The Man Who Calls BS On AI: AI Is The World’s Greatest SCAM, And They All Know It! | Ed Zitron

The rapid rise of the AI industry has been accompanied by an unprecedented level of hype, investment, and promotional promises. However, beneath this surface-level optimism lies a deeply flawed financial structure that Ed Zitron calls little more than a "con," built on unsustainable losses, misleading metrics, and speculative bubble economics, far removed from genuine technological or business progress.

Deceptive Business Model With Circular Financing Over Real Profits

AI's current boom is not supported by traditional market demand or profitability. OpenAI and Anthropic, the industry's two largest generative AI labs, are emblematic: both are deeply unprofitable and cannot exist without continuous infusions of cash from Microsoft, Amazon, and Google—the very tech giants that use their services and supply them with cloud and GPU infrastructure. This circular financial relationship sees money endlessly flowing between these tech companies, AI labs, and hardware suppliers like Nvidia, rather than being generated from sustainable customer revenues.

Amazon, for instance, sent $50 billion to OpenAI this year and $5 billion to Anthropic, while Google invested $10 billion in Anthropic alone. OpenAI lost $20.9 billion last year, and Anthropic's business model is equally unsustainable. These companies do not present a clear path to profitability. In fact, they forecast that tens or even hundreds of billions of dollars must be injected annually to keep their operations afloat and to fulfill projected revenue numbers—OpenAI alone expects to need $100 billion yearly and a staggering $750 billion by 2030.

The illusion of financial health is maintained by vague and intentionally opaque 'run rate' metrics, rarely defined with any consistency—sometimes referring to a single month's revenues multiplied by twelve or another arbitrary figure. Cloud providers such as Microsoft and Amazon obscure the true scale of their AI revenues, using metrics like 'annualized run rate' without clear explanation, so investors and analysts have little sense of how these enterprises are really performing. Many times, financial statements do not break out AI at all, further shielding performance from scrutiny.

Operationally, the companies run at sustained losses. Subscribers pay $20-40 per month for ChatGPT or Anthropic access, but these subscriptions actually deliver $400 to $14,000 worth of AI computation. A power user on a $200 monthly plan can burn through $14,000 in actual cloud costs; a $20 plan might subsidize $400 in AI usage for the customer. Companies absorb these massive costs, hoping to build market share rather than charging the true value of service, thus relying even more on ongoing investment rather than viable economics.

The economic flywheel is completely dependent on continued, escalating capital injections—a phenomenon reminiscent of past tech bubbles. Once these injections slow or stop, progress in generative AI stutters or halts entirely because there is no market-driven basis for continued investment.

Exorbitant Trillion-Dollar Data Center Costs Unjustified by Minimal Revenue

The scale of infrastructure spending is equally staggering and unsustainable. Over $1 trillion in capital expenditures has already been committed to AI data centers and the GPUs powering them; another trillion is set to be spent in the next year. Yet, the total AI revenue across all major players—Microsoft, Google, Amazon, Nvidia, OpenAI, and Anthropic—barely exceeds $100 billion. Of that, 70% comes directly from OpenAI and Anthropic, which are in turn only kept afloat by cash from Microsoft, Amazon, and Google, creating a feedback loop where industry giants are both the main clients and main benefactors of AI labs.

This loop has led to herd behavior, with Amazon, Microsoft, and Google accelerating their capital expenditures after observing competitors purchasing massive numbers of GPUs from Nvidia. The result is a speculative boom: companies see their rivals spending billions, so they mimic the spending, believing it signals genuine demand when in reality, most "demand" is artificially created by propping up two loss-making companies.

The actual returns on these investments are abysmal. For example, Microsoft reported $34.33 billion in AI revenue (driven by OpenAI), but spent $115 billion on capital expenditures in a single year and plans to spend $175 billion the next. Nvidia sold $215.9 billion in GPUs last year, but the true independent revenue generated by AI usage, outside of the OpenAI and Anthropic circular system, is as low as $22 billion globally.

Despite this, companies continue to take on more debt to fund expansions. Amazon, Google, and Meta have collectively invested over $700 billion in new physical assets in just four years, shifting from cash-rich, asset-light software firms to infrastructure-heavy, highly leveraged businesses.

No breakthrough in hardware efficiency has occurred or appears imminent. Even with every chipmaker—Nvidia, Broadcom, Google, etc.—pushing to improve GPU efficiency, energy costs for running data centers remain stubbornly high. Data centers built today will cost as much to operate in 2050 as they do now, barring some unforeseen leap in technology, making the prospect of eventual profitability even bleaker.

In short, nearly the entire AI "revenue" pool comes from two unprofitable firms funded by three tech giants. The cycle is unsustainable: tens of billions flow in, are spent on training and renting cloud GPUs, and return as revenues only in a closed, synthetic market.

Marketing of Ai Capabilities Misleads Investors, Analysts, Public

The AI industry's problems are compounded by relentless marketing campaigns and media hype, which serve to mislead investors, the press, and the general public about the state of AI and its ec ...

Despite Better Benchmark Scores, Large Language Models Are Still Unreliable and Prone to Hallucinations, Making Them Unsuitable for Mission-Critical Applications

Large language models (LLMs) show improvements on standardized benchmarks, but Ed Zitron and Steven Bartlett highlight that these advances do not translate into effective reliability for mission-critical tasks. Hallucination rates, or the generation of plausible but factually incorrect or nonsensical outputs, remain a core problem. According to historical data cited by Bartlett, hallucination rates for simple, well-defined tasks have dropped dramatically from around 21.8% four years ago to 0.7% in top frontier models like Gemini and ChatGPT. However, these low rates only apply to such basic tasks—complex or nuanced real-world applications still suffer from much higher error rates, often making these models unsuitable for high-stakes use cases such as transcribing medical information or operating financial systems.

The issues are compounded by the underlying economics of token-based pricing, as Zitron explains. Every query incurs a cost, whether the AI delivers a correct answer or an erroneous one. Especially for enterprise customers—who now often pay per million tokens—hallucinations and resulting errors lead to unnecessary expenses. Subscription models have somewhat hidden these inefficiencies, but direct token-based charges reveal how costly errors can be, drastically affecting perceived value and making it hard to justify spending for ambiguous results.

Due to LLM unpredictability, human oversight remains essential. Users are forced to build intricate systems of prompts, model selection, and verification steps—what Zitron likens to "Pee Wee's breakfast machine"—to get dependable results. This heavy need for orchestration and oversight negates much of the promised simplicity and productivity gains of intelligent systems. Rather than AI handling tasks autonomously, the user must constantly watch for mistakes, often putting in as much or more effort than traditional workflows.

Generative AI Is Limited To Content Creation and Coding Help, Lacking Autonomous Decision-Making

Despite bold public promises about AI’s transformative power, LLMs principally excel only at standard content generation, summarization, and code completion. Zitron makes clear that these models are essentially advanced pattern matchers, not reasoning agents. Their strength is situated in summarizing large volumes of text, generating new documents in the style of their inputs, and offering code suggestions. Even the workflow automation or "agentic AI" described in industry hype is not autonomous intelligence, but rather a layered construction of brittle and unreliable subsystems working together.

Attempts to go beyond these capabilities reveal further limitations. Video generation, a much-hyped field, still faces major obstacles. Zitron points out that state-of-the-art video generators struggle to produce consistent, photorealistic content suitable for film or media production. The effort required to produce even a short, compelling AI-generated video is enormous, and claims of recent breakthroughs—such as a supposed Cannes Film Festival screening—are frequently exaggerated or misleading. The improvements in quality are steady but linear, not exponential, and researchers including Gary Marcus have noted that practical utility begins to plateau as diminishing returns set in.

Software Quality and Stability Decline as Companies Rely On AI-assisted Coding Without Proven Productivity Gains

The rapid adoption of AI tools for software development has begun to ...

Ed Zitron outlines a scenario in which the collapse of the AI bubble begins with OpenAI’s financial exhaustion, triggering a chain reaction throughout the tech and broader economic landscape. The argument rests on the unsustainability of current AI valuations, the fragility of huge tech [restricted term], and the overexposure of retail investments to an artificial growth narrative.

Openai to Exhaust Funds By 2027, Causing Tech and Economic Collapse

Openai's Public Offering Delayed To 2027+ Amid Unsustainable Losses Revealed by Cfo Sarah Fryer

OpenAI, previously a private company, planned to go public this year, but the timeline has slipped to 2027 or later, according to CFO Sarah Fryer. When OpenAI sought a $1 trillion valuation for its IPO, advisors rejected this as unrealistic, especially after the company’s audited financials revealed unsustainable losses. OpenAI needs “perpetual amounts of money”—having raised $122 billion recently, with projections showing at least $100 billion more needed annually for survival. If it cannot go public, raising further funds at similar valuations will be extremely difficult, very likely forcing OpenAI to accept flat or even dilutive funding rounds. Zitron highlights Amazon’s $35 billion commitment to OpenAI, reportedly made contingent on an early IPO, as evidence of the company’s desperate need for cash.

Openai's $1 Trillion Ipo Deemed Unrealistic; Faces Faster-Growing Anthropic With Better Unit Economics, Likely Making Openai the Ipo Loser

Zitron argues that OpenAI is at the “catastrophe center” because its main rival, Anthropic, is positioned to go public first. Anthropic, though also unprofitable, has better business fundamentals and is growing faster than OpenAI. If Anthropic goes public ahead of OpenAI, it will be nearly impossible for OpenAI to match those valuations, especially while maintaining its current economic model, and could “get savage[d]” if it rushes the IPO process. The comparison is drawn to previous failures like WeWork.

Softbank's $100b Openai Stake Unusable as Collateral Without Ipo

SoftBank holds about $100 billion in OpenAI stock, but without an IPO, this stake becomes practically useless—it can’t be sold or used as collateral. OpenAI’s failure to go public would thus harm SoftBank’s liquidity and its very ability to operate, since its business model depends on regularly liquidating investments.

Anthropic's Growth and Business Model Suggest It Will Go Public First, Preventing Openai From Achieving Similar Valuations, Forcing It to Accept Dilutive Funding or Risk Collapse

Zitron asserts that Anthropic is likely to beat OpenAI to the public market, which would set a ceiling on AI company valuations and make it almost impossible for OpenAI to replicate those numbers. Without IPO funds, OpenAI would have to accept highly dilutive funding to survive or face collapse.

Ai Failures to Force Microsoft, Amazon, Google to Revise Earnings and Admit [restricted term] Spending Shortfalls

Stock Rise Driven by Ai Speculation, Not Real Revenue due to Growth Saturation in Current Businesses

Zitron says the meteoric rise in valuations for Microsoft, Amazon, Google, and Meta has been driven by AI speculation and not real, durable revenue stemming from new products. Their core businesses are saturated, so sustained market optimism increasingly depends on perpetual growth—a scenario which cannot deliver.

Amazon's $55b Cloud Commitment to Openai; Similar Deals by Microsoft, Google in Ai-dependent Gpu Centers

Major tech companies like Amazon, Google, and Microsoft have committed vast sums—Amazon alone has pledged $55 billion—to AI projects and GPU-centered cloud deals with players like OpenAI. These expenditures rest on continued growth and AI-driven demand for data centers.

Ai Labs Deplete Funds, Gpu Demand Collapses, Nvidia's Revenue Falls, "Magnificent Seven" Stock Narrative Crumbles

Once AI labs like OpenAI and Anthropic deplete their funds and data-center demand stalls, GPU purchases collapse. This triggers a severe drop in Nvidia’s revenue, destabilizing the entire “Magnificent Seven” tech stock narrative that props up the wider market. Oracle’s multi-billion-dollar data center buildout for OpenAI would become unsustainable, threatening its own stagnating business.

Meta, Apple, and Tesla, Key Market Giants, Aren't Tied to Ai Infrastructure Spending, Leaving Aging Tech Companies Without Growth Stories to Sustain the Market

Amongst the largest US companies, only Meta, Apple, and Tesla have significant market share without direct exposure to the AI infrastructure bubble. This leaves the likes of Microsoft, Amazon, and Google strug ...

Steven Bartlett and Ed Zitron compare the current artificial intelligence (AI) frenzy with the dot-com bubble of the late 1990s, noting fundamental differences in scale, speed, economic foundation, and social effects.

Ai Bubble Fundamentally Differs From Dot-com Bubble in Scale, Speed, and Lack of Viable Replacements

During the dot-com bubble, there was rampant hype and overinvestment. Bartlett recounts how companies touted exaggerated capabilities, and 90% of them failed after the crash. Skeptics such as Paul Krugman and Clifford Stoll doubted the long-term impact of the internet, and Stoll noted early on that the proliferation of bad information online could harm society. The bubble was, in fact, two separate manias: the “website bubble,” which produced myriad useless sites, and the “dark fibre” bubble, in which massive, expensive fiber optic infrastructure was built in anticipation of explosive demand.

Yet, after the collapse, the fiber optic cables—initially seen as wasted investment—became essential as internet use grew. The capital spent on infrastructure during the dot-com era laid the foundation for the real internet revolution, birthing generational companies like Amazon and Oracle.

By contrast, Zitron believes the AI bubble is fundamentally different because its market demand is predominantly subsidized, not organic. Unlike fiber optics, which later served a vital purpose, the GPU-powered data centers now being built for generative AI will have little secondary value if demand for generative AI collapses; the capital investment is unrecoverable. Microsoft and other tech giants invest in these facilities, but the real-world demand for the expensive AI software remains limited, allowing even the largest cloud providers to capture only single-digit billions in revenue, which does not justify the massive capital outlay.

AI also differs in adoption. The spread of the internet in the 1990s required an immense logistical effort to physically connect homes. In contrast, generative AI deployment simply requires a web browser, making technical integration trivial—but sustained by forced integration and constant subsidy, not organic adoption. Zitron highlights that, in many companies, employees are pressured to “AI wash” their outputs or risk professional consequences, a form of coerced adoption not seen with the internet. Unlike the gradual, voluntary uptake of internet tools, today employees may be required to claim increased AI productivity on threat of losing their jobs.

Furthermore, Zitron asserts AI adoption is propped up by the largest marketing campaign in history. This pressure to integrate, underwritten by job threats and the “AI or else” mentality from managers, far outpaces the internet era’s organic enthusiasm.

Ai Bubble's Danger: Combines Capital Destruction With Societal Harms Unlike the Dot-com Bubble

The dangers of the AI bubble go beyond investor losses. Zitron details several forms of societal harm unique to the current boom. Gas turbine-powered data centers are frequently sited in low-income neighborhoods, bringing noise and environmental degradation. This infrastructure, unlike the fiber cables of the dot-com era, directly harms the communities that host it.

Another major danger is the impact on the quality of information onlin ...

AI’s explosive growth is leaving deep, uneven marks on society and the environment, from pollution and infrastructure strain to wealth concentration and the erosion of public trust.

Data Center Growth Harms Communities Via Pollution, Power Use, Infrastructure Strain

Gas turbines powering AI data centers create significant noise and air pollution in communities such as Louisiana and Vineland, New Jersey. Ed Zitron describes the use of gas turbines as disgraceful, noting their “reckless” impact on neighborhoods that overwhelmingly oppose their construction. Despite resident resistance, local planning boards continue to approve these projects, often amid suspicions of cozy relationships with developers.

Data center projects are happening at a massive scale and with a hunger for power. For example, OpenAI and Oracle's Stargate Abilene project in Texas will cram 1.2 gigawatts of power—more electricity than the city of Bristol uses in a year—into a facility occupying a fraction of Bristol's physical space. This concentration of power demand strains local electrical grids and results in higher power bills for residents.

The allocation of public and private investment to data center expansion is also criticized. Rather than funding crucial societal needs such as poverty alleviation, healthcare, or education, capital is diverted into speculative infrastructure for AI. Ed Zitron describes this as the “directionless arrogance of capitalism,” with resources funneled into tech “growth” at the expense of labor and community well-being. Ultimately, regular people bear the costs—through pollution, higher bills, and infrastructure stress—while most benefits accrue to tech executives and venture capitalists.

Ai Industry Reveals Stark Wealth Inequality As Ai Ventures Surpass Productive Businesses in Capital Access

Ed Zitron highlights a gaping divide in how capital is accessed in today’s economy. Profitable, traditional businesses routinely struggle to secure loans from banks, facing insurmountable scrutiny and demanding requirements. In contrast, AI companies—regardless of profitability—are showered with billions by venture capitalists and hardware vendors.

Zitron gives the example of Coreweave, a company operating AI data centers, which easily secured a $1.3 billion contract to rent GPUs (graphics processing units), using agreements with hardware sellers to validate additional debt. This circular logic allows companies like NeoCloud and Coreweave to access immense funding with minimal oversight, simply by showing intent to buy more AI hardware. By comparison, ordinary people attempting to get a modest loan or mortgage are subjected to intense vetting and high interest—even as stable, productive businesses struggle for basic financing.

The system thus rewards speculation and rapid, sometimes unprofitable, technological scaling. It punishes steady, profitable entrepreneurship and the broader society, all while multiplying returns to venture capital and tech giants.

Misleading Through Exaggeration and False Timelines Erodes Trust in Institutions and Media

AI companies and the media contribute to anxiety and public confusion through exaggerations and misinformation. Zitron points to widespread, m ...