OpenAI: Still hallucinating?

LiveWire12-01

In our previous GQG Research, Dotcom on Steroids, we explored the striking parallels between the current AI landscape and the euphoric rise—and subsequent fall—of the late 1990s technology, media, and...

Source Link

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Comments

We need your insight to fill this gap

Leave a comment

{"i18n":{"language":"en_US"},"isChannel":false,"data":{"share":"https://ttm.financial/m/news/2588482017?lang=en_US&edition=fundamental","thumbnail":"","is_english":true,"pubTime":"2025-12-01 12:59","share_image_url":"https://static.laohu8.com/e9f99090a1c2ed51c021029395664489","id":"2588482017","market":"nz","top_or_hot":-1,"title":"OpenAI: Still hallucinating?","media":"LiveWire","content":"<div>\n<p>In our previous GQG Research, Dotcom on Steroids, we explored the striking parallels between the current AI landscape and the euphoric rise—and subsequent fall—of the late 1990s technology, media, and...</p>\n\n<a href=\"https://www.livewiremarkets.com/wires/openai-still-hallucinating-2025-11-30\">Source Link</a>\n\n</div>\n","source":"livewire_stock","html":"<!DOCTYPE html>\n<html>\n<head>\n<meta http-equiv=\"Content-Type\" content=\"text/html; charset=utf-8\" />\n<meta name=\"viewport\" content=\"width=device-width,initial-scale=1.0,minimum-scale=1.0,maximum-scale=1.0,user-scalable=no\"/>\n<meta name=\"format-detection\" content=\"telephone=no,email=no,address=no\" />\n<title>OpenAI: Still hallucinating?</title>\n<style type=\"text/css\">\na,abbr,acronym,address,applet,article,aside,audio,b,big,blockquote,body,canvas,caption,center,cite,code,dd,del,details,dfn,div,dl,dt,\nem,embed,fieldset,figcaption,figure,footer,form,h1,h2,h3,h4,h5,h6,header,hgroup,html,i,iframe,img,ins,kbd,label,legend,li,mark,menu,nav,\nobject,ol,output,p,pre,q,ruby,s,samp,section,small,span,strike,strong,sub,summary,sup,table,tbody,td,tfoot,th,thead,time,tr,tt,u,ul,var,video{ font:inherit;margin:0;padding:0;vertical-align:baseline;border:0 }\nbody{ font-size:16px; line-height:1.5; color:#999; background:transparent; }\n.wrapper{ overflow:hidden;word-break:break-all;padding:10px; }\nh1,h2{ font-weight:normal; line-height:1.35; margin-bottom:.6em; }\nh3,h4,h5,h6{ line-height:1.35; margin-bottom:1em; }\nh1{ font-size:24px; }\nh2{ font-size:20px; }\nh3{ font-size:18px; }\nh4{ font-size:16px; }\nh5{ font-size:14px; }\nh6{ font-size:12px; }\np,ul,ol,blockquote,dl,table{ margin:1.2em 0; }\nul,ol{ margin-left:2em; }\nul{ list-style:disc; }\nol{ list-style:decimal; }\nli,li p{ margin:10px 0;}\nimg{ max-width:100%;display:block;margin:0 auto 1em; }\nblockquote{ color:#B5B2B1; border-left:3px solid #aaa; padding:1em; }\nstrong,b{font-weight:bold;}\nem,i{font-style:italic;}\ntable{ width:100%;border-collapse:collapse;border-spacing:1px;margin:1em 0;font-size:.9em; }\nth,td{ padding:5px;text-align:left;border:1px solid #aaa; }\nth{ font-weight:bold;background:#5d5d5d; }\n.symbol-link{font-weight:bold;}\n/* header{ border-bottom:1px solid #494756; } */\n.title{ margin:0 0 8px;line-height:1.3;color:#ddd; }\n.meta {color:#5e5c6d;font-size:13px;margin:0 0 .5em; }\na{text-decoration:none; color:#2a4b87;}\n.meta .head { display: inline-block; overflow: hidden}\n.head .h-thumb { width: 30px; height: 30px; margin: 0; padding: 0; border-radius: 50%; float: left;}\n.head .h-content { margin: 0; padding: 0 0 0 9px; float: left;}\n.head .h-name {font-size: 13px; color: #eee; margin: 0;}\n.head .h-time {font-size: 11px; color: #7E829C; margin: 0;line-height: 11px;}\n.small {font-size: 12.5px; display: inline-block; transform: scale(0.9); -webkit-transform: scale(0.9); transform-origin: left; -webkit-transform-origin: left;}\n.smaller {font-size: 12.5px; display: inline-block; transform: scale(0.8); -webkit-transform: scale(0.8); transform-origin: left; -webkit-transform-origin: left;}\n.bt-text {font-size: 12px;margin: 1.5em 0 0 0}\n.bt-text p {margin: 0}\n</style>\n</head>\n<body>\n<div class=\"wrapper\">\n<header>\n<h2 class=\"title\">\nOpenAI: Still hallucinating?\n</h2>\n\n<h4 class=\"meta\">\n\n\n2025-12-01 12:59 GMT+8&nbsp;&nbsp;&nbsp;<a href=https://www.livewiremarkets.com/wires/openai-still-hallucinating-2025-11-30><strong>LiveWire</strong></a>\n\n\n</h4>\n\n</header>\n<article>\n<div>\n<p>In our previous GQG Research, Dotcom on Steroids, we explored the striking parallels between the current AI landscape and the euphoric rise—and subsequent fall—of the late 1990s technology, media, and...</p>\n\n<a href=\"https://www.livewiremarkets.com/wires/openai-still-hallucinating-2025-11-30\">Source Link</a>\n\n</div>\n\n\n</article>\n</div>\n</body>\n</html>\n","isBrief":false,"type":0,"news_type":1,"symbol":"BK4551","symbol_name":"寇图资本持仓","start_time":0,"source_url":"https://www.livewiremarkets.com/wires/openai-still-hallucinating-2025-11-30","article_id":"2588482017","we_media_id":null,"thumbnails":[],"rights":{"source":"livewire_stock","url":"https://www.livewiremarkets.com/wires/openai-still-hallucinating-2025-11-30","rn_cache_url":null,"directOrigin":true},"url":"https://stock-news.laohu8.com/highlight/detail?id=2588482017","pubTimestamp":1764565171,"columns":[],"sourceInfo":{"source_id":"livewire_stock","name":"Livewire Markets"},"weMediaInfo":null,"summary":"In our previous GQG Research, Dotcom on Steroids, we explored the striking parallels between the current AI landscape and the euphoric rise—and subsequent fall—of the late 1990s technology, media, and telecommunications bubble. In this follow-up piece, we delve into OpenAI, and take a hard look at the numbers, narratives, and risks separating the technological marvel from the economic reality.On the surface, OpenAI may sound like a Mag 7 company in its early days—unprofitable, burning through cash, and a magnet to investors who think they have found the next pot of gold. Peel one layer or dare to bring out your calculator for a closer look, and OpenAI goes from being the world’s “most valuable start-up” to what we believe is one of the most overvalued and overhyped companies in history.","collect":0,"end_time":0,"defaultTopTitle":"livewiremarkets.com","property":[],"viewcount":null,"language":"en","relate_stocks":{"BK4551":"寇图资本持仓","BK4587":"ChatGPT概念","LU0498741114.HKD":"FRANKLIN GOLD & PRECIOUS METALS \"A\" (HKD) ACC","AGI.AU":"AINSWORTH GAME TECHNOLOGY LT","BK4528":"SaaS概念","BK4588":"碎股","LU0055631609.USD":"贝莱德世界黄金基金A2","LU0368265764.SGD":"Blackrock World Gold Fund A2 SGD-H","BK4585":"ETF&股票定投概念","LU0498741890.SGD":"FTIF - Franklin Gold and Precious Metals A (acc) SGD","BK4023":"应用软件","AGI":"Alamos Gold Inc","BK7078":"赌场与赌博","LU0496367417.USD":"富兰克林黄金和贵金属A（acc）","BK4017":"黄金","AI":"C3.ai, Inc.","BK4543":"AI"},"translate_title":"OpenAI：还在产生幻觉吗？","themeId":null,"isJumpTheme":false,"ttsUrl":null,"symbols_score_info":{"AGI.AU":1,"AGI":1,"AI":0.9},"content_text":"In our previous GQG Research, Dotcom on Steroids, we explored the striking parallels between the current AI landscape and the euphoric rise—and subsequent fall—of the late 1990s technology, media, and telecommunications bubble. In this follow-up piece, we delve into OpenAI, and take a hard look at the numbers, narratives, and risks separating the technological marvel from the economic reality.\nScaling the heights, ignoring the cracks\nOn the surface, OpenAI may sound like a Mag 7 company in its early days—unprofitable, burning through cash, and a magnet to investors who think they have found the next pot of gold. Peel one layer or dare to bring out your calculator for a closer look, and OpenAI goes from being the world’s “most valuable start-up” to what we believe is one of the most overvalued and overhyped companies in history.\nIn our view, the company’s financials are broken and unrealistic even if you lower the standards and look at them through the lens of a start-up with potentially revolutionary technology. OpenAI is more capital intensive than any other start-up we have seen, lacking the stickiness, the moat, and the network effect that have paved the way for other tech success stories. In addition, the disconnect between OpenAI’s revenue and valuation is alarming by any historical standard. As a comparison, consider Amazon or Google in their earlier days: when these companies had a $500B valuation, their revenues were roughly 10x and 4x as large, respectively, in comparison to OpenAI’s current run-rate revenue of ~$20B.1\nAs a long-only large cap manager investing in listed equities on behalf of its clients, most of the time we have the luxury of ignoring the less transparent world of private companies and some of their egregious valuations. So why is that different today? We are facing a unique situation where most of the large public players in tech are intertwined in a web of circular financing where money and deals flow through OpenAI. These relationships have direct implications for not just the large tech companies in our investable universe, but also many other companies in the S&P 500 in our view.\nOpenAI is also playing a key role in driving the “hopes and dreams” factor, which we believe is helping to fuel this AI bubble. They believe Large Language Models (LLMs) will continue seeing meaningful step-function improvements toward this goal of Artificial General Intelligence (AGI), a notion a number of AI experts have started to pour cold water on.2,3 In fact, just recently we attended an AI conference in Miami, in which OpenAI’s CFO Sarah Friar conveyed to a room filled with stakeholders, including investors and lenders, that the company’s “true north” is “ AGI for the benefit of humanity.” In short, we see OpenAI as a capital-intensive business model that offers a product that faces meaningful risks of becoming a commodity. Yet, it continues to magically prop up valuations of its “partners” after each deal announcement.\nExplosive growth, unprecedented spending\nTo give credit where it is due, OpenAI has firmly established itself as a leading innovator with its LLMs. Its market leadership is evidenced by rapid user adoption of their ChatGPT product which has swelled to over 800M weekly active participants and many large enterprise customers, in addition to revenue of billions of dollars, in the span of a few years.4,5\n\n\n\n\n\n\n\n\n\n\nHowever, success is rarely free.\nThis explosive growth has been financed by record-breaking capital infusions and expenditures to build the infrastructure to support it.6 This unprecedented level of spending is underwritten by market forecasts of what we view as extraordinary economic gains that may ultimately prove to be unsustainable. While these investments signal widespread enthusiasm, they obscure the structural challenges that threaten OpenAI’s path to durable profitability. Despite the hype, the translation of this breakthrough technology into a resilient business is far from certain.While we will not go into the details here, refer to the links at the end of this piece to read about what has been widely reported on both the circularity and domino effect OpenAI is having on other companies.7,8,9 Instead, we will delve into OpenAI’s dynamics from an economic and technology perspective to illustrate why we believe it fails the proverbial “smell test” given that most analysts seem to be ignoring the facts.\nThe myth of the metric\nWhile impressive on the surface, high usage metrics (of which OpenAI has many to showcase) can often mask fundamental weaknesses in the quality of demand for a product and its ability to retain users. Specifically, a significant portion of OpenAI’s current user engagement is largely driven by free-tier users using it for non-productive reasons.10,11 This type of usage inflates activity metrics used for raising copious amounts of capital, but we believe it holds little signaling power for future revenue, making it a poor proxy for sustainable customer value. Not to mention that 90% of ChatGPT users are outside of the United States, according to Friar. Furthermore, we believe the translation of enterprise use into durable and sticky revenue is undermined by several key factors:\n\nPersistent Reliability Gaps: LLMs continue to struggle with hallucinations.12 Their lack of consistency erodes trust and makes it difficult for the models to be embedded in mission-critical, multistep applications or operations that generate recurring, high-value revenue. The primary workaround, involving re-runs (increased token13 usage) and human-in-the-loop verification, directly counteracts the core promise of LLM efficiency and cost reduction, ultimately slowing adoption.14\nFragile Adoption and Low Switching Costs: The current user base is heavily weighted toward free-tier participants who have no financial commitment to the platform.11 Even for paying enterprise customers, switching costs remain low for now. Currently, 28% of OpenAI’s API usage flows through low-code platforms like Zapier, Bubble, and Retool.5 These companies are model agnostic and can easily redirect workflows to better or more cost-effective models as they become available, including open-source models.\n\n\n\n\n\n\n\n\n\n\n\n\nQuestionable Performance Benchmarks: Extreme growth expectations are often justified by referencing the improving model performance on AI benchmarks.15 However, this too gives us pause. Some studies are beginning to show how these models have “gamed” benchmark results by “memorizing” the questions rather than understanding the subject.16 Beyond this technicality, the benchmarks create what we believe is a conceptual mismatch between what they measure and what enterprises value. They reward a model’s breadth of capability, making it an impressive generalist, yet corporations tend to be structured around the reliability of specialist employees performing a certain set of tasks with consistency. Because standard benchmarks do not adequately measure this consistency, a model’s top-tier score in a benchmark test can be a misleading indicator of its real-world utility.\nThe cheaper it gets, the more you pay\nEven if user engagements were a perfect proxy for value, the basic economics of generative AI present a challenge. While the cost per token has plummeted due to hardware and software innovations, the cost per query has not seen the same dramatic decrease.17\nWith the mainstream adoption of reasoning models, the economics of LLM usage have shifted dramatically.18 Paying consumers are charged for each token the model takes as input and each token that the model outputs. It is important to note that users are charged more for tokens generated by the model than the tokens that are input. While the general cost per token has plummeted due to hardware enhancements, the number of tokens consumed per task has not.\nAn analysis by an AI gateway provider revealed an interesting trend: simple, single-turn queries, which constituted 80% of enterprise usage in early 2024, dropped to just 20% by year end. They were replaced by multi-step, complex workflows that drive up token consumption for each request, with most of them being the costlier output tokens.19\nThis shift is stark when comparing different model generations. Early models were often tasked with simple Q&A, where a query might consume a few hundred tokens. Today’s reasoning models are used for complex, multi-turn tasks like analyzing a lengthy document, searching the web, or creating large pieces of working software, which increases the token count by a couple orders of magnitude.\nThis “token inflation” is further exacerbated by the models’ tendency to be overly verbose and the operational necessity of re-running queries to suppress hallucinations and ensure output consistency. Ultimately, these factors combine to drive up the cost per answer, leaving users to question true efficiency gains.\nWill generative AI get sufficiently better from here?\nIt is not unusual for companies to invoke the promise of AGI when pressed on whether today’s use cases justify the massive amount of AI infrastructure investment companies are making. What should be unsettling is that the notion of near-term AGI is increasingly challenged by researchers and operators (including OpenAI founders themselves), with evidence of continued scaling and deployment realities pointing to a longer, harder road for advancement.2\nBut let us lower the standards a bit and put AGI to the side—can AI, as we know it today, meaningfully improve? Was GPT-5 materially better than its predecessor, and was its rate of improvement comparable to the step up from GPT-3 to GPT-4? A cursory glance at release cadence, benchmark deltas, sharply rising compute, and power requirements suggests step-function gains are becoming rarer and more expensive—an uncomfortable backdrop for valuations and hype that implicitly assume (need?) continued leaps.\nRising Costs, Shrinking Gains\n\n\n\n\n\n\n\n\n\n\nThe uncomfortable truth is that this technology is running into hard constraints: the reservoir of high-quality human data is finite, returns to scale are slowing, and a recursive reliance on synthetic data risks degrading the signal. As a result, products like ChatGPT look a lot less like software with zero-marginal-cost scale and more like metered compute with rising variable inputs and operational choke points.\nThe illusion of scale\n\nThe unit economics of LLM providers do not seem to align with the high-margin, scalable models of successful Software-as-a-Service (SaaS) and marketplace companies. Unlike a SaaS model, the cost of adding a new LLM customer is not zero because the cost of compute for LLMs scales with the users, leading to an organic barrier to economies of scale that this business model can achieve. And unlike Uber, as an example, an LLM does not benefit from network effects in that the value of its service does not inherently improve as more users engage with it.\nHowever, this flawed comparison appears to be driving a classic Venture Capital-subsidized “blitz-scaling” strategy that is now playing out globally. As shown in the graph, AI adoption has been unequal between geographies with varying GDP per capita. The lowest adoption rates can be seen in low GDP per capita countries, which tend to be more price sensitive, so providers seem to be using deeply subsidized pricing to capture users in these markets.\nOpenAI, for instance, has introduced local plans in India, its biggest market after the US, at a fraction of the global price. Also in India, Perplexity, an AI-powered search engine, is bundling $20 per month subscriptions for free in mobile plans where telecom providers generate as little as $3 per month in revenue per user. It is difficult to rationalize how acquiring these highly price-sensitive users contributes to the future profitability that current valuations demand.\nChatGPT Weekly Average Users\n\n\n\n\n\n\n\n\n\n\nIf growth within the consumer market relies on unsustainable subsidies, the hope for profitability must lie with enterprise adoption. This has been a comparatively bright spot, with enterprise spending on foundation models more than doubling in the first half of 2025. However, we believe this optimism is tempered by significant headwinds.\n\nA few recent studies found that either companies remain in a perennial pilot phase for AI projects, or that most pilot programs for incorporating AI end up failing.20,21 Furthermore, even successful adoption may yield only marginal gains. A recent study estimated AI’s net impact on enterprise profit margins by 2030 at a mere 50-70 basis points.22 While not immaterial, this margin uplift is expected to come primarily from job automation (which we think is becoming an increasingly questionable assumption23) and offset by the significant depreciation expenses of the required AI infrastructure (which is a known and unavoidable cost).\nThe hard financials make this picture even bleaker. In the first half of 2025, OpenAI generated $4.3B in revenue while posting a net loss of $13.5B and burning through $2.5B in cash. Despite this staggering burn rate, the company projects it will be profitable by 2030, with revenues soaring to $200B and gross margins over 60%.24 This type of long-range, hockey-stick forecast for a business with such a murky path to profitability is eerily reminiscent of the dotcom era, where many companies extrapolated short-term hype into extreme future earnings that never materialized.25\nFor context, a $200B revenue target would exceed the 2024 sales of Nvidia, a company that operates as a near-monopoly with immense barriers to entry and is years ahead of its closest competitors, something we believe OpenAI can hardly claim considering how easily interchangeable frontier language models are.26 To suggest that OpenAI—a company burning through cash by the billions amid fierce competition with seemingly no durable moat—will achieve a similar financial profile in just a few years seems less like a forecast and more like a work of speculative fiction in our view.\nLarge-Capitalization Stocks\n\n\n\n\n\n\n\n\n\n\nThis opacity is a private company’s privilege, but it also raises a simple question: if the underlying economics were as strong as the hype, why not disclose them?\n\nThe reluctance to share a complete financial picture suggests the story of hyper growth may be masking a reality different from the hype. And when leadership can float a federal ‘backstop’ one day and swear off bailouts the next, it does not quite read like conviction. In our opinion it reads like pre-wiring a safety net while pretending not to need one.27,28 We found Sam Altman’s defensive reaction when pressed by Brad Gerstner of Altimeter Capital on the math behind the planned $1.4 trillion spend to be similarly telling.29 And prior court depositions and board accounts corroborate the concern about there being a pattern here, not to mention OpenAI’s previous board of directors fired Altman in November 2023 for not being “consistently candid in his communications” with the board! 30,31\nWe believe there is another underappreciated risk: the intense competition for talent among the AI labs. OpenAI is on track to spend close to $6B on stock-based compensation in 2025 alone.32 Since these are all based on the company’s ~$500B valuation, even a modest slowdown could heighten the challenge of retaining key employees. However, despite the possible headwinds, OpenAI continues to raise its revenue expectations.\nThis all leads back to one unavoidable consequence: the math seemingly does not work, and the capital-intensive nature of the business creates relentless liquidity pressure. Despite record-breaking revenue, OpenAI has raised over $40B this year, up from $6.6B in 2024, all in addition to revolving credit lines and infrastructure partnerships. To keep up their growth, arguably by subsidizing the cost of compute for users and maintaining their high-in-demand workforce, we think they will need to keep raising money at record-breaking rates.\nUsing a quick back-of-the-envelope calculation, taking the 136M US private workforce at the going subscription rate of $20 per user per month yields less than $3B in revenues—and of course that is before discounts, usage caps, or reseller splits. To put this into perspective, Google raised just $26M (~$60M in today’s terms) before becoming profitable, while Meta raised about $480M (~$800M today).33,34 By comparison, OpenAI is projected to burn through a staggering $115B before it even hopes to reach free cash flow profitability in 2030.35 The scale of this capital consumption is without precedent.\nRising Revenue\n\n\n\n\n\n\n\n\n\n\n\n\nNot so open to other AIs\nThis constant need for capital exposes a deeper vulnerability: the absence of a durable competitive moat. In a head-to-head matchup against other well-funded labs like Google and Anthropic, any performance advantage is fleeting. Minute differences between top models and low switching costs prevent the formation of a durable price premium or sticky user base, as evidenced by Anthropic’s recent capture of a major share of the software coding market.36\nThis vulnerability is magnified by the disruptive and often understated force of the open-source community. While closed-source models have historically maintained a performance lead, that gap is closing. Notably, China’s open-source Kimi K2 Thinking model has now demonstrated performance results that match or exceed the leading closed-sourced reasoning systems from the likes of OpenAI and Anthropic, underscoring just how quickly the frontier is shifting.37\nAs the graph shows, this is far from anomalous: open-source models consistently catch up to (and now seemingly surpass) the state-of-the-art performance of their closed-source counterparts.\nFurthermore, we have seen that open-source models optimize resource utilization that ultimately improves efficiency and lowers the cost of inference resulting in even more adoption and innovation.\nClosed-Source vs Open-Source Models\n\n\n\n\n\n\n\n\n\n\nThis dynamic mirrors the rise of Linux, which, despite being perpetually behind the polished proprietary systems of its day, steadily grew a loyal following. That following eventually built the foundation for Android, an open-source project that now dominates the mobile operating system market. The lesson is clear: pricing power is not always tied to having the absolute state-of-the-art model. A “good enough” alternative can become the market standard just because it is accessible to more people.\nUltimately, we believe this will democratize access and distribute the costs of development to more companies and builders, making it increasingly difficult for closed-source players to obtain and keep market share at sufficiently attractive economics. We believe this will drive the industry to become more like a marketplace, aiding in more innovation as more builders and consumers come together with aligned incentives. This organic growth tailwind will be difficult to replicate with the centralized, capital-intensive nature of the closed-source AI labs.\nMoreover, closed-source models face challenges with the way established SaaS incumbents like Microsoft and Salesforce are positioned to extract value for their users. Their value proposition is not the raw capability of the AI model, but the productivity increase delivered through an integrated workflow that they already serve to their clients. They can treat the increasingly commoditized AI models as an interchangeable component, while capturing the lion’s share of the value creation.\nThis leaves the foundational model providers like OpenAI bearing the immense and perpetual cost of R&D while the incumbents capture the lion’s share of the profits, relegating the innovators to the position of a low-margin utility supplier. This is not dissimilar to what we saw (and indeed recently wrote about in Dotcom on Steroids) during the TMT bubble of the late 1990s where the big spenders generally did not end up capturing the lion’s share of the economics which that infrastructure enabled.38\nImmortal tech, mortal company\nAs Howard Marks once said, “There are no bad assets, only bad prices.” As the saying goes, LLMs represent a genuine technological innovation with the potential to create immense value for humanity. However, as an investment proposition, we think the sticker price to participate in them is untenable given the uncertain path to translating the innovation into a resilient, capital-efficient business. The current market, fueled by a venture capital culture that has long prioritized growth over fiscal discipline, appears to be conflating headline usage, benchmark victories, and run-rate revenues with durable economics.\nThe current exuberance is particularly jarring because participants seem to be completely discounting signs that the exponential improvements driven by scaling laws may be slowing. Research from major tech labs has highlighted the inherent limitations of LLMs like the ‘illusion of thinking’ paper from Apple.39 Other studies have also cast doubt on whether simply scaling current architectures can lead to true AGI, the very goal that was supposed to justify this level of investment.40 This is a critical point: while LLMs are impressive, in our view, it is still just a technology with its own inherent constraints, not a magic pill to solve the world’s problems.\nThis exuberance also seems to overlook the fundamental fragility of the current business model. OpenAI’s explosive revenue growth, on track to have delivered ~$13B by the end of 2025, has been matched by an equally explosive burn rate, requiring massive and repeated infusions of capital just to sustain operations.41\nMeanwhile its $500B valuation rests on the assumption that this growth can be sustained and eventually converted into profit.42 This premise appears to be increasingly questionable given the structural headwinds and fierce competition from both closed- and open-source AI labs alike.\nAs investors, our role is to allocate capital efficiently, which requires separating the technological marvel from its economic impact. The current approach of funneling capital into a few leaders in a winner-take-all race overlooks the risk of creating entities so large that their potential failure becomes a systemic threat with second-degree effects on the wider economy, in our view.\nWe believe the real test will come when the current wave of enthusiasm and subsidies fades. As capital tightens and competition accelerates, the market will be forced to distinguish between speculative and sustainable enterprises. LLMs are not a bad asset, but the price being paid for a business with such immense capital costs, no obviously durable moat, and questionable unit economics may prove to be fatal. We believe the story to watch is not whether the technology is immortal (it most likely is), but whether the companies building it are.\n\n\n\n\n\nNever miss an update\n\n    Enjoy this wire? Hit the ‘like’ button to let us know. \n    Stay up to date with my current content by \n    following me below and you’ll be notified every time I post a wire","kind":"highlight","is_publish_news":true,"is_publish_highlight":false,"is_publish_live":false,"is_publish_wemedia":null,"editions":null,"column":"","sentiment":"0","news_tag":"","news_rank":0,"symbols":[],"gpt_button":0,"need_auth":false,"code":"91000000","status":"200"},"commentList":[],"isCommentEnd":true,"newsSizeData":{"likeSize":0,"commentSize":0,"repostSize":0,"favoriteSize":0,"likeStatus":false,"favoriteStatus":false},"APP":{"userAgent":"Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; ClaudeBot/1.0; +claudebot@anthropic.com)","isDev":false,"isTTM":true,"deviceId":"web-server-community-laohu8-v3","version":"4.35.9","shortVersion":"4.35.9","platform":"web","vendor":"web","appName":"ttm","isIOS":false,"isAndroid":false,"isTiger":false,"isTHS":false,"isWeiXin":false,"isWeiXinMini":false,"isWeiBo":false,"isQQ":false,"isBaiduSwan":false,"isBaiduBox":false,"isDingTalk":false,"isToutiao":false,"isOnePlus":false,"isHuaWei":false,"isXiaomi":false,"isXiaomiWebView":false,"isOppo":false,"isVivo":false,"isSamsung":false,"isMobile":false},"href":"/m/news/2588482017","isCrawlerRequest":true}