Back to Home
The AI Gold Rush: A Retail Investor's Guide to Navigating the Hype
October 26, 2025 — Your essential guide to understanding AI investments
Welcome to the Age of AI Investment Mania
Artificial Intelligence dominates headlines, captivates the imagination, and moves trillions of dollars in global financial markets. Technology giants and fledgling startups alike are pouring unprecedented sums into building the future of AI, with hundreds of billions in capital expenditure announced for the coming year alone. As a retail investor, it's natural to feel urgency and a fear of missing out (FOMO) on what could be the next great technological revolution.
This feeling is not new. It's a recurring theme in financial history, from the railway manias of the 19th century to the dot-com bubble of the late 1990s. In every cycle, a transformative technology promises to remake the world, and in every cycle, fortunes are both made and lost in the ensuing speculative frenzy.
Our Mission: Cutting Through the Noise
This special report aims to cut through the hype and provide you with clear, actionable intelligence. We will delve into what AI, in its current form, is truly providing to society. We will scrutinize the astronomical costs associated with this technological arms race and ask the hard questions about profitability and realistic return on investment.
Is the current AI boom a sustainable wave of innovation, or is it, as some astute observers have suggested, the largest and most dangerous financial bubble in history? Our goal is to equip you, the individual investor, with the knowledge to distinguish between the dazzling mirage of hype and the solid ground of tangible value.
"The greatest risk in investing is not taking calculated risks, but rather succumbing to the emotional pull of narratives detached from financial reality."
The Picks and Shovels Strategy
Identify the Gold Diggers
Recognize who is frantically digging for AI gold with unsustainable business models and burning through capital at alarming rates.
Find the Tool Sellers
Discover the companies providing essential infrastructure—the digital picks and shovels—that profit regardless of whether the miners strike gold.
Make Informed Decisions
Apply this time-tested investment philosophy to navigate the AI boom without exposing yourself to catastrophic bubble risk.
In the context of AI, we will identify who is selling the digital picks and shovels, and who is merely digging in the dirt. This classic gold rush strategy has proven profitable throughout history: when everyone is frantically digging for gold, the surest profits are often made by selling the tools to the miners.
Part I: The Sobering Reality of Modern AI
More Parrot Than Oracle
Before we can intelligently invest in AI, we must first understand what it is we are investing in. The term "Artificial Intelligence" conjures images of sentient machines and creative problem-solving, a vision heavily promoted by the companies seeking billions in investment. The current reality of the technology is far more mundane.
What AI Actually Is: Rote Learning at Scale
As macro strategist Julien Garran articulated in a recent podcast, today's AI is fundamentally a system of "rote learning." Large Language Models (LLMs) like ChatGPT are trained on vast datasets of text and code from the internet, and their primary function is to recognize patterns and regurgitate information in a statistically probable sequence.
They do not understand context, nuance, or truth in the way a human does. They are, in essence, incredibly sophisticated parrots—a concept academics have termed "stochastic parrots."

Key Limitation
LLMs cannot distinguish between true and false information. They only generate what is statistically likely based on training data patterns.
Commercial Failures of AI Technology
Legal Brief Disasters
Over 5,000 documented cases in the United States where lawyers using AI submitted legal briefs citing fabricated case law imagined by LLMs.
Deloitte's Expensive Mistake
Major consulting firm forced to repay significant fees to government client after AI-driven work was riddled with incorrect references.
Hallucination Problem
AI "hallucinations" are not a bug that can be fixed—they are an inherent characteristic of the underlying technology.
These are not isolated incidents; they are symptoms of a technology that is not yet ready for high-stakes, mission-critical commercial use. The promise of AI replacing skilled human workers remains, for now, a distant and unproven proposition.
The Copyright Time Bomb
Beyond the issue of accuracy, the very data that these models are trained on presents a legal and ethical minefield. The vast majority of the text and images used to train AI models were scraped from the internet without the permission of the original creators.
This has led to a wave of copyright lawsuits from authors, artists, and media companies whose work has been used to create a commercial product without their consent. The New York Times, Getty Images, and numerous individual artists have filed lawsuits that could have profound implications for the industry.

Investor Risk: The outcome of these legal battles could result in massive financial liabilities for AI companies and potentially force them to rebuild their models from scratch using licensed, and therefore much more expensive, data.
Part II: The Staggering Costs of the AI Gold Rush
The gap between the current capabilities of AI and its perceived potential has not stopped an unprecedented torrent of capital from flowing into the sector. The sheer scale of the investment is difficult to comprehend, but it is essential for investors to grasp these numbers to understand the financial mountain that AI companies must climb to achieve profitability.
Breaking Down the Astronomical Costs
AI Model Training: The Exponential Cost Curve
The computational cost to train a single, frontier-level AI model has been escalating at a truly staggering rate. OpenAI's GPT-4 cost an estimated $100 million to train—a figure that already seemed astronomical at the time. However, newer models are approaching half a billion dollars for a single training run, and disturbingly, these massive investments are not producing proportionate improvements in performance.
This exponential cost increase suggests we may be hitting a "scaling wall," where throwing more money and data at the problem yields diminishing returns. For investors, this is a critical signal. If the technology cannot improve without exponentially larger investments, the path to profitability becomes increasingly uncertain. The industry may be chasing a mirage, spending billions in the hope that the next breakthrough is just one more expensive training run away.
Physical Infrastructure: Building an Empire on Sand
$364B
Hyperscaler Spending 2025
Amazon, Google, Microsoft, and Meta are spending this amount annually on data centers alone.
$500B
Total AI Infrastructure 2025
The complete investment in physical hardware to power AI, including chips, servers, and buildings.
$3T
Projected by 2029
The total cumulative investment in AI infrastructure is on track to reach nearly three trillion dollars.
These figures represent not just the chips themselves, but the servers, high-speed networking equipment, cooling systems, and physical buildings required to house and operate them. This is an investment on a scale rarely seen in industrial history, and it's all predicated on the assumption that profitable AI applications will eventually emerge to justify these costs.
Energy Consumption: Powering an Insatiable Beast
Data centers are voracious consumers of electricity, and AI workloads are dramatically accelerating this trend. In 2024, US data centers consumed approximately 183 terawatt-hours of electricity—roughly 4% of the nation's total power consumption. By 2035, their power demand is expected to more than double, placing immense strain on the existing power grid.
To put this in perspective, a single AI query can use 10-15 times the electricity of a standard web search. When you multiply this by billions of queries, the numbers become staggering. This will require the construction of new power plants, potentially including nuclear facilities, just to keep the lights on in AI data centers.
GPU Economics: The Math Doesn't Work
At the heart of every AI data center is the Graphics Processing Unit (GPU). A single, top-of-the-line NVIDIA Blackwell chip costs around $50,000, with an additional $25,000 required for installation and integration into a data center. This brings the total cost per GPU to approximately $75,000.
Here's where the economics become deeply troubling: at current market rental rates of approximately $3.79 per hour, data center companies are losing money on every hour that a GPU is used. The revenue generated from renting out these expensive assets does not cover the capital costs, let alone the operational expenses of power, cooling, and maintenance.

Depreciation Crisis: With NVIDIA releasing a more powerful chip every year, these $75,000 assets depreciate rapidly, making long-term profitability even more challenging.
Part III: The Profitability Mirage
Where Are the Profits?
For an industry to be sustainable, it must eventually generate more money than it spends. In the current AI landscape, profitability is not just elusive; it appears to be a distant mirage shimmering on the horizon, always seeming close but never actually within reach.
OpenAI: The Poster Child of Unsustainable Growth
Revenue: $4.3 Billion
Generated in the first half of 2025—an impressive figure that garnered significant media attention.
Net Loss: $13.5 Billion
During that same period, the company burned through more than three times its revenue in losses.
Loss Per Dollar
For every dollar of revenue, OpenAI lost more than three dollars—a ratio that is completely unsustainable.
OpenAI, the poster child of the AI boom, generated an impressive $4.3 billion in revenue in the first half of 2025. However, during that same period, it reported a net loss of $13.5 billion. This is not a company that is close to profitability—it is a company that is burning capital at an extraordinary rate.
Industry-Wide Profit Drought
OpenAI is not an outlier. Anthropic, a major competitor and another darling of the AI investment community, lost $5.3 billion in 2024 alone. The company is now valued at an eye-watering $183 billion despite having no clear path to profitability and mounting losses that show no signs of abating.
The sobering reality is that not a single major AI model developer is currently profitable. In fact, an estimated 95% of all corporate AI pilot programs are failing to generate a return on investment. These are not just small startups struggling to find product-market fit; these are well-funded companies with access to the best talent and technology, and they still cannot figure out how to make money.
95%
AI Pilots Failing
Corporate programs with no ROI
0%
Profitable AI Developers
Major model companies
Understanding Round-Tripping Revenue
How is it possible for an industry to generate billions in revenue while losing even more billions? The answer lies in a combination of unsustainable business models and questionable accounting practices reminiscent of past financial bubbles. A key concern is the phenomenon of "round-tripping" revenue.
This is where a company invests in its own customers, who then use that money to buy the company's products, creating an illusion of organic demand. This practice was rampant during the dot-com bubble, and it appears to be making a comeback on a much grander scale. When Company A invests $100 million in Company B, and Company B then spends $100 million buying products from Company A, it looks like $100 million in revenue, but no real economic value has been created.
NVIDIA and CoreWeave: A Case Study in Circular Financing
1
NVIDIA Invests in CoreWeave
NVIDIA invested $1 billion in CoreWeave's IPO, providing the data center operator with capital.
2
CoreWeave Buys NVIDIA Chips
CoreWeave uses the investment money to purchase billions of dollars worth of NVIDIA GPUs.
3
NVIDIA Rents from CoreWeave
NVIDIA commits to renting back the capacity from CoreWeave, completing the circular flow.
4
Illusion of Revenue
Both companies report impressive revenues, but no real economic value has been created.
The relationship between NVIDIA, the dominant chip supplier, and its customers provides a textbook example of round-tripping. This creates a closed loop of money that inflates the revenues of both companies without generating any real economic value or sustainable profits.
NVIDIA's Accounts Receivable: A Red Flag
This circular flow of money is further evidenced by NVIDIA's own financial statements. The company's accounts receivable—the money owed to it by its customers—has exploded by 626% over the past 30 months. This is a far more extreme figure than the 140% increase seen in Cisco's receivables just before the dot-com crash of 2000.
What this suggests is that NVIDIA is not just selling chips; it is lending its customers the money to buy them. This is not a sign of a healthy, thriving market. It is a classic red flag of a bubble nearing its breaking point. When the music stops and the funding dries up, these IOUs will be exposed as bad debts, and the entire house of cards could come tumbling down.
"NVIDIA's accounts receivable growth of 626% is more than four times the increase Cisco experienced before the dot-com crash."
Insider Selling: A Vote of No Confidence
Adding to the concern about NVIDIA's long-term prospects is the massive wave of insider selling. While the stock has been a favorite of retail investors, company executives and long-time employees have been selling their shares at an unprecedented rate.
This is a clear signal that those who know the company best—the people with access to internal data and strategic plans—are taking their profits off the table. They are not buying more shares at these elevated prices; they are selling. This represents a vote of no confidence in the sustainability of the current business model and valuation.

What Insider Selling Means
When insiders sell aggressively, it often precedes significant price corrections. They have information that retail investors don't, and they're acting on it.
Part IV: Investment Strategy - Navigating the Minefield
Given this backdrop of immense costs, non-existent profits, and circular financing, how should a retail investor approach the AI sector? The key is to differentiate between the speculative, high-risk application layer and the tangible, essential infrastructure layer. This requires a disciplined, contrarian approach.
Overvalued & Risky: Potential Short Candidates
AI Model Developers
Private companies like OpenAI and Anthropic with sky-high valuations, no profits, and unsustainable burn rates. Their valuations influence the entire market despite having no proven business model.
Data Center Operators
Companies like CoreWeave (CRWV) that went public at $40/share and now trade at $60B+ valuation with virtually no profit. They operate with negative unit economics and depend on vendor financing.
NVIDIA (NVDA)
Despite being the AI boom's darling, massive insider selling and alarming accounts receivable growth suggest an unsustainable trajectory. Shorting is high-risk, but long-dated puts offer defined-risk exposure.
CoreWeave: A Bubble Within a Bubble
CoreWeave represents a particularly concerning case study in speculative excess. The company went public at $40 per share and now trades at a valuation of over $60 billion. This would be remarkable for a company with strong profits and a proven business model. For CoreWeave, which operates data centers with negative unit economics—meaning they lose money on every GPU they rent—this valuation is simply staggering.
The company is entirely dependent on vendor financing from NVIDIA, as discussed earlier. If that financing were to dry up, or if GPU rental rates fail to increase enough to cover costs, CoreWeave's business model would collapse. For investors with a high risk tolerance and a contrarian mindset, this represents a potential short opportunity, though it should be approached with extreme caution given the market's irrational exuberance.
The Case for Shorting or Avoiding the Hype Layer
For investors considering short positions or put options on these overvalued names, it's important to understand both the potential rewards and the substantial risks. Markets can remain irrational longer than you can remain solvent, as the saying goes. The AI hype has powerful momentum, driven by fear of missing out and a compelling narrative.
However, the fundamentals paint a clear picture: unsustainable burn rates, no path to profitability, circular financing, and insider selling. These are the classic warning signs that have preceded every major market crash in history. For those unwilling to take the risk of shorting or buying puts, simply avoiding these names and staying on the sidelines is a perfectly valid strategy.
Undervalued & Solid: The Infrastructure Play
The Picks and Shovels Strategy
In contrast to the speculative frenzy at the application layer, the infrastructure that underpins the digital world offers more grounded investment opportunities. These are the "picks and shovels" plays of the AI gold rush—the companies that profit from the boom without being exposed to the speculative risk of whether the gold is ever found.
Power and Utilities: The Certain Winner
Constellation Energy (CEG)
Leading nuclear power operator with significant capacity to meet 24/7 data center demands. Operates in regulated markets with guaranteed returns on infrastructure investments.
Vistra (VST)
Diversified power generation company with substantial nuclear and natural gas capacity. Well-positioned to benefit from the projected 120 GW of additional electricity demand by 2030.
Utility Companies
Traditional utilities with infrastructure expansion plans. They profit from increased electricity consumption regardless of whether AI models achieve commercial success.
Why Power Companies Are a Safe Bet
The one certainty in the AI boom is the immense and growing demand for electricity. Data centers require massive amounts of power to run their servers and, critically, to cool them. This demand is not speculative—it is real, tangible, and growing exponentially.
Companies with significant nuclear and natural gas capacity are particularly well-positioned because data centers require baseload power—electricity that is available 24/7, not just when the sun is shining or the wind is blowing. Nuclear and natural gas plants can provide this consistent power supply.
120%
Demand Growth by 2035
US data center electricity needs
100%
Baseload Reliability
Nuclear and gas generation
Moreover, these companies operate in regulated markets, where they often receive guaranteed returns on their investments in new power plants. The risk profile is dramatically different from the speculative AI application layer. Whether or not AI models ever become profitable, the data centers will still need power, and these companies will still generate revenue and earnings.
Semiconductor Equipment: The Monopolists
Before a chip can be put in a data center, it must be manufactured. The companies that build the complex machinery for semiconductor foundries are essential to the entire technology ecosystem. These are not speculative plays; these are highly profitable businesses with substantial competitive moats.
01
ASML Holding (ASML)
Holds a monopoly on EUV lithography machines required for the most advanced chips. This technological moat is virtually impossible for competitors to cross, providing pricing power and stable demand.
02
Applied Materials (AMAT)
Dominates different steps of the chip manufacturing process with market-leading deposition and etching equipment. Strong market position across multiple critical technologies.
03
Lam Research (LRCX)
Another key player in semiconductor manufacturing equipment with specialized capabilities. Benefits from the overall growth in computing across all applications.
Why Semiconductor Equipment Companies Win
ASML Holding deserves special attention. The company holds a monopoly on Extreme Ultraviolet (EUV) lithography machines, which are absolutely essential for manufacturing the most advanced semiconductor chips. Without ASML's machines, you cannot make cutting-edge processors. This monopoly position gives ASML tremendous pricing power.
The barriers to entry in this market are enormous. EUV lithography requires some of the most advanced physics and engineering on the planet. Even if a competitor wanted to challenge ASML, it would take decades and tens of billions of dollars in R&D with no guarantee of success. This is as close to an impregnable competitive moat as exists in the technology sector.
Applied Materials and Lam Research, while not monopolies, also hold dominant positions in their respective niches of the semiconductor manufacturing process. Crucially, these companies are already highly profitable and have strong balance sheets. They benefit from the overall growth in computing, not just the AI hype, making them more diversified and resilient investments.
Chip Manufacturers: Diversified and Essential
TSMC (Taiwan Semiconductor)
The world's largest and most advanced chip foundry. Manufactures chips for Apple, NVIDIA, AMD, and virtually every major tech company. Success not tied to single technology trend.
Samsung Electronics
Major chip manufacturer with diversified business across memory chips, processors, and consumer electronics. Provides geographic and technological diversification.
Intel Corporation
Investing heavily in US-based manufacturing capacity. Benefits from government subsidies under the CHIPS Act and strategic importance to national security.
The Geopolitical Consideration: TSMC
Taiwan Semiconductor Manufacturing Company (TSMC) is arguably the most critical company in the global technology supply chain. It manufactures the vast majority of the world's most advanced chips, including those for Apple's iPhones, NVIDIA's GPUs, and countless other applications. The company's technological lead over competitors is measured in years, not months.
However, TSMC's location in Taiwan presents a significant geopolitical risk. The potential for conflict between China and Taiwan is a real concern for investors. If Taiwan were to be invaded or blockaded, the global technology supply chain would be thrown into chaos. This is not a hypothetical scenario; it is a risk that the U.S. government and major tech companies take very seriously.

Balancing Risk and Opportunity: Despite the geopolitical risks, TSMC is so critical to the global economy that it is, in many ways, "too big to fail." Governments worldwide have a vested interest in ensuring its operations continue. For investors, this means weighing the technological advantages against the geopolitical uncertainties.
Commodities and Raw Materials: The Foundation
Copper: The Essential Conductor
Data centers and power grid expansions require vast quantities of copper for electrical wiring and power infrastructure. Global copper demand is projected to increase significantly as AI infrastructure buildout accelerates.
Lithium and Battery Metals
Energy storage solutions are needed to balance intermittent renewable energy sources. Lithium, cobalt, and nickel are essential for battery production and grid-scale energy storage.
The construction of data centers and the expansion of the power grid will require vast quantities of raw materials. These commodities provide another way to gain exposure to the infrastructure build-out without betting on the success of any particular AI application.
Commodity Investment Strategies
Mining Companies
Direct investment in producers of copper, lithium, and other essential materials with proven reserves and production capacity.
Commodity ETFs
Diversified exposure through exchange-traded funds focused on raw materials and natural resources sectors.
Futures Contracts
For sophisticated investors, commodity futures provide direct price exposure with leverage (higher risk, higher potential reward).
Copper is particularly interesting because it is essential for electrical wiring and power infrastructure. As the power grid expands to meet the demands of data centers, copper demand will increase substantially. Similarly, lithium and other battery metals are needed for energy storage solutions that will help balance the intermittent nature of renewable energy sources.
Part V: Historical Context and Lessons
Echoes of Past Bubbles
To fully appreciate the magnitude of the current AI bubble, it is instructive to look back at previous speculative manias. History does not repeat itself, but it often rhymes. The patterns we see today are eerily familiar to anyone who has studied financial history.
The Dot-Com Bubble: An Obvious Parallel
1
1995-1999: The Mania Builds
Companies with little more than a ".com" at the end of their name saw valuations soar to absurd heights. Profitability was dismissed as "old economy thinking."
2
2000: The Peak and Crash
The NASDAQ peaked in March 2000 and then proceeded to lose 78% of its value over the next two years. Hundreds of companies went bankrupt.
3
2000-2002: The Aftermath
Even companies that survived saw their stock prices collapse by 90% or more. Amazon fell from $100 to $6. Investors lost trillions.
4
Lessons Learned
The internet did revolutionize the world, but the vast majority of companies that rode the first wave went bankrupt. The survivors were exceptions, not the rule.
The Pets.com Story: A Cautionary Tale
Pets.com is perhaps the most iconic symbol of dot-com era excess. The company sold pet supplies online—a perfectly reasonable business concept. However, the economics were fundamentally flawed. They lost money on every sale due to high shipping costs for heavy bags of dog food.
Despite these terrible unit economics, the company spent millions of dollars on a Super Bowl ad featuring their now-famous sock puppet mascot. The company went public in February 2000 at $11 per share. By November 2000, just nine months later, they had completely shut down and liquidated. Investors lost everything.
The lesson is clear: a good narrative and heavy marketing cannot overcome bad business fundamentals. Eventually, the market demands profitability, and companies that cannot deliver are ruthlessly discarded.
Today's Bubble vs. Historical Precedents
According to Julien Garran's analysis using the Wicksell spread, the current misallocation of capital in the U.S. is approximately 65% of GDP, or roughly $17 trillion. This is 17 times the size of the accumulated misallocated capital at the peak of the dot-com bubble and four times the size that led to the 2008 financial crisis.
The Systemic Risk: When Bubbles Burst
These are not just large numbers on a chart; they represent a systemic risk to the entire financial system. When a bubble of this magnitude bursts—and history suggests that all bubbles eventually do—the economic fallout can be severe and far-reaching. The question is not if it will burst, but when and how catastrophic the consequences will be.
The bursting of the AI bubble could force a massive monetary and fiscal response from governments and central banks. This might include aggressive interest rate cuts, quantitative easing programs, and large fiscal stimulus packages. While these measures might prevent a complete economic collapse, they would likely come at the cost of significant currency debasement and inflation, particularly in commodities.
Part VI: The Scaling Wall - A Fundamental Barrier
Why More Money Won't Fix the Problem
One of the most critical, yet least understood, aspects of the current AI boom is the concept of the "scaling wall." For years, the AI industry operated under the assumption that simply making models bigger would lead to continuous improvements in performance. This belief has now run into a hard reality.
The Scaling Paradigm Breaks Down
GPT-3: $50 Million
The first major breakthrough in large language models, trained at significant cost but showing impressive capabilities.
GPT-4: $100 Million
Double the training cost with notable improvements in performance and reasoning ability. The scaling paradigm seemed to be working.
GPT-5: $500 Million
Five times the cost of GPT-4, but with no significant performance improvement. OpenAI delayed release multiple times.
Released as GPT-4.5
Eventually launched with only incremental gains, signaling that the scaling approach has hit a wall before achieving commercial viability.
What the Scaling Wall Means for Investors
OpenAI spent $500 million training what was supposed to be GPT-5, but the model did not show a significant improvement over GPT-4. The company delayed the release multiple times and eventually launched it as GPT-4.5 with only incremental gains. This is not just an OpenAI problem; it appears to be a fundamental limitation of the current approach to AI.
The statistical complexity of language may simply not require the massive number of parameters that these models are using. This means that throwing more money at the problem yields diminishing, or even zero, returns.

Critical Investment Insight
If the technology has hit a fundamental ceiling before achieving commercial viability, then no amount of additional investment will change the outcome. The billions being poured into training ever-larger models may simply be money down the drain.
Alternative Approaches: Will They Work?
Recognizing the scaling wall, AI companies are now exploring alternative approaches to improve model performance. These include techniques like reinforcement learning from human feedback (RLHF), chain-of-thought prompting, and various forms of model distillation. While these techniques can produce some improvements, they come with their own costs and limitations.
Reinforcement learning from human feedback, for example, requires extensive human labor to rate model outputs. This is expensive and time-consuming. Chain-of-thought prompting increases the number of tokens the model must process, which increases both latency and cost. None of these approaches has demonstrated the ability to overcome the fundamental accuracy and reliability issues that plague current AI systems.
For investors, this means that the path to profitability remains uncertain even if companies pivot to new training methodologies. The problem is not just one of scale; it may be a problem with the underlying architecture and approach.
Part VII: The Break-Even Question
How Much Capital Is Needed?
One of the most pressing questions for any investor is: what would it take for AI to become profitable? The answer, unfortunately, is deeply uncertain and potentially astronomical. Let's examine what break-even might actually look like.
OpenAI's Break-Even Math: A Sobering Calculation
OpenAI is projecting revenues of $12.7 billion for 2025, but it is also projecting losses of at least $8 billion, and likely much more when all costs are accounted for. To reach break-even, the company would need to either dramatically increase revenues or dramatically decrease costs—both of which present enormous challenges.
The Revenue Challenge: Limited Pricing Power
Increasing revenues in the AI model space is extraordinarily challenging because the company faces intense competition, including from open-source models that are free to use. When your competitor's product costs zero dollars, your pricing power is inherently limited. This is fundamentally different from traditional software, where network effects and switching costs can create powerful moats.
Users can switch between AI models with essentially no friction. If OpenAI raises prices too much, customers will simply move to Claude, Gemini, or one of the many open-source alternatives. This competitive dynamic makes it very difficult to achieve the kind of gross margins that would be necessary to offset the enormous fixed costs of model development and infrastructure.
The Cost Challenge: Everything Is Getting More Expensive
GPU Costs Rising
The cost of cutting-edge GPUs continues to increase with each generation. NVIDIA's pricing power means that hardware costs are unlikely to decline meaningfully.
Electricity Prices Increasing
As data centers strain power grids, electricity costs are rising. Some regions are seeing power costs double or triple for large industrial users.
Talent Costs Escalating
The salaries of skilled AI engineers continue to rise due to intense competition for talent. Top researchers can command multi-million dollar compensation packages.
Potential Data Licensing Costs
If companies must rebuild models using licensed data to avoid copyright lawsuits, training costs could increase by orders of magnitude.
The Capital Requirement: An Almost Impossible Sum
OpenAI has committed to a $300 billion deal with Oracle for computing capacity, though this commitment is non-binding. To meet this obligation, the company would need to raise an additional $250 billion on top of the tens of billions it has already raised. This is an almost unimaginable sum of money.
To put this in perspective, $250 billion is larger than the GDP of most countries. It's more than the market capitalization of all but a handful of the world's largest companies. Where would this capital come from? The venture capital industry, which has been a major source of funding for AI startups, is already showing signs of strain. The number of new mega-rounds has been declining, and many VCs are becoming more cautious.
"The sobering reality is that it may require hundreds of billions, or even trillions, of additional dollars to bring AI to a point of commercial viability—and even then, there is no guarantee of success."
Part VIII: Macro Implications of the Bubble
A Bubble That Could Reshape the Global Economy
The potential bursting of the AI bubble is not just a concern for tech investors; it has the potential to reshape the entire global economy. The ripple effects would be felt across markets, currencies, and geopolitical relationships.
AI's Contribution to GDP Growth
The AI boom has contributed approximately 3% to U.S. GDP growth in recent years, split between the direct investment in data centers and the wealth effect from rising tech stock prices. If this bubble deflates, that 3% boost could turn into a 3% to 6% drag on GDP, potentially tipping the U.S. economy into a recession.
The Policy Response: Aggressive Stimulus Ahead
Fed Rate Cuts
Federal Reserve would likely cut interest rates aggressively, potentially back to zero or even negative rates.
Quantitative Easing
Restart of bond-buying programs and balance sheet expansion to inject liquidity into the financial system.
Fiscal Stimulus
Large government spending packages to support employment and prevent deflationary spiral.
Dollar Debasement
These measures would likely lead to currency debasement and surge in inflation, particularly in commodities.
Capital Flows: The Great Reallocation
As U.S. assets become less attractive due to currency debasement and economic uncertainty, investors would seek opportunities elsewhere. This represents a significant shift from the disinflationary, tech-driven boom of the past decade to a more inflationary, commodity-driven environment.
Julien Garran identifies India and Vietnam as potential beneficiaries of this capital reallocation. These countries have large, emerging middle classes and are well-positioned to attract manufacturing investment as companies diversify away from China and seek stable growth markets.
The Commodity Supercycle Thesis
In this scenario, commodities—particularly those needed for infrastructure development such as copper, aluminum, and lithium—would likely see strong demand and price appreciation. This would mark a reversal from the past decade, where commodities generally underperformed financial assets.
The logic is straightforward: if governments respond to an AI bubble collapse with massive infrastructure spending and monetary stimulus, the prices of physical goods and materials will rise. Unlike financial assets, which can be created with a few keystrokes, commodities require physical extraction, processing, and transportation. Supply cannot be increased instantly, which leads to price spikes when demand surges.
300%
Potential Copper Appreciation
Historical commodity supercycles have seen increases of this magnitude over 5-7 years
500%
Potential Lithium Appreciation
Battery metal demand could drive even larger gains as electrification accelerates
Part IX: The Psychology of FOMO
The Most Dangerous Emotion in Investing
As we conclude this analysis, it is important to address the psychological challenge that many investors face: the fear of missing out. The AI boom has created a powerful narrative, and it is natural to feel anxious about being left behind.
Why FOMO Is So Dangerous
1
It Overrides Rational Analysis
When you're gripped by FOMO, you stop analyzing fundamentals and start chasing prices. The fear of missing gains becomes more powerful than the fear of losing money.
2
It Drives You to Buy at the Top
The strongest FOMO typically occurs at the peak of bubbles, right before they collapse. This is precisely when investments are most dangerous.
3
It Ignores Historical Lessons
Every bubble in history has been driven by the fear of missing out. "This time is different" are the four most expensive words in investing.
Learning from History: It's Never Different
The history of financial markets is littered with examples of investors who bought at the top of a bubble, convinced that "this time is different." It is never different. The fundamentals always matter. Companies must eventually generate profits. Technologies must eventually deliver on their promises.
When the gap between the hype and the reality becomes too large, the correction is inevitable and often brutal. Those who bought Pets.com or countless other dot-com darlings at their peak lost everything. Those who bought real estate in 2007 thinking it could only go up suffered devastating losses.
The AI bubble will follow the same pattern. The only question is timing. By the time the bubble is obvious to everyone, it will be too late to avoid the losses.
The Retail Investor's Advantage: Patience
As a retail investor, your greatest advantage is that you are not subject to the same career pressures as professional fund managers. You do not need to chase performance every quarter to justify your fees or keep your job. You do not need to invest in AI stocks just because everyone else is doing it. You can afford to be patient, to wait for valuations to become more reasonable, and to focus on companies with proven business models and real earnings.
"In the long run, this disciplined approach will serve you far better than chasing the latest hot stock. The goal of investing is not to get rich quick, but to build wealth steadily over time."
Strategic Summary: Positioning for the Future
Avoid the Hype Layer
Stay away from unprofitable AI model developers and data center operators with unsustainable economics
Embrace Infrastructure
Focus on power companies, semiconductor equipment manufacturers, and chip foundries with proven profitability
Consider Commodities
Build positions in copper, lithium, and other materials needed for the physical buildout
Maintain Discipline
Resist FOMO and stick to fundamental analysis. Wait for reasonable valuations and clear paths to profitability
Final Thoughts: Preparing for the Aftermath
The AI bubble will eventually burst, as all bubbles do. When it does, the investors who positioned themselves wisely—long the infrastructure, short or avoiding the hype—will be the ones who emerge with their capital intact and their portfolios positioned for the next phase of the market cycle.
The picks and shovels strategy has proven successful in every gold rush throughout history. While the prospectors often went broke, the merchants who sold them supplies made steady, reliable profits. In the AI gold rush, the merchants are the power companies, the semiconductor equipment manufacturers, and the chip foundries. These are the companies that will profit regardless of whether the current generation of AI models ever achieves commercial success.
Remember that the goal of investing is not to get rich quick, but to build wealth steadily over time. Approach this exciting but treacherous market with a healthy dose of skepticism. Conduct thorough research. Focus on fundamentals. And above all, resist the siren song of FOMO. Your future self will thank you.
References & Further Reading
Primary Sources
  1. The Grant Williams Podcast: Episode 109 - Julien Garran (October 13, 2025)
  1. Zitron, E. "Why Everybody Is Losing Money On AI" - Where's Your Ed At (September 5, 2025)
  1. NASDAQ: "AI Infrastructure Spending Is Projected to Hit $490 Billion" (October 26, 2025)
  1. International Energy Agency: "AI is set to drive surging electricity demand from data centres" (April 10, 2025)
Additional Reading
  1. Reuters: "OpenAI generates $4.3 billion in revenue in first half of 2025" (October 1, 2025)
  1. Fortune: "MIT report: 95% of generative AI pilots at companies are failing" (August 18, 2025)
  1. Bender et al.: "On the Dangers of Stochastic Parrots" - ACM Conference on Fairness, Accountability, and Transparency (2021)
This report represents analysis and opinion for educational purposes. Always conduct your own research and consult with financial advisors before making investment decisions. Past performance does not guarantee future results.