On one side, the income expectations of leading companies are quietly "cooling down," while on the other side, the valuations of infrastructure startups have nearly doubled in less than a year. Previously, Axios cited sources claiming that OpenAI's annual recurring revenue had approached $70 billion, a figure once viewed by the market as an optimistic anchor point for the acceleration of AI commercialization; however, according to the Financial Times, which cited internal financial documents shared with investors, OpenAI’s revenue by the end of September was only close to $50 billion, and multiple Chinese media outlets interpreted this as a reduction of approximately $20 billion in actual levels compared to earlier signals. Concurrently, there is another capital story: Bloomberg reported that Arena, an AI model evaluation platform, was valued at about $1.7 billion in January this year, and in the latest round of financing led by Lightspeed Venture Partners and Khosla Ventures, with participation from Salesforce Ventures and Dell Technologies Capital, it raised $200 million, boosting its valuation to approximately $3.1 billion, nearly doubling in about nine months. When these two sets of figures, both derived from media sources, are juxtaposed, a clear contradictory picture forms: as the revenue reality represented by OpenAI begins to recalibrate the market's bullish advance, capital continues to escalate prices in infrastructure and tools; concerns over an AI bubble and bets on the essential demand for underlying computational power and evaluation capabilities are unfolding on the same balance sheet.
OpenAI Revenue Signal Correction: $20 Billion Optimism Gap
The previously widely quoted "$70 billion" comes from Axios' sources, referring to "annual recurring revenue." This concept in financial language usually represents sustainable revenue streams from subscriptions, long-term contracts, and so forth, thus resembling a growth trajectory pointing to the future. Subsequently, the Financial Times obtained internal documents shared with investors, stating, “as of the end of September, the annual revenue was close to $50 billion,” shifting the language from "recurring" back to the more general "revenue" dimension. When these two sets of numbers are placed side by side, Chinese media quickly seized upon this difference of about $20 billion, contrasting the previously viewed $70 billion commercialization progress anchor with the $50 billion that is closer to the reality of operational status, constructing a narrative line of "optimistic expectations being corrected."
The issue is that this $20 billion gap is more like a mirror reflecting not the details of OpenAI's financial statements, but rather the market's systematic overestimation of the commercialization speed of large models. Research briefs have already pointed out the key uncertainties: currently, no one can confirm what these "annual revenues" actually cover, whether different businesses such as API calls, ChatGPT subscriptions, and enterprise licensing are all included, as media reports do not provide a breakdown. In this information structure, the market selectively embraced the more optimistic $70 billion number, taking it as a signal that "the leader has successfully commercialized the model," and only upon the release of the $50 billion figure did it passively acknowledge that the growth pace might not be as drastically accelerating as imagined. More importantly, neither of these two sets of figures came from official financial reports or statements but rather from internal documents and hearsay; without a breakdown of items and multi-source cross-validation, the so-called "$20 billion optimism gap" can only be regarded as a reminder of an emotional bubble, rather than a solid data foundation for reconstructing industry profit models.
Revenue Not as Desired, AI Bubble Theory Again in the Spotlight
When the Financial Times presented internal documents stating "as of the end of September, annual revenue was close to $50 billion," what was first ripped apart was not OpenAI's trade secrets but rather the overly optimistic narrative that had been repeatedly amplified over the past year within social networks and capital markets. Previously, Axios' cited sources had tossed out the "annual recurring revenue close to $70 billion" as an industry consensus. The $70 billion had become the foundation for all stories about the commercialization of generative AI; now, multiple Chinese media outlets cited both sets of figures side by side, directly translating the approximately $20 billion difference into "overly optimistic," accompanied by the phrase "is the bubble about to burst?" which easily wins in terms of emotional impact.
However, if we slightly widen the perspective to view this from the entire technology cycle and corporate procurement rhythm, this linear reasoning that "revenue not as desired = bubble burst" is itself a form of bubbling. Over the past two years, large models have been pushed onto nearly every industry agenda, and the actual implementation into budgets, process transformations, and risk assessments will naturally lag behind the expansion speed of marketing narratives, especially during a phase where data compliance, computational cost, and organizational migration are still in transition. The market previously used the nearly $70 billion annual recurring revenue signal to support the optimistic anchor point about commercialization speed and valuation; now the approximate $50 billion annual revenue revealed by internal documents pulls that optimism back somewhat, representing more of a correction from reality to the narrative, rather than the end of the story.
Moreover, the research brief repeatedly reminds us: whether it's $70 billion or $50 billion, all current revenue signals stem from media reports of internal documents and insiders; lacking official disclosures of itemized accounts and growth rates, it is impossible to seriously reconstruct a set of arguments indicating that "profit models have peaked" based on items and growth rates. Under the premise of incomplete information, simply taking "missing $20 billion" as ironclad evidence of a bubble burst while neglecting the fact that OpenAI is still viewed as a leader in the generative AI and large model track, with annual revenue still around the $50 billion level, is, in itself, another form of emotional interpretation. Rather than proclaiming that the bubble has burst, it would be better to acknowledge that expectations are being cooled; what is truly worth observing is how the market will seek to redefine future pricing coordinates between the reality of high income and the corrected narrative.
Arena's Valuation Almost Doubles in Nine Months: Capital Favors the "Shovel Sellers"
At the same time that the income story of leading companies is being recalibrated, Arena stands on the other side. According to research briefs, it is not a consumer-facing large model application but rather an AI model evaluation platform focused on "benchmark testing" and "evaluation tools," closer to a "quality gate" following computational power and data: various models must first pass this assessment gate before entering the market. In light of this positioning within the infrastructure and tools layer, Bloomberg reported that Arena raised $200 million in its latest round of financing, resulting in a post-financing valuation of approximately $3.1 billion; while earlier in January, the same media outlet mentioned its valuation was only about $1.7 billion, nearly doubling in the short span of nine months. This round of financing was led by Lightspeed Venture Partners and Khosla Ventures, with participation from Salesforce Ventures and Dell Technologies Capital, showcasing a mixed lineup of venture and industry capital that embodies a stance of "lining up with the shovel sellers."
As OpenAI's revenue signal reminds the market to lower its fantasies about commercialization speed, Arena has instead been elevated to a higher valuation range in the assessment and benchmark sector. This contrast indicates that the story has not extinguished but rather has had its protagonist rewritten by capital. In the context of a rapidly increasing number of models with increasingly nuanced performance differences, whoever can define "good or bad" and provide a unified comparable evaluation standard holds the discourse power in this new round of competition. As an assessment platform situated within the infrastructure layer, it does not directly bet on the success or failure of any particular terminal application but has the opportunity to charge all participants a "gate fee." This typical "shovel seller" position precisely aligns with the risk preferences of venture and industry capital during a phase of cooling expectations. Therefore, even though Arena has yet to disclose details like revenue or user scale, top-tier capital is still willing to nearly double its valuation in less than a year, betting on a more horizontal pipeline, reflecting the current AI industry's shift from a narrative focused on a single hero to a competition over tools and infrastructure pricing.
Where Is the Money Going? Models Make Money Slowly, Yet Tool Valuations Soar
When the Financial Times produced internal financial documents indicating "as of the end of September, OpenAI's annual revenue is close to $50 billion," the market realized that the previously released "$70 billion annual recurring revenue" by Axios' sources was more like an overly magnified optimistic anchor point. The research brief tagged OpenAI with "the application/model layer revenue recognition speed is below expectations," underscoring that top players are not meeting expectations, implying that the storytelling cycle solely relying on models and direct applications is taking longer, slower, and more circuitous paths than capital imagined.
In juxtaposition, there is another curve where Arena's valuation has been raised from $1.7 billion to $3.1 billion in under a year. The brief emphasizes that this tool layer player, positioned as an "AI model evaluation platform," has not been asked to deliver a clear revenue report like OpenAI. Instead, benefiting from a mixed lineup of Lightspeed, Khosla Ventures, and various industrial capitals, it completed a reevaluation of infrastructure value with $200 million in financing. The shift of money from revenue expectations in the application layer towards supporting selection, comparison, and evaluation across the entire industry in a “shovel seller” position reflects the real migration of the current AI allocation logic: Technology and crypto market investors, if they continue to mix all AI-related targets into a single “story stock” list without distinguishing between storytelling application parties and those with long-term tool attributes in the infrastructure layer, will struggle to stand on the more risk-return matched side in this misalignment of “revenue expectations cooling vs. valuation enthusiasm remaining strong.”
From Valuation to Revenue: The Cool Calculation of AI's Next Phase
The timelines of OpenAI and Arena juxtaposed provide the market with a dual "valuation and revenue calibration lesson": On one hand, the Financial Times cites internal documents pulling the annual revenue back to the reality range of close to $50 billion by the end of September, revealing an about $20 billion gap between this and the previously widely circulated optimistic anchor point of "close to $70 billion," compelling everyone to recalculate the real monetization pace of leading models; on the other hand, Bloomberg records Arena's leap from a valuation of about $1.7 billion in January this year to about $3.1 billion after the latest round of financing, raising $200 million, while capital still remains willing to pay higher prices for infrastructure platforms during the cooling of revenue expectations. Looking ahead, the storytelling in the model layer is transitioning from "volume first, then reconciling" to detailed accounting focused on annual revenue, renewal rates, and cost structures, while the infrastructure and tools layer is increasingly being viewed as productive factors to be acquired long-term for years to come, competing for quasi-public facility pricing rights amidst the interplay of industrial capital and venture capital. For technology and crypto market investors, this means they need to utilize a more rational framework to filter targets: First, check whether revenue signals are showing embryonic forms of sustainable cash flow; second, assess whether valuations are habitually overstretching the commercialization timeline; and finally, distinguish between storytelling application parties and those bearing long-term tool attributes in the infrastructure layer, placing capital in those enterprises that can still reliably generate cash flow, even as expectations are downgraded.
Join our community to discuss and grow stronger together!
AiCoin exclusive Hyperliquid benefits: https://app.hyperliquid.xyz/join/AICOIN88
AiCoin exclusive Aster benefits: https://www.asterdex.com/zh-CN/referral/9C50e2
On-chain Telegram community: https://t.me/AiCoinWhaleData
On-chain community: https://www.aicoin.com/link/chat?cid=N6OVMor5g
AiCoin on-chain Twitter: https://x.com/aicoinwhaledata
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
