Key Moments
Is AI Trending Up or Down in 2026? (Let’s Take a Closer Look)
Deep Questions with Cal NewportKey Moments
AI's promise of a new technological era is misleading; most startups are mere wrappers for models, facing unsustainable costs and an inability to scale profitably, unlike SaaS models.
Key Insights
The "Open Claw" story, initially hyped as AGI or a major leap, was primarily a Python library enabling calls to existing LLMs, leading to widespread misinterpretations and credulous media coverage.
Anthropic's claims of ethical AI and opposition to military use were contradicted by their ongoing involvement and contracts with the Department of War, highlighting a disconnect between public statements and actions.
Estimates suggest only about a third of announced US data centers (15.2 GW out of 115 GW by 2028) are actually under construction, indicating significant potential for project cancellations and fraud.
NVIDIA has likely pre-sold more GPUs (potentially hundreds of billions of dollars worth) than there are currently feasible data center or installed capacity to accommodate them, raising questions about their revenue recognition.
The core business model of many AI startups is problematic, as increased user engagement directly translates to higher operational costs due to LLM compute expenses, unlike traditional SaaS models.
The current AI landscape is characterized by "dread laundering," where amplified anxieties about job displacement or artistic destruction are used to obscure weak technological underpinnings and unsustainable economic models.
The "Open Claw" frenzy was a misinterpretation of a simple library facilitating LLM calls, not a breakthrough.
The initial media and public reaction to "Open Claw" in early 2026, which saw it hailed as a potential AGI or a significant leap forward, was largely based on a misunderstanding of the technology. Ed Zitron clarifies that Open Claw is fundamentally a Python library designed to make it easier for developers to write code that interacts with Large Language Models (LLMs). It doesn't introduce a new AI or LLM but rather simplifies the process of building agentic applications that leverage existing LLM capabilities. The hype surrounding its "autonomous" agents posting on social networks like "Molt Book" was attributed to LLMs generating plausible, albeit often generic or sci-fi-flavored, content when prompted, rather than genuine emergent intelligence. This phenomenon was amplified by what Zitron calls "credulous media coverage," where outlets sensationalized the capabilities, equating the integration of LLMs with APIs to a "ChatGPT moment" or even the singularity, despite the underlying technology being derivative of existing LLM functionalities. The ease with which people were "freaked out" by the prospect of AI agents interacting, often fueled by prompts that primed LLMs towards dystopian themes, highlights a collective lack of understanding about LLM mechanics.
Anthropic's ethical posturing clashes with its military contracts and financial reporting.
The narrative surrounding Anthropic's role in the Department of War contracts reveals a significant gap between their public image and their business practices. Despite public statements expressing reservations about AI's use in mass domestic surveillance and fully autonomous weapons, Anthropic had already been embedded with classified access within the US military since June 2024, involved in operations like incursions in Venezuela and the war in Iran. The timing of their "narrowing" of contract terms, just before the war escalated, was seen as strategic rather than purely ethical. Furthermore, in a sworn affidavit for a lawsuit, Anthropic's CFO revealed that the company had only made $5 billion in revenue to date, a figure that appears to significantly understate reported revenues and investments. This discrepancy, coupled with allegations of pushing enterprise users onto APIs even when using their core services and the general lack of transparency around their revenue streams (with a disputed 85% claimed from API calls), paints a picture of a company prioritizing growth and funding through aggressive means, even at the expense of its ethical claims. The media's focus on Anthropic as an "ethical company" was largely seen as a form of "dread laundering," using the war context to amplify their perceived moral high ground without scrutinizing their actual conduct or financial realities.
The data center boom is significantly overhyped, with a large portion of announced projects facing delays or cancellations.
Concerns about the pace and reality of AI-driven data center expansion are growing, with evidence suggesting a substantial divergence between announcements and actual construction. Reports indicate that only about a third of the announced 115 GW of data center capacity slated for 2028 is currently under construction, totaling roughly 15.2 GW. This means a substantial portion of proposed projects could be delayed or canceled. The problem is exacerbated by the sheer scale of GPU demand; NVIDIA has claimed visibility into half a trillion dollars in GPU sales by the end of 2026, far exceeding the capacity being actively built. This mismatch raises significant questions about where these components are going, with speculation that NVIDIA may be pre-selling future production or engaging in accounting practices that recognize revenue without immediate shipment. Hyperscalers like Microsoft are ordering servers through Original Design Manufacturers (ODMs) in Taiwan, which then warehouse the components, leading to mounting inventories and potential financial risks. Many smaller, non-hyperscale data center announcements are likely speculative ventures with little to no actual construction, suggesting a potential for widespread fraud or at least significant financial miscalculations.
NVIDIA faces scrutiny over GPU sales outpacing data center construction.
NVIDIA's immense market capitalization and projected revenue growth are intertwined with a massive demand for its GPUs. However, the pace of data center construction lagged significantly behind GPU sales, suggesting a potential disconnect. While NVIDIA claims billions in future GPU sales, only a fraction of the planned data center capacity is actively being built. This raises questions about the logistics of GPU deployment. A plausible explanation offered is that hyperscalers are purchasing GPUs through ODMs, which then warehouse them, anticipating future data center availability. This could allow NVIDIA to book revenue prematurely, potentially through transfer of ownership agreements, even if the hardware remains in warehouses, as suggested by growing inventory levels reported by NVIDIA. The sheer volume of GPUs sold, potentially exceeding what can be installed in the near future, points to a reliance on future builds that may not materialize, creating a fragile economic situation.
Widespread media hype and "dread laundering" obscure the true state of AI.
A recurring theme is the media's tendency to amplify AI narratives, often engaging in what Cal Newport terms "dread laundering" – using one set of AI-related anxieties (e.g., job automation) to bolster less substantiated fears (e.g., existential threats, artistic destruction). This sensationalism creates an environment where the actual progress, economic viability, and limitations of AI are obscured. Both hosts criticize the persistent, uncritical coverage, noting that many bold predictions, like widespread job losses or immediate singularity events, have not materialized. The aggressive dissemination of "doom porn" and hyperbolic claims, often unchecked by rigorous reporting, contributes to a public perception that is disproportionate to current AI capabilities and economic realities. This creates a cycle where hype generates investment and attention, which in turn fuels more hype, making it difficult to have a grounded discussion about AI's true impact and challenges.
The economic model of AI startups is fundamentally flawed due to escalating compute costs.
Unlike traditional Software as a Service (SaaS) models where increased user adoption typically leads to greater profitability and marginal costs per user, AI startups built around LLMs face an inverse dynamic. The core issue is that serving more users or more intensive usage of LLM-based applications directly increases operational costs due to the high compute requirements. This means that the most engaged and valuable customers, who use the services the most, are also the most expensive to support. This antithetical business model—where growth amplifies costs rather than scaling profits—renders many AI startups inherently unprofitable. While chatbots and coding assistants may be popular, their underlying economics are unsustainable, making it difficult for these companies to achieve profitability or secure viable exits, especially as many are essentially "wrappers" for foundational models owned by larger entities.
The AI sector faces significant financial risks, potentially leading to a market correction or "AI winter."
The confluence of overhyped data center construction, inflated GPU sales, and unsustainable AI startup business models points towards an inevitable financial reckoning. The sheer volume of debt financing the speculative data center projects and VC funding in AI startups creates a fragile system. When these ventures fail to generate returns—due to high operating costs, lack of scalable profitable products, or an inability to exit—a cascade effect is likely. This could manifest as a stock market hit affecting NVIDIA and related companies, a crisis in the private credit market where much of this debt is held, and a collapse in venture capital valuations for AI startups. The expectation is not necessarily a complete market crash akin to the 2008 financial crisis, as the current situation is less complex and derivative-heavy, but a significant "AI winter" where investment dries up, startups face a fire sale, and the industry experiences a prolonged downturn. The lack of genuinely profitable AI products and the difficulty in controlling user costs are central to this looming instability.
Mentioned in This Episode
●Products
●Software & Apps
●Companies
●Organizations
●Concepts
●People Referenced
Potential Data Center Capacity vs. Actual Construction (GW)
Data extracted from this episode
| Year | Announced Capacity (GW) | Under Construction (GW) | Percentage Under Construction |
|---|---|---|---|
| 2026 | 12 | 4 (approx. 1/3) | ~33% |
| By End of 2028 | 115 | 15.2 | ~13.2% |
Common Questions
Based on the discussion, 2026 has been characterized by significant hype and media frenzy around AI, overshadowing underlying economic and performance issues. The overall sentiment leans towards a critical view, suggesting it hasn't been a definitively 'good' year for AI in a fundamental sense, despite the constant stream of new developments.
Topics
Mentioned in this video
AI commentator who is a guest on the show, providing a reality check on AI news and trends.
CEO of Nvidia, discussed in the context of their GTC conference and the company's market capitalization and GPU sales.
Mentioned in the context of OpenAI's actions and claims regarding negotiations with the US military.
An individual from the US military who commented on the use of AI, specifically Claude, in military operations.
CEO of Anthropic, whose statements regarding the company's contract with the Department of War and AI ethics are central to the discussion.
The host of the podcast 'Deep Questions', who invites Ed Zitron to discuss AI.
Developer of the Cicero AI system, which plays Diplomacy.
A character from the Wall Street movies, quoted for his views on speculation.
A philosopher whose work is implicitly contrasted with the media's current approach to AI, suggesting a need for deeper critical analysis.
Director associated with the movie Avatar, whose name similarity to 'Ocarina of Time' inspired the AI assistant's name choice is clarified.
Journalist whose article about Anthropic's revenue presentation to venture capitalists is referenced.
A celebrity mentioned for her social media post supporting Anthropic's Claude, highlighting the disconnect between public perception and company actions.
A baseball player, mentioned humorously in the context of which Mets players should ideally be fired.
Mentioned in relation to the acquisition of Inflection AI by Microsoft.
Founder of XAI, mentioned as a customer of cloud data centers renting GPUs.
A news publication whose headline about 'Mold Book' is quoted, portraying the event as significant and alarming.
A market intelligence company whose research on data center construction is highlighted.
A news outlet whose reporting on data centers is criticized for 'laundering' reputations of companies involved.
A publication that provided credulous coverage of Open Claw and is mentioned for its role in spreading AI hype.
The military branch with which Anthropic has contracts, leading to disputes over AI usage in surveillance and autonomous weapons.
A publication where Jack Clark of Anthropic used to write critical articles.
Mentioned in the context of financial fraud and scams, comparing the AI industry to previous speculative markets.
Mentioned in the context of financial fraud and scams, comparing the AI industry to previous speculative markets.
Mentioned in the context of financial fraud and scams, comparing the AI industry to previous speculative markets.
A baseball team whose fans' experience of dread is used in a comparison to the existential dread associated with AI news.
News agency, mentioned as one of the few publications that touched on the 'shaky revenue math' in Silicon Valley.
News agency, mentioned as one of the few publications that touched on the 'shaky revenue math' in Silicon Valley.
GPU manufacturer whose sales and inventory appear to be outpacing data center construction, leading to potential oversupply and accounting concerns.
AI company whose financial reporting and military contracts are under scrutiny, contributing to the discussion of AI's ethical and business realities.
A company frequently mentioned in conjunction with Anthropic, discussing their acquisition of Open Claw, their business model, and their financial state.
A financial news publication whose reporting on data center delays and cancellations is cited.
Mentioned in the context of their Project Rainia data center and its claimed capacity versus actual GPU power.
Mentioned as a hyperscaler that purchases servers from ODMs, contributing to the GPU supply chain and demand for data centers.
An ODM (Original Design Manufacturer) that builds servers and incorporates GPUs.
Mentioned as a company potentially involved in speculative data center projects that are not actually being built.
Mentioned as a customer of cloud data centers, signing deals to rent GPUs.
An AI company acquired by Microsoft, used as an example in the discussion about the acquisition of AI startups.
An AI company bought by Google, highlighted in the context of AI startup exits and acquisitions.
A company from which ODMs purchase components, noted for a co-founder's arrest related to selling chips to China.
Mentioned as a supplier to ODMs like Corey, which are involved in building data center infrastructure.
Mentioned as an acquirer of AI startups, and as a hyperscaler procuring servers from ODMs for data centers.
Mentioned as a hyperscaler that purchases servers from ODMs for data center infrastructure.
Mentioned as a hyperscaler that purchases servers from ODMs for data center infrastructure.
Music streaming service controlled by a personalized AI assistant.
Speaker system controlled by a personalized AI assistant.
Mentioned as Meta, a hyperscaler involved in data center infrastructure.
A platform for code hosting, mentioned in the context of 'slop commits' related to Open Claw's adoption.
Manufacturer of the Mac mini, mentioned in the context of its sales potentially increasing due to AI hype around Open Claw.
Anthropic's large language model, discussed in the context of LLM capabilities, military contracts, and revenue.
An AI startup that might face a fire sale due to the current market conditions and difficulty in exiting AI companies.
Anthropic's AI model/service, mentioned in the context of its marketing pushes and throttling of services.
A specific version of Anthropic's Claude model, mentioned as the base for a personalized digital assistant setup.
Mentioned as a hyperscaler that purchases servers from ODMs for data center infrastructure.
An AI system developed by Noam Brown that plays the board game Diplomacy, used as an example of modular AI architecture.
A search engine where one of the co-founders of a recursive self-learning company also runs a related entity.
An AI-generated image of Nvidia's CEO Jensen Huang with lobster claws, released at the GTC conference and discussed in the context of AI hype.
A well-known AI chatbot and the benchmark against which new AI platforms like Open Claw are often compared, sometimes inaccurately.
A family of large language models from OpenAI, mentioned as a source of expensive API calls that Open Claw users encountered.
Email service controlled by a personalized AI assistant.
A coding harness or AI tool popular among programmers, mentioned as an example of a non-profitable AI product.
Refers to Anthropic's AI model, potentially used in a military context.
Productivity and note-taking app used by a personalized AI assistant's owner.
Mentioned in relation to Mac Minis selling out due to the hype around Open Claw.
Mentioned as the location where Nvidia GPUs might be warehoused and ODMs build servers.
Mentioned in the context of Nscale's data center project in Lton, which is described as having only scaffolding.
Mentioned as the location of a war where Anthropic's Claude was reportedly used.
Mentioned in relation to Anthropic's prior use in a military incursion.
Mentioned via Cincinnati and Pittsburgh as cities whose sports fans experience dread, contrasted with Mets fans' potential relief from AI news.
Mentioned in relation to Super Micro's co-founder being arrested for selling chips to China.
Mentioned in relation to Mac Minis selling out due to the hype around Open Claw.
Mentioned again as the location for project Matador, an 11 GW data center that reportedly has fraud issues.
Location of Amazon's Project Rainia data center, discussed in terms of its claimed capacity.
Mentioned in the context of the war in Iran and potential AI use.
The United States, where many of the AI developments, data center constructions, and financial concerns discussed are focused.
The cost incurred when using AI models through an interface, a significant expense for AI startups and users.
A hypothetical future point where technological growth becomes uncontrollable and irreversible, incorrectly associated with early AI releases.
A newly coined term describing the practice of amplifying fear about one AI issue to mask or amplify fear about another, often less substantiated one.
A vision for AI where numerous small, task-specific systems work together, rather than a single, all-powerful AI entity.
A period of reduced investment and interest in AI research and development, predicted as a potential outcome of the current AI bubble bursting.
Large Language Models, the underlying technology for many AI applications, discussed in terms of their cost, capabilities, and limitations.
A theoretical AI capability claimed by some companies, used as an example of potentially unsubstantiated AI advancements that raise funding.
The market for loans and debt outside of traditional public exchanges, discussed as being at risk due to AI-related investments.
Investment firms mentioned in the context of the SAS model and how a focus on scaling might overlook underlying profit mechanics.
The expansion of data centers fueled by AI demand, critically assessed for its actual construction status versus announcements.
More from Cal Newport
View all 292 summaries
86 minHow to Build Discipline in a Distracted World
26 minIs Claude Mythos “Terrifying”? (According to Experts: No.)
91 minCan Downgrading Your Tech Upgrade Your Results? | Cal Newport
64 minRules For Deep Work — Updated for 2026 | Cal Newport
Found this useful? Build your knowledge library
Get AI-powered summaries of any YouTube video, podcast, or article in seconds. Save them to your personal pods and access them anytime.
Get Started Free