The Information Economy in Collapse: Profit Motivated Journalism, Public Relations, and the Anatomy of a $2.5 Trillion Speculative Cycle

On the structural relationship between the decline of accountability journalism, the rise of the influence industry, and the conditions that produced the largest technology investment bubble in recorded history

Clear Lake Coffee Roasters · Clear Lake, Iowa · July 2026

---

There is a newspaper that used to cover this town.

It covered city council meetings. It covered the school board. It covered plant closings, hospital consolidations, flooding, and the slow decisions that shape a community's life over decades. It employed people whose job was to attend things, ask questions, write down the answers, and publish them — creating the public record of a place's choices and their consequences.

It is gone. In its place: a website updated with press releases, one overworked journalist covering four counties, and an information vacuum that has been filled, as nature abhores a vacuum, by people with a profit motive, something to sell.

Right now what they are selling is large language models, also known as,  artificial intelligence. And the price tag — $2.5 trillion in projected capital expenditure for 2026 alone — makes every previous technology bubble look like a rounding error.

This is the story of how those two things are connected.

---

I. The Structural Decline of Local Journalism: Scale, Cause, and Consequence

American newspaper newsrooms shed approximately 57 percent of their employees between 2008 and 2023. From roughly 71,000 newsroom jobs to under 31,000. Local television news followed: down 13 percent over the same period, accelerating through 2025. Profit motivated digital outlets that promised to fill the gap have, with a handful of exceptions, either failed, pivoted to entertainment, or been absorbed by the same consolidation forces that destroyed long-form, enterprise, accountability journalism .

The Pew Research Center's most recent data documented more than 200 American counties with no local news outlet of any kind — news deserts, the researchers call them, though the word desert implies something natural and geological when the reality is something deliberately produced by market forces and policy choices that made it more profitable to strip journalism than to sustain it. A further 1,500 counties had a single remaining outlet, typically a weekly paper operating with minimal staff.

What does the absence of local journalism actually cost? The research is clear. Academic studies comparing municipalities before and after newspaper closures have documented: higher municipal borrowing costs, as bond markets price in the loss of financial accountability. Measurable increases in local government corruption. Declining voter participation in local elections. Unchecked school board decisions, i.e. 'book bans'. Police misconduct that goes unreported. Zoning decisions that transfer public wealth to private developers without scrutiny.

The social utility of journalism is a measurable public good. Its destruction is a measurable public harm. The communities that lost their newspapers are poorer, less accountable, and less democratic in ways that compound over time.

---

II. The Ascendancy of Public Relations and the Inversion of the Information Economy

In 1980, there was approximately one public relations professional for every working journalist in the United States. By 2024, that ratio had inverted to roughly six PR professionals for every journalist — and that figure excludes the vastly larger ecosystem of marketing strategists, brand managers, content creators, social media managers, communications directors, and the sprawling influence advertising apparatus that has grown around the PR core.

The Bureau of Labor Statistics counted approximately 92,000 working journalists in 2024. Against that: 320,000 PR specialists, and by some estimates more than 800,000 people employed across the broader marketing and communications sector. The number of people employed to find true things and tell you about them has collapsed. The number of people employed to tell you things on behalf of institutions with a financial interest in your belief has exploded.

This is not coincidence. It is a market clearing. As journalism contracted, the communications industry expanded into the vacuum. The press release — once a tool for attracting a journalist's attention, filtered through editorial judgment before reaching a reader — became the content itself. "News" about a company is now typically drafted by that company's communications team, distributed through wire services that charge for placement and employ no editors, published by websites that replaced their reporters with content aggregators, and amplified by social media algorithms that reward engagement volume over accuracy.

The result is an information environment in which the quantity of available information is without historical precedent and the proportion of it that has been subjected to adversarial scrutiny is vanishingly small. In addition to this 'firehose of shit' the attention economy deploys every possible trick to keep us endlessly scrolling throughout our waking hours. Our attention spans are diminishing, at the same time literacy rates are falling, as well as reading comprehension levels for the average adult in the United States at or below grade six. All this compunded by an unalloyed increase in misinformation and disinformation by State and non-state actors.  

Into this environment, in November 2022, came the largest and most consequential marketing campaign in the history of technology.

---

III. The Speculative Cycle: Capital Allocation, Promotional Culture, and the Absence of Scrutiny

The press coverage of OpenAI's release of ChatGPT in late 2022 was, in the information environment we have just described, essentially unmediated by the kind of scrutiny that a healthy journalism ecosystem would apply to an extraordinary claim. As the saying goes, extraordinary claims require extraordinary evidence.  The profit motivated technology press — consolidated, understaffed, increasingly dependent on the access and advertising relationships of the companies it covers — responded with coverage that read, in many cases, as barely reformatted promotional material. The financial press covered the investment story rather than the technology story. The basic questions a properly functioning press would foreground — what are the documented failure rates? What does the unit economics look like at scale? Who is actually generating the revenue figures being cited, and what precisely are they purchasing? — were largely unasked, and where asked, largely unanswered.

What filled the vacuum was, as we have established, marketing puffery.

The AI marketing apparatus in this current iteration that has operated since, at least late 2022 is without precedent in the history of technology hype cycles — and that history includes the dot-com bubble, the blockchain, 'crypto' and NFT cycle, the Big Data era (WEB 3.0, Internet of things; i.e. IOT) 'quantam computers' and the metaverse. What distinguishes the current cycle is the scale of capital commitment, the ubiquity, sophistication of the communications operation sustaining it, and a specific, structurally peculiar mechanism by which the major players have managed to generate the appearance of validated commercial returns.

By 2026, global capital expenditure on AI infrastructure — GPU clusters, data centers, power systems, networking, cooling — is projected to reach $2.5 trillion. Against this: the revenues generated by AI products and services in 2025 and 2026 amount to tens of billions of dollars across the entire sector. The ratio of capital deployed to revenue generated is, by any conventional investment metric, extraordinary. To put it in the language of a business owner in Clear Lake, Iowa: if we spent $250,000 building out our roastery and generated $2,500 in annual revenue, we would not describe this as a promising investment awaiting the inflection point.

---

IV. Circular Financing and Structural Overvaluation: The Doctorow Analysis

Cory Doctorow — the writer, technologist, and one of the most consistently accurate critics of the technology industry over the past two decades — has identified the specific mechanism by which the revenue figures sustaining AI valuations have been constructed.

The handful of companies that dominate AI development and deployment — Microsoft, Google, Amazon, Meta, Cisco and a smaller number of specialized players, e.g. Coreweave — are simultaneously the primary customers for Nvidia's GPU hardware and significant cross-investors in each other's AI ventures. Microsoft invested $13 billion in OpenAI. Google invested in Anthropic. Amazon invested in Anthropic. These investments create a structure in which the revenue flowing to Nvidia comes largely from this same small cluster, and the revenue these companies report from AI services flows substantially back through the same interconnected patronage network.

Doctorow's point is not that anyone has necessarily committed fraud in a technical, legal sense, at least this has not been uncovered, as of yet — though questions about the accounting treatment of these circular relationships have been raised by analysts and are worth the scrutiny that a functioning financial press might apply to them. His point is structural: a significant portion of the AI revenues being cited to justify the capital deployment represents money circulating within a closed loop of interconnected corporate entities rather than genuine, independent, end-user demand at prices that would sustain the projected returns.

The tulip bulb changes hands between wealthy Amsterdam merchants in the 17th century. Each transaction establishes a new, all time high, price. The price keeps rising because each participant needs it to rise in order to justify what they paid. This continues until someone actually needs to sell, at which point the question of what a tulip bulb is worth to someone who simply wants to grow a tulip becomes suddenly, painfully relevant.

Doctorow's further contribution, laid out across his writing and his books including The Internet Con and Picks and Shovels, is about the nature of the technology itself. Large language models are, at their technical core, sophisticated statistical pattern-matching systems. They are trained on vast quantities of human-generated text and produce outputs that resemble that text. They are not reasoning systems. They do not have knowledge, beliefs, or understanding in any sense those words carry when applied to human cognition. These systems will never become intelligent. Slime mold has more material attributes of intelligence than any LLM will ever produce.  The much-discussed aspiration toward "recursive self-improvement" — the idea that AI systems could iteratively improve themselves toward superintelligence without bound — is, Doctorow and a growing number of technical critics argue, not a roadmap toward a near-term technological reality but a category error: confusing the ability to generate plausible sounding, probabalistic text chains about a concept with the ability to implement that concept.

The systems hallucinate — producing confident, fluent, entirely fabricated statements — at rates, in some case studies,higher than 39% of the time, that make them unreliable for high-stakes autonomous applications without expensive, continuous human verification. The applications that would justify the capital investment are precisely the ones where hallucination is most costly. The applications where the systems perform reliably are ones where the value generated does not support the infrastructure cost at scale in a competitive market.

---

V. Independent Critical Analysis: Zitron, Shearer, and the Accountability Gap

Ed Zitron has written, with increasing precision and increasingly validated accuracy, about the structural gap between LLM industry claims and LLM industry reality. His newsletter, Where's Your Ed At, has documented in detail: the percentage of enterprise LLM deployments failing to demonstrate positive return on investment; the growing evidence that early adopters are abandoning LLM subscriptions and contracts after initial novelty diminishes; the mounting costs of hallucination in enterprise deployment; and the specific, uncomfortable arithmetic of inference costs of frontier models versus the prices users will pay for LLM outputs in a competitive commodity market.

Zitron's essential point is straightforward and requires no technical expertise to evaluate — only honest arithmetic and the willingness to ask the question that the financial press has largely declined to ask: does the business model work? The cost of training and running frontier large language models at the scale required by current valuations is not recoverable, in any market, at the prices users will actually pay for the outputs. The pricing power needed to make the numbers work requires a degree of product differentiation that does not exist across an industry where the major players are producing increasingly similar outputs from increasingly similar architectures.

Harry Shearer — whose Le Show has documented media dysfunction and institutional self-dealing for decades with both precision and the kind of persistent, patient mordancy that the current moment particularly rewards — has made a related observation about the role of financial media in sustaining the current cycle. The structural forces that hollowed out local and investigative, accountability journalism have simultaneously captured financial journalism: access dependence, advertiser relationships with covered companies, the replacement of experienced beat reporters with generalists who cannot independently evaluate a technical claim or a financial projection.

The functional result is a financial press that serves, in large measure, as a distribution mechanism for investor relations departments. Nvidia reports record revenues; the figure is reported as newsworthy, without sustained examination of who is generating those revenues, through what relationships, at what effective margins, and with what prospects of continuation if the circular investment structure unwinds.

---

VI. Existential Risk Discourse as a Capital Allocation Strategy

There is a specific phenomenon of the current LLM moment that requires direct and uncomfortable examination: the use of existential risk discourse as a capital allocation strategy.

In recent years, researchers, executives, and prominent voices at major AI companies — OpenAI, Anthropic, DeepMind, and others — have issued increasingly bombastic, dramatic public statements about the potential for LLM systems to cause civilizational harm. The language escalates: "extinction-level risk," "loss of human control," "the most consequential technology in human history," "a potential crime against humanity." The statements are made in op-eds, in Congressional testimony, in podcast appearances that generate enormous press coverage in an information environment hungry for dramatic narratives.

These statements share several consistent features. They are made by people who have spent their careers building the very technology they describe as potentially catastrophic. They are made without concrete technical prescriptions for how the specific risk might be mitigated — the mechanisms of the feared catastrophe are described in general terms, while the pathways to prevention are left vague. And they converge, with remarkable consistency, on a single actionable conclusion: that the organizations making the warning require substantially more resources — more capital, more compute, more researchers, more time — to conduct the safety research that might address the risk, 'otherwise China will beat us to it.'

The prescription for avoiding the catastrophe is to accelerate the work producing it, under the supervision of the people currently producing it, funded by a further enormous transfer of capital to those same people, after all their paychecks and stock options require a near endless flow of capital allocation prior to their exit strategy of the initial public offering (IPO). Don't worry about the fact that Chinese computer engineers can produce similiar or superior models for pennies on the dollar. 

But the functional effect of the "doomer" discourse — in this specific information environment, in the absence of a journalism infrastructure capable of applying sustained adversarial scrutiny to both the technical claims and the financial interests of those making them — is to crowd out the more mundane but more immediately consequential questions. Does the product work reliably? A: of course not, it never will! Who is paying for it, and what are they receiving? What does the business model look like at maturity without the subsidy of circular investment? What is the honest account of where the technology actually is, as opposed to where the promotional materials say it is heading?

The extinction warning is, among other things, the most effective possible distraction from those questions. If the stakes are civilizational, quarterly revenue figures seem almost impolite to discuss.

---

VII. What It Costs to Burn $2.5 Trillion

We should be concrete about the scale of this.

$2.5 trillion in 2026 capital expenditure. To put this in context: the entire GDP of France is approximately $3.1 trillion. The United States spends approximately $900 billion annually on its entire national defense budget. The global pharmaceutical industry spends approximately $260 billion on research and development per year.

Against this capital deployment: LLM revenues across the entire sector — subscriptions, enterprise contracts, API usage, cloud AI services — of tens of billions of dollars. The ratio is, in historical terms, without precedent for a technology in commercial deployment rather than in early research. The dot-com bubble at its peak involved approximately $5 trillion in market capitalization losses, but the capital actually deployed into internet infrastructure was a fraction of what is currently being committed to AI data centers and chip manufacturing.

The power consumption alone is staggering. The data centers required for frontier AI training and inference are projected to consume, by 2030, approximately 9 percent of total US electricity generation — more than the entire current residential sector of several major states. The water consumption for cooling these facilities runs to billions of gallons annually. The rare earth mineral requirements for GPU manufacturing are creating supply chain pressures with their own geopolitical and environmental consequences.

These are real costs — physical, infrastructural, environmental — being incurred today, in expectation of returns that have been consistently described as eighteen months away since the cycle began.

VII.I The Convergence: When the Speculative Cycle Meets the Real World

This post was substantially complete when the Guardian published, on September 20, 2026, a piece whose headline asked a question that would have seemed alarmist eighteen months ago and reads today as a straightforward empirical inquiry: Are Global Stock Markets Heading for a Crash?

The article's subtitle is worth quoting in full, because it names, in fourteen words, the convergence of every structural risk we have been examining: "Economies thrown into renewed turmoil as AI debt, Iran war and soaring government bond yields fuel alarm."

AI debt. That phrase — used without quotation marks, without qualification, as ordinary financial journalism — marks a significant shift in how the mainstream press is beginning to describe the $2.5 trillion capital commitment we have analyzed throughout this piece. The money is no longer simply "AI investment" or "AI capex." It is, increasingly, in the language of analysts and now in the language of The Guardian, debt — borrowed capital deployed against projected returns that have not materialized and are, as the cycle matures, looking progressively less certain to materialize on any timeline that the debt's terms can accommodate.

The Guardian reported that at the height of the summer, the mood was optimistic in the world's financial capitals — powered by the AI revolution, the US stock market had rallied to a fresh all-time high, as investors bet the multitrillion-dollar investment spree would overshadow the hit from the Iran war. That sentence encapsulates the information environment we have described with precision: the AI narrative was functioning, at its peak, as a story powerful enough to override geopolitical crisis in the calculus of global equity markets. The marketing had worked. The tulip was at its highest price.

Now, the Guardian reports, the warning lights are flashing red. As the fighting in the Middle East intensifies without clear sign of a resolution, financial markets have been thrown into renewed turmoil. A slowdown looms in the AI arms race, and tinderbox conditions in the market for government debt are fuelling alarm.

The specific market mechanics are stark. In the past week the US government's borrowing costs climbed to the highest level since 2007, with knock-on consequences for the finances of households, businesses, and other governments worldwide. The 10-year Treasury yield crossing 5 percent is not an abstract financial data point. It is the price of every mortgage, every business loan, every municipal bond in the country repricing upward simultaneously. It is, in the language of an Iowa business owner, the cost of the money going up at the exact moment the revenues from the thing you borrowed it for are failing to meet projections.

With the S&P 500 index of leading US companies 3% below an all-time high, and a combined value of more than $20 trillion for the "magnificent seven" tech stocks — Nvidia, Apple, Google, Microsoft, Meta, Amazon and Tesla — the concern is that markets are overextended just as the storm clouds gather for the world economy. The CAPE ratio — the cyclically adjusted price-to-earnings measure that Warren Buffett and others use as a long-term valuation indicator — is, according to market analysts cited in reporting around this period, at its highest level since the dot-com peak of 2000. The Magnificent Seven, which constitute a historically unprecedented share of S&P 500 market capitalization and whose valuations rest substantially on AI return projections, are the specific concentration risk that the "overextended" characterization names.

As Albert Edwards, senior analyst at Société Générale, wrote in a note to clients and quoted by the Guardian: "These are febrile times. The key worry for investors and policymakers alike is the extent to which the current oil price 'shock' will ripple through the global economy and whether it will necessitate sharply higher, recession-inducing, interest rates."

The word febrile is carefully chosen. It means feverish — the condition of a system running hotter than it can sustain, where the heat itself is a symptom of the underlying pathology rather than a sign of productive energy. Febrile markets are not strong markets. They are markets in which the temperature has become the story, obscuring the question of what is actually wrong.

What is wrong, we would argue, is exactly what we have been describing throughout this piece: a speculative cycle built on circular financing, amplified by a marketing apparatus that replaced the journalism capable of scrutinizing it, sustained by existential risk discourse that redirected public attention from financial accountability to philosophical speculation, and now encountering the physical constraints — rising interest rates, geopolitical instability, energy costs, and the simple arithmetic of capital deployed versus returns generated — that no promotional campaign can indefinitely postpone.

The Guardian's question — are global stock markets heading for a crash? — is, at its core, a question about whether the information environment will correct before or after the financial one. In a healthy information ecosystem, the scrutiny would have arrived years ago, modulating the speculative excess before it reached systemic scale. In the information environment we have described — hollowed journalism, ascendant PR, financial media serving as investor relations distribution — the scrutiny arrives late, in the form of a headline asking whether the crash is coming rather than why the bubble was allowed to inflate to this scale without adequate examination.

The answer to the Guardian's question is, in our view, not the most important question. The most important question is the one that should have been asked three years and $2.5 trillion ago: by whom, precisely, are these returns expected to be generated, at what price, through what mechanism, and on what timeline? Those questions were available to be asked. The journalists who might have asked them were not, in sufficient numbers, available to ask them.

That is the connection this piece has been building toward. And the Guardian's headline, arriving on September 20, 2026, is the bill coming due.

---

VIII. The Connection

Return to Clear Lake. Return to the hypothetical journalists who are gone.

They would have covered the data center permitted on the edge of town — one of the facilities housing the infrastructure for the LLM systems we have been discussing. They would have asked about the power consumption figures, the water usage, the property tax abatements schemes, the temp jobs 'created', the wages those jobs pay, and who bears the costs of the infrastructure required to support the facility. They would have written a story that created a public record and enabled democratic accountability for the decision.

That story was not written. The data center's communications team issued a press release describing the economic benefits of the investment. It was published, largely verbatim, on the local news website. The questions were not asked.

This is the deepest connection between the journalism crisis and the LLM bubble: the same structural forces that destroyed the information environment capable of scrutinizing the bubble created the conditions in which the bubble could inflate unchecked. The PR apparatus that replaced journalism is the apparatus generating and sustaining the hype. The financial media that lost its investigative capacity is the media reporting circular revenues without scrutiny. The information vacuum left by the collapse of accountability journalism is the space in which $2.5 trillion in capital has been misallocated on the basis of promotional materials, circular investment relationships, and extinction warnings issued by the people most financially positioned to benefit from them.

The analogous tulip is very beautiful. The merchant is very persuasive. The journalist who might have asked what a tulip is actually worth, in a field full of tulips, in a country full of tulip merchants all selling to each other — Their entire newsroom was laid off in 2019.

The journalists from that newsroom, most of them, anyway, got a pay bump and all of them are in marketing communications and / or public relations now.

The silence they left behind is the most expensive silence in history. We are all paying for it with higher utility costs, persistent inflation, misallocation of capital, crumbling infrastructure, climate breakdown and the bill has not yet come due. Once this bubble pops we shall see the pretender emperor, behind the curtain, has no clothes; if you permit me these mixed metaphors ( I consulted an LLM.) 

---

Clear Lake Coffee Roasters LLC · 15068 Hill St · Clear Lake, Iowa 50428 hello@clearlakecoffeeroasters.com · clearlakecoffeeroasters.com

Sources and further reading: Cory Doctorow, The Internet Con (Verso, 2023); Picks and Shovels (Tor Books, 2025); Ed Zitron, Where's Your Ed At (wheresyoured.at); Harry Shearer, Le Show (KCRW, 1983–present); Pew Research Center, State of the News Media 2024–2025; Northwestern University Medill Local News Initiative, News Desert Report 2024; Bureau of Labor Statistics, Occupational Employment and Wage Statistics; Goldman Sachs, "AI Capex and the Return Question" (2024); Gary Marcus and Ernest Davis, Rebooting AI (2019); Emily Bender, Timnit Gebru, et al., "On the Dangers of Stochastic Parrots" (FAccT 2021); Penelope Muse Abernathy, "The State of Local News 2023," Northwestern Medill. The authors note, with the transparency this subject demands, that this post was drafted with assistance from an Anthropic large language model — a tool whose specific, bounded utility we find genuine, and whose industry's capital economics, promotional culture, and structural incentives we have nonetheless attempted to examine honestly and without deference to our own tool use.

 

 Six reasons for making Clear Lake Coffee Roasters - CLCR - your go-to coffee roaster:


☕️ We are a local family-run business located in the heart of Clear Lake, Iowa.

☕️ We go to great lengths to find only the finest and ethically sourced coffee around, from the top 2% of coffee beans in the world.

☕️ We only source 100% certified Arabica coffee beans, carefully hand-selecting each coffee based on specific quality and taste attributes.

☕️ Our roasting process has been refined over the years and each roast profile is individually designed to complement the nuances of the coffee we source, from Cup of Excellence (COE) award-winning producers.

☕️ By roasting in smaller batches, we can ensure our coffee is ALWAYS fresh, in fact, we roast your coffee only after you place an order - the same day your order ships out.

☕️ At CLCR, we are dedicated to a single mission: the unyielding pursuit of coffee perfection in every cup.

We would give you more reasons, but rather than reading it's better if you visit our website, purchase a bag or two, and experience a unique caffeinated or half-caff journey for yourself 😊!
Explore goodness. Click. Buy. Smile.

Clear Lake City Park Beach - Named one of USA Today's top 50 Beaches in the United States 2025

---