Anthropic, Posting Over 7 Trillion Won in Monthly Revenue, Nears IPO: Should You Become a Shareholder?

Following SpaceX, buzz is now building around an IPO for Anthropic. A massive amount of capital flooded into SpaceX's recent public debut, and talk is circulating that Anthropic could smash even that record. What is driving such confidence? Today on OhGraph, we dive into the story behind Anthropic's potential IPO. Through various data and charts, let's look at the strengths fueling its listing push, as well as the critical vulnerabilities worth watching closely.


Surpassing SpaceX? Is an Earth-Shattering IPO on the Horizon?
On June 1, Anthropic took its first step toward going public by submitting a confidential draft filing to the U.S. Securities and Exchange Commission (SEC). Recently, a string of reports have emerged suggesting an IPO is imminent. The New York Times reported that the investment banks underwriting Anthropic's public offering recently held preliminary meetings with prospective investors.

Enlarge image

Listing on the U.S. stock market largely requires navigating three major phases. First, a company must select investment banks to lead the listing process and review its financial condition. Next is submitting a registration statement to the SEC. The filing known as Form S-1 outlines the company's business model and financial statements. Once submitted, it is eventually made public for scrutiny. Finally, the offering price is determined, and once listed, shares begin trading publicly.

Anthropic has submitted its registration filing, but without making it public yet, it is currently meeting with prospective investors. Word has emerged that during these preliminary pitches, Anthropic floated a fundraising target of around $100 billion through the IPO. Let's look at the chart to see just how extraordinary this figure is.

Enlarge image

This chart from Bloomberg displays data on the world's top 100 public companies. The X-axis represents the listing date, while the Y-axis shows the amount of capital raised at IPO. The size of each circle reflects the company's market capitalization. This past June, SpaceX raised $85.7 billion and saw its market cap surge past $2.1 trillion on its first day of trading.

Plotting Anthropic's target against this landscape makes the comparison clear: a $100 billion fundraising target would instantly eclipse SpaceX's record. If the offering proceeds as targeted, Anthropic's valuation is projected to reach $2 trillion, matching SpaceX.

Enlarge image

In fact, Anthropic has consistently secured staggering funding sums throughout its financing rounds. Funding rounds refer to the multi-stage equity investment process startups undergo as they grow, moving from initial seed capital to Series A, Series B, Series C, and beyond.

Looking at coverage of these rounds, each successive stage also reveals the company's valuation, as investors state what value they place on the firm when investing. Anthropic has massively expanded both its fundraising scale and corporate valuation, especially through rounds held since 2025.

Enlarge image

In its Series E round in March of last year, Anthropic raised $3.5 billion. It then secured $13 billion in Series F, $30 billion in Series G, and an eye-popping $65 billion in Series H this past May. What is even more striking is how much faster Anthropic's valuation grew: from $61.5 billion during Series E to $965 billion in Series H, multiplying more than 15 times in just 15 months.

Emboldened by these metrics, Anthropic is reportedly preparing to project its total addressable market (TAM) at more than $30 trillion. Converted to Korean currency, $30 trillion exceeds 40 quadrillion won. Ahead of an IPO, tech startups frequently calculate and pitch their TAM to investors, representing the theoretical annual revenue a company could generate if it captured 100% of its market. Anthropic's calculation treats the AI domain not merely as a software market, but as a technology capable of handling most human work.

Enlarge image

This figure exceeds the $28.5 trillion TAM previously claimed by SpaceX. Debate continues over whether a $30 trillion market size is realistic. In the community section of major U.S. tech outlet The Verge, some sarcastic commentary quipped that Anthropic was being even more delusional than Elon Musk.

Yet there is a clear rationale behind Anthropic's aggressive market projection: its recent earnings growth has been extraordinarily steep. Let's explore that aspect in more detail.


Generating Over 7 Trillion Won a Month... Backing It Up with Revenue
Anthropic's greatest source of confidence lies in its current revenue and operating performance. The recent numbers disclosed to key investors are remarkably strong. Here is the third chart from OhGraph.

Enlarge image

This chart tracks annualized run-rate revenue, calculated by annualizing monthly earnings. While the figure stood at around $9 billion at the start of the year, it surged more than fivefold to $47 billion in May, overtaking OpenAI. Looking at July, it climbed by nearly another $20 billion to reach $65 billion. An annualized run rate of $65 billion means Anthropic pulled in $5.4 billion in a single month—more than 7.6 trillion won. For perspective, South Korea's Ministry of Culture, Sports and Tourism has an annual budget of 7.8555 trillion won this year, meaning Anthropic generated an equivalent sum in just one month.

How can Anthropic make this much money? The answer lies in its firm grip on token-heavy corporate clients and software developers rather than general consumers. Anthropic has pursued a strategy of securing enterprise customers through the technological capabilities of Claude Code. In February of this year, around 500 enterprises spent more than $1 million annually with Anthropic; by April, that figure doubled to 1,000. Real-world data clearly demonstrates how successfully Anthropic is capturing enterprise clients. Let's look at this chart tracking spending by U.S. enterprises on services and tokens from leading AI firms.

Enlarge image

This shows the trajectory from 2023 to the present. In the beginning, OpenAI held an overwhelming share. This year, however, Anthropic's rise has been dominant. Anthropic now ranks first among all AI companies with a 43.5% share, while former front-runner OpenAI has slid to 39.7%.

Sustaining this momentum naturally requires deep technological capabilities. That is why Anthropic remains fully committed to talent acquisition and technical investment. Among these moves, the recruitment of Andrej Karpathy stands out as particularly noteworthy.

Who holds the greatest influence in the AI tech market today? Sam Altman? Dario Amodei? Elon Musk? Demis Hassabis? While these tech leaders certainly wield enormous influence, Andrej Karpathy arguably ranks first when it comes to communicating, explaining, and evangelizing AI technology to developers and the broader public.

Enlarge image

Karpathy laid the groundwork for early models as a founding research scientist at OpenAI. He later joined Tesla at Elon Musk's request, overseeing the Autopilot program and spearheading autonomous driving development. Beyond his technical prowess, what makes him stand out is his communication ability: he possesses an exceptional gift for translating research findings and real-world large-scale system deployments into concepts accessible to other developers.

The phrase "vibe coding" originated from Karpathy. The concept of an "LLM Wiki"—archiving personal knowledge for AI models to draw upon—became a trend among developers after Karpathy mentioned it. The term "AutoResearch," describing AI agents running autonomously through the night without human intervention, followed the same path. In short, Anthropic recruited a figure who wields immense sway in shaping the culture and lexicon of AI developers today.

Enlarge image

Since joining Anthropic, Karpathy has been conducting research aimed at applying his concept of AutoResearch to model improvement. Instead of humans fine-tuning, adjusting, and refining large language models step by step, AI handles the process autonomously. Leading AI models form hypotheses, write code to test them, conduct experiments, and analyze results to build superior AI models—a mechanism known as recursive self-improvement. Andrej Karpathy is driving research at Anthropic to bring this mechanism to fruition.

Enlarge image

Another key hire commands attention: Amir Salek, formerly of Google. Salek was an essential architect of Google's proprietary AI chip program, the TPU, leading development from the first generation through TPU v7. Industry observers believe Salek's arrival could reshape the trajectory of Anthropic's custom silicon development.

Among major frontier AI rivals, Anthropic was notably absent from the in-house chip race. Google has aggressively pushed model development using its TPUs, while OpenAI's custom-designed chip "Jalapeño" recently demonstrated superior speed and efficiency over Nvidia chips in inference workloads. Bringing Amir Salek on board allows Anthropic to address this long-standing vulnerability.


Hurdles Ahead for Anthropic: Data Centers Less Welcomed Than Coal Plants
With surging revenue and plugged gaps, does investing right after the IPO guarantee instant wealth? That conclusion is premature. While Anthropic boasts stunning numbers today, assuming an effortless path forward would be naive. Currently, Anthropic's biggest dilemma is that its most advanced model, developed at immense expense and effort, simply is not selling well.

Enlarge image

This chart illustrates the breakdown of Anthropic's usage by model following the release of its flagship model, Fable 5, on June 10. Fable 5 accounted for just 6% of purchased tokens and 11.4% of total customer spending. Because Fable 5 token pricing is so high, its share of spending appears somewhat elevated, but actual adoption remains marginal.

Until now, Anthropic's business model operated on a familiar cycle: invest massive sums to develop more powerful models. When a new model launched, customers rushed in to use the expensive, high-performing flagship. Reaping profits from that top-tier model allowed Anthropic to recoup its R&D investments and fund the creation of even larger models.

With Fable 5, however, the premise that "if capability improves, customers will follow" broke down. It proved too costly. Moreover, users found that previous models could comfortably handle most workloads, removing any compelling incentive to upgrade. Unless exceptional capabilities are required, cutting-edge frontier models may no longer feel necessary. Users can achieve ample productivity with cheaper older models or substantially less expensive Chinese alternatives.

If this dynamic persists, Anthropic's investors will inevitably grow concerned. Compute costs will continue to escalate while revenue from flagship model sales stagnates. What happens if open models close the gap with frontier models even faster? Adoption of cheap, highly efficient open-weight models is already becoming the dominant trend.

Enlarge image

At the end of June, open-weight models represented only 28.4% of total token volume, but by August, that share jumped to 62.0%. Some observers suggest that if this trend continues, the business logic behind frontier model development could fundamentally shift. Rather than generating profits from multi-billion-dollar top-tier models, flagship systems may increasingly serve as promotional showcases to display technological prowess.

Enlarge image

In addition, growing public resistance to AI and AI data centers presents another major concern. Anthropic is reportedly planning to list public backlash as a key business risk factor in its IPO prospectus. Some observers argue that this negative sentiment was fueled in part by Anthropic's own CEO, Dario Amodei, given his persistent warnings regarding AI risks.

Negative perceptions of AI in the United States are increasingly entering the mainstream. Communities are showing greater reluctance toward AI data centers than toward nuclear or coal-fired power plants. In fact, survey responses favoring coal-fired power plants over data centers were double those supporting data centers.

Enlarge image

Why would residents favor facilities that consume more water and cause greater environmental pollution? The reasoning is that a coal plant at least creates local jobs. In contrast, data centers offer minimal employment benefits, and as the tech industry constantly proclaims, AI threatens to displace human jobs—leaving locals little reason to welcome them.

Electricity bills pose an additional problem. Because data centers consume tremendous amounts of power, regional grids must construct extra transmission infrastructure. Under U.S. regulatory frameworks, infrastructure expansion costs are shared across all local utility subscribers. Residents end up paying higher electric bills for power they never used. As backlash mounts, tech companies are scrambling for solutions; Microsoft pledged earlier this year not to pass additional data-center-driven electricity costs onto local residents.

Political circles are also shifting their stance rapidly. Both major parties previously dismissed opposition to data centers as the rhetoric of radical outliers. Yet as public discontent swells, politicians on both sides of the aisle are competing to propose stricter regulations and development caps.

Judged purely by its financial metrics, Anthropic's IPO looks set to be a blockbuster debut. Yet the paradox of its most powerful model failing to become its biggest seller, combined with mounting public pushback against AI infrastructure, leaves Anthropic facing substantial challenges. Will Anthropic's public debut lead down a path of triumph or one of tribulation? That concludes today's edition of OhGraph. Thank you very much for watching this video to the end.

References
- Anthropic | YouTube
- Andrej Karpathy(@karpathy) | X
- Anthropic confidentially submits draft registration statement for proposed initial public offering | Anthropic
- August 2026 Ramp AI Index: Cracks in the AI Thesis | Ramp Economics Lab
- AI Model Leaderboards: Open vs. Closed Token Volume | Vercel AI Gateway
- Americans' Openness to Data Centers Has Kept Declining | Heatmap Pro·Embold Research
- Institutions of Democracy Survey: Opposition to Local Data Centers | University of Pennsylvania Annenberg Public Policy Center
- Anthropic Bankers Pitch $100 Billion I.P.O. at Roughly $2 Trillion Valuation | The New York Times
- Jalapeño's first results show industry-leading speed and efficiency in AI inference | OpenAI

Text: An Hyemin Design: Ahn Jun-seok Intern: Shin Yeon-seong
※ Please note: This article was translated by AI and may contain errors.