Breaking: RTX Pro 6000 Payback Time and the Math Behind Anthropic's IPO

An initial public offering is a process where a private company sells shares to the public for the first time, allowing it to raise capital and giving investors a stake in its future. Anthropic, the developer of the Claude large language models, is preparing for such an event that it projects at a valuation between 2.5 trillion dollars and 3 trillion dollars. This figure stands in stark contrast to the company's current financial reality, which includes a deficit of about 500 billion dollars and annual revenue of less than 20 billion dollars. The gap between this projected value and its actual earnings forms the core of the debate over how a company losing so much money can be worth so much.

The answer lies in the economics of the hardware required to run the models. Nvidia, a major supplier of the graphics processing units that power these systems, sells the RTX Pro 6000 card at a retail price of 20,000 dollars. This price tag does not include the servers, switches, central processing units, data center facility costs, or electricity needed to keep the machines running. However, when purchased directly from Nvidia at a wholesale price of 3,000 to 4,000 dollars, the math for a single card becomes more tangible for potential investors and operators.

A single RTX Pro 6000 card can serve approximately 15 to 20 light users who consume about 60 to 70 tokens per second. Alternatively, it can support 1 to 2 heavy coding users running Claude Code at a rate of 40 to 50 tokens per second, assuming constant use for 24 hours a day. Heavy users on the maximum plan pay 200 dollars a month, while light users pay an average of 20 dollars a month for a base plan. These figures are the building blocks of the revenue model that justifies the company's massive valuation.

For a heavy user paying 200 dollars a month, the GPU alone takes about 10 months to pay back its wholesale cost of 3,000 to 4,000 dollars. The same 10 month payback period applies to the group of 15 to 20 light users paying 20 dollars a month. While most light users do not use the chat for more than a few hours a day, which could stretch the bandwidth capacity by a factor of 10, the acquisition cost and breakeven point remain substantial. This bottleneck in compute and GPU supply is a central challenge for the industry, and Nvidia is well aware of its leverage in this market.

Nvidia drives demand up and lowers supply, prioritizing its allocation to large players like Anthropic, Grok, and OpenAI, leaving the consumer market to fit the bill. This dynamic was highlighted when Nvidia agreed to acquire Hugging Face, the platform where most open AI models are distributed, for 12.9 billion dollars. Hugging Face generates about 150 to 200 million dollars a year in its peak years, making the acquisition price about 70 times its revenue. Although the stated reason for the deal was control over CUDA, the software that runs these models, the deeper motive was control over model distribution, particularly in the open source sector.

The open source market had been flooded with advanced Chinese models from companies like Moonshot AI, Alibaba, and Xiaomi. These models, including Kimi K3, began to outperform Anthropic's Opus 5 models and were made available to the public. After the acquisition, advanced models were deleted from the Hugging Face platform, with anything over a quant 4 removed. This action led developers to rally on Discord and create Pirate Face, an alternative site for uploading and downloading models, after large 400 gigabyte models were scrubbed from the network.

In response to the competition, Anthropic demanded that Congress ban Chinese models on national security grounds. The effort gained momentum after a publicity stunt involving a disgruntled employee, Jacob Coxon, who quit amidst disappointment in AI, while Anthropic circulated media suggesting AI would take over the world. The panic worked, and Congress passed a bill banning any government and its affiliates from using Chinese models. Anthropic is now attempting to pass additional bills to stop other companies from launching their own in-house AI models using these advanced open source tools.

The regulatory climate also affected Anthropic's product releases in the first half of the year. Fable 5 and Mythos 5 were released on June 9, 2026. However, the U.S. Department of Commerce issued an export control directive that forced Anthropic to suspend access to these models worldwide on June 12, 2026. The models were solving too many problems, leading Anthropic to pull them off the shelf. This move signaled that the company was focused on controlling the AI race rather than simply winning it.

In recent months, Anthropic has been suspending users for reasons such as abuse, potential fraud, and usage violations. The appeal process is largely ineffective because it is fully run by bots, and accounts are rarely reactivated once suspended. Once Anthropic goes public, these suspensions are expected to multiply significantly because the company will have deep pockets and access to unlimited funds. The end user may become less important to the organization as it scales.

Anthropic does not currently make money off of daily or heavy users in a direct manner. Instead, the company is likely to create programs where user ideas and data belong to Anthropic. If the company identifies valuable ideas, it may suspend users for unproven reasons and claim intellectual property rights to the work. Dario Amodei, the CEO of Anthropic, stated in an interview that some ideas are worth billions and that Anthropic will likely implement a royalty or partnership program. Under this plan, if a person or company finds a major breakthrough, such as a cure for cancer, some of the accreditation and funding will go to Anthropic. This structure suggests that the company hosting your work may claim a share of the prize when your work becomes worth billions.