increasing. All true. All directional. None of them guarantee profitability at the pace investors are demanding. You can expand margins by pricing power, yes. You can also expand them by forcing your way through a supply chain that still treats advanced silicon like it might run out tomorrow.
Here is what actually happens next. The companies that lock compute agreements early either achieve margin inflection inside the window those deals provide, or they bleed capital trying to fill the gap. Anthropic has the flexibility argument. Flexibility helps during shortages. It does not create capacity where none exists. The question is whether $200 billion in Nvidia commitments translates into deployed TFLOPS fast enough to support thirty-five percent operating margins within three years, or whether those commitments become balance sheet anchors during a hardware correction cycle.
I would buy the stock if I thought margin expansion was inevitable. I am not sure it is. What I am sure about is that whoever controls silicon allocation right now controls pricing power for the next generation of enterprise AI contracts. Anthropic just proved they can navigate multiple chip ecosystems. That alone makes this one of the most interesting valuations in technology right now.