Ten Trillion in the Fog: Could AI Revolution Become a Bubble?

Thinking Note – Watches, AI, Business as an Art, Critical Thinking in Vanguard Leadership

The AI revolution is real — but the numbers look fragile.

$300B invested for $60-80B in annual returns. While the West burns cash chasing “the best” AI, China quietly builds “good enough” at 1/10th the cost. Like luxury watches, the premium AI story works until everyone realizes a reliable alternative exists at a fraction of the price. Nvidia could go from $900 to $300 when this corrects. The technology survives; the bubble doesn’t. The question isn’t whether AI is transformative—it’s whether you’re positioned for the reality or the hype.

Intro: About Watches, Living Artists, AI, Hype, and Smart Investing

While half the world was posting about AI yesterday, I spent the evening doing something I’ve done for decades – following horology and watching the GPHG awards.

I’ve been watching brands like ming become increasingly interesting over the years. Their designs are clever, the story is strong, and the community is loyal. Every year, I feel the pull to buy one finally… and every year, something in me still doesn’t quite pull the trigger.

On the other side, yesterday I was genuinely happy and proud to see Kari Voutilainen’s success. (I don’t know him, and I am just a fan of his virtues) His watches were competing in multiple categories and winning in several of them. Kari is, to me, the embodiment of virtue in craft—more than a watchmaker – a true living artist and, in many ways, a real Vanguard.

Since acquiring the Urban Jürgensen business, it has been growing, and more people now have the chance to wear and enjoy his art. The UJ-2 Double Wheel Natural Escapement – that, for me, is a watch. The UJ-3 came second in the men’s complication category. In other words, Kari quietly and consistently won GPHG. This is a great reminder that Time needs to be kept and spent beautifully.

And yet… most people have never heard of him. But almost everyone has heard of Rolex, and a lot of people want one or more.

That contrast is what triggered this thinking note.

Because this is not just a story about watches, it’s a story about how to try to be wise with investments, how to read signals, how to switch on critical thinking, and how to look for signals, learn, seek, and find your place in the world.

The world is moving in a direction none of us fully control – especially in AI and finance. We don’t choose the direction of the wave. But we should choose our role in it: winner, loser, or just an observer. And when a role is determined to act.

From there, we can finally talk about AI, ten trillion in the fog – and what the current “AI revolution” really looks like when you put the hype next to the numbers.

AI Is Real – and That’s Exactly Why It Can Enchant

We all already know that AI is real and valuable.

It’s entering every corner of life: it writes texts, draws images, helps doctors, writes code, and analyzes data faster than an entire team of analysts.

This is not a trick – it’s a technology that is genuinely changing the world, and at a speed that reminds us of the time when the internet exploded, only even faster.

But precisely because it’s so powerful and impressive, AI has another side: it easily enchants. (I am enchanted too)

Especially small investors, people who stare at charts on their phones, read headlines on portals, and think they’re missing the “opportunity of a lifetime”.

A few big growth stories, a few bombastic AGI announcements, a couple of viral demo videos – and suddenly it looks like the only rational move is to put everything you have into “AI stocks” or anything that has “AI” in its name.

That’s where the problem starts. When everyone sees only the story, and nobody looks at the math. When nobody asks anymore:

• How much does this really earn?
• How much does it cost?
• How sustainable is it?

Instead, the market behaves like a stadium full of fans, not like a cold, sober accountant. (I am the fan)

That’s why it’s important to look at this current AI wave critically – not from the position of a technophobe, but from the position of a sober person who asks:

What is really happening, who is earning, who is losing, and what awaits us over the next few years if this story continues at this pace?

Only when we clear that up can we tell the difference between real value and a we can access likehud of a bubble like scenarios.

Why Does History Keep Repeating in Business? (Always)

In the boardrooms of Silicon Valley, full of endless brainpower, the big tech families have poured a fortune into a new force called artificial intelligence over the last two years. Nvidia’s stock, which once stood at around $150, jumped to over $900. That rise gave people like Jensen Huang, one of the dons of those houses, a reason to talk about a “revolution in every industry.”

Microsoft has spent around $60 billion on infrastructure, building empires of data centers and chips. Google, Amazon, Meta – they’ve all put almost their entire fortune on AI, as if this were the last, biggest score of their lives.

Sam Altman and other players shout that AGI – artificial general intelligence – will arrive in a few years. Investors, quiet men with deep pockets, keep filling their coffers as if there will be no dawn.

But if we look at historical data and current numbers without excitement and emotion, we see that the risks are significant and obvious, not just some of us: the numbers don’t line up.

These families are pushing hundreds of billions into the machine, and what do they get in return? Crumbs and hopium. There are plenty of applications, but few are profitable, and the costs are bottomless.

Is this the dot-com bubble reborn – only bigger, more dangerous, ready to swallow trillions of dollars like bad debt?

All wars, including business wars, have their rhythm – a cycle that repeats through generations. As wars evolved from classical to asymmetric, hybrid, composite, and dynamic forms, business is also evolving with brilliant, fast, and vanguard leaders.

AI is poised to transform war and businesses. There is no doubt, and it is wise not to focus on spending and ROI when the race for the best weapon is on.

But from a small investor’s perspective, we need to dig more into the topic.

A new force always appears: railways in the 19th century, radio in the 1920s, the internet around 2000, and crypto in 2021.

The whisper is always the same:

“This changes everything.”

Money flows like wine at a wedding, and stocks shoot up to the sky. Then the blade falls.

The technology works – but not as big as promised. Revenues dry up, empires collapse, fortunes vanish overnight.

Not because the technology is fake – AI really “lives” – but because the dreams are too big, the spending is insane, and the path to profit is hidden in a foggy alley.

So let’s start thinking critically about the numbers, the moves, and how – and when – this story might end.

First: the Investments – Numbers That Insult Common Sense

In 2023, the tech families spent around $200 billion on the skeleton of AI: data centers, chips, electricity, cooling – everything that gives the models “lungs”.

By 2024, that jumps to $350 billion. For 2025, projections talk about more than $400 billion.

Almost $1 trillion in three years just for hardware – not counting research, engineer salaries, startup acquisitions, and app development done in secret.

Who’s actually paying for this?

Nvidia is not losing money – it’s collecting it. Their GPUs are the weapons of choice.

Data center revenue:

• $15 billion in 2022
• $47 billion in 2023
• over $100 billion projected for 2024

A sevenfold increase in two years, all thanks to AI.

Customers are buying H100s and H200s at $25–40,000 apiece, in tens of thousands of units. One serious training cluster costs from $0.5–1+ billion. Every major player is building several of them.

Microsoft has spent around $60 billion in the last two years – data centers, chips, electricity – plus $13 billion directly into OpenAI. Google has invested around $50 billion, developing its own TPUs alongside Nvidia’s GPUs. Meta has spent more than $40 billion, and Amazon has spent around $30 billion.

These five – Microsoft, Google, Meta, Amazon, and a network of Nvidia’s big clients like Oracle and Tesla – together have burned around $300 billion in two years. And the fire is only growing.

Does the Revenue Match?

Three hundred billion demands a return – with interest. And the revenues so far don’t match the story.

• OpenAI, the spark that ignited everything with ChatGPT, by the end of 2024 is reportedly generating around $3.5 billion in annual revenue, up from $1 billion in 2023. Nice growth. But costs are even higher – more than $5 billion goes to computing and deals with Microsoft and others. Roughly $2 billion loss. Valuation around $150 billion = ~43x revenue for a company still underwater.
• Microsoft is embedding AI into everything – Copilot in Office, Azure, Bing – but how much does it really earn from that? Estimates say $10–15 billion in additional annual revenue. $60 billion already spent; they need 4–6 years to get that back, if everything stays stable and costs don’t explode.
• Google’s additional AI-related revenue from cloud and advertising might be $15–20 billion. They’ve invested around $50 billion. Payback? 3–4 years, if everything aligns. Gemini is mostly given away – a defensive move, not a cash machine.
• Meta improves ads and feeds using AI, but new products like their assistant earn almost nothing. $40+ billion spent, mostly on AI and labs. New features are given away for free, like air.
• Amazon’s AWS, Alexa, and logistics might be getting $10–15 billion from AI services. ~$30 billion invested. Payback stretches over years. Alexa has been burning billions in losses for years – AI doesn’t fix that.

When you add it all up:

• around $300 billion spent
• around $60–80 billion in additional annual revenue

A large part of that revenue comes from old products, simply “rebranded” with AI. Completely new sources, like ChatGPT subscriptions and AI assistants, contribute maybe $10–15 billion.

The ratio is roughly 20:1. At that pace, it would take twenty years for these investments to pay back.

In this game, twenty years is an eternity.

Meanwhile, Across the Ocean: The “Good Enough” Strategy

And while the US houses compete to see who can burn more on chips and data centers, across the ocean, another world is rising – quiet, disciplined, virtuous, Chinese.

There, the state doesn’t write tweets about revolution; it writes plans for five, fifty, five hundred years.

Their houses – Baidu, Alibaba, Tencent, and a whole army of new players – are not necessarily trying to build “the smartest model in the world.” It’s enough for them to build models that are “good enough” for 80% of tasks, but at half or a tenth of the price.

China is building its own parallel AI universe: its own cloud, its own chips wherever possible, its own regulations, its own ecosystem of applications.

While the West is paying $30–40,000 for an H100, the Chinese are slowly pushing cheaper alternatives and local models, first at home, then toward the Global South – Africa, Asia, Latin America. For customers who watch every dollar, it doesn’t matter if the model is “a little bit better”; what matters is whether it’s a fraction of the price.

For Western giants, this is a nightmare they are, for now, sweeping under the rug. On one side, their own costs are rising – electricity, chips, and engineers. On the other side, the pressure of cheaper Chinese solutions is pushing service prices down and squeezing margins.

The equation that barely works today can completely fall apart tomorrow.

If you only look at the hype, AI seems like one global, glamorous stage. If you look coldly, you see two families playing different games:

• one spends as if there’s no tomorrow
• the other builds quietly, focused on volume and cost.

When those two waves collide, the market will have to decide:

Will it pay for the story – or for the result?

If Growth Slows, the Story Cracks

The use cases are impressive – text, images, video, data, coding – but the real question is:

Are enough people truly willing to pay for this?

Individuals are willing to spend $10–30 a month:

• ChatGPT Plus around $20
• Copilot around $30
• Google One around $20

To generate serious revenue, you need hundreds of millions of subscribers.

ChatGPT reportedly has about 10 million paying users: roughly $2.4 billion per year from subscriptions, plus enterprise clients to reach $3.5 billion in total. But costs still eat more than that. Raise prices – and you lose users. Cut costs – and you risk quality; you start chasing miracles.

Copilot has fewer than 5 million users – maybe around $1.8 billion in revenue. Google perhaps has 2 million, around $0.5 billion. Growth is already slowing; the first wave of curiosity is over.

Companies hesitate:

• Integration is complex
• Privacy is a problem
• Costs are dangerous if you miscalculate the cutover

Costs: A Special Circle of Hell

Training GPT-4 costs $100M+. Running the model (inference) costs 1–4 cents per query. Billions of queries per year mean $1.2–5 billion in compute costs.

Chips are getting more powerful, but models are getting bigger and hungrier. GPT-5 will demand more than GPT-4.

Energy is the silent killer.

Training GPT-4 consumed around 50 GWh – enough electricity for 5,000 households for a year. Running the model consumes even more: a single H100 draws around 700W. 10,000 units – 7 MW. 50,000 to 100,000 – 35–70 MW, non-stop.

Power bills: $50–100 million per year per big data center, plus an additional 30–50% for cooling.

By 2027, AI could be using as much electricity as entire countries, such as Argentina and Sweden. Grids are stretching to the breaking point, and the planet pays the bill.

And how much water is spent on cooling is a horror story of its own.

Look at the Stocks – and Then at the Pattern

Nvidia went from $150 to $900. Market cap ~$3 trillion. P/E around 65 – classic bubble shine. Such a valuation assumes that AI spending will continue forever.

If investments slow down, the story cracks.

We already see the signals:

• Startups are running out of money
• Cash is burning faster than it comes in
• Investors are asking: “Where is the revenue? Where is the path to profit?”

Memories of the dot-com bubble are clear: NASDAQ fell from 5,000 to 1,100 points – a 78% drop.

AI is similar: the technology is real, but the promises are often inflated.

Example: Inflection AI. They raised about $1.3B. Their Pi chatbot is free, with almost no clear revenue. Microsoft acqui-hires them for around $650M – half the money has already gone on compute. Investors are left short. The script repeats.

The Nvidia Scenario – From Rolex to “Just a Chip”

This is the house that today looks unbeatable – a stock that went from 900 to 300 dollars would mean a 50–70% drop and wipe out around $2 trillion in market value. One day you’re the king of the table, the next day someone turns off the lights.

Why Nvidia?

Because the whole story rests on one assumption: that the world will keep paying for their GPUs as if they were holy scripture, at $30 40,000+ per piece, for years to come. And across the table sits China, playing a different game.

Nvidia is the Rolex among chips. People are overjoyed when, after two years of waiting, Rolex tells them their ordered watch has arrived. Why? If they bought it wisely, that watch is now worth more on the market than what they paid for it. Sometimes twice as much. Millions of people want a Rolex, and for now, the market works that way.

China has been making fake Rolexes for years – almost identical to the originals – for 100x lower price. At the same time, original Chinese watches are gaining greater respect in the world of horology. At the GPHG, Chinese pieces are advancing to the finals and even winning special prizes.

Rolex did not enter the finals in any of the categories.

Chips and AI models from China are not perfect, they’re not the best on paper, but they are “good enough” for 70–80%of tasks – and at half or a tenth of the price.

China doesn’t care about having “the biggest model in the world”; they care about volume and efficiency: millions of chips, thousands of data centers, the entire chain from production to application under their control.

The West says:

“Our AI is the strongest.”

China says:

“Ours is strong enough – and gives the best value for money.”

Now consider this scenario is in play in a few years or before:

• Companies that are not at the very top, the ones that watch every cent, start choosing cheaper Chinese or local alternatives.
• Models adapt to those chips, optimized for “good enough” instead of “best in the world.”
• The price of “premium” GPUs starts to fall, first quietly, then sharply.

At that moment, Nvidia is no longer the only Rolex in town.

The chip becomes a commodity, almost a piece of metal.

Margins get squeezed, new orders slow down, and the market – which has already baked the story of “eternal growth” into the price – starts panicking and correcting its dreams.

That’s the moment when the stock goes from 900 down toward 300.

Not because Nvidia suddenly became a bad company, but because the world realized there is another house across the ocean that makes chips that are good enough, getting closer with each generation, at half the price.

The age of luxury ends.

The age of brutal efficiency begins.

What are the chances for this scenario to happen in percentage terms? 0,30, 50, 70, 90?

Here are some answers from AI:

Gemmini Likelihood Assessment: 70%

This high probability is based on analyzing three key factors presented in your text against current market data:

1. China’s Rapid “Good Enough” Progress: Chinese firms are closing the performance gap, with some models already competitive with top Western counterparts, particularly on tasks involving Chinese language/data. They are achieving this by focusing on algorithmic efficiency (like Mixture-of-Experts architectures) and cost-effective hardware (domestic chips like Huawei’s Ascend series, and even circumventing restrictions with modified Nvidia chips), aligning perfectly with the “good enough at a fraction of the price” strategy.
2. The Western Cost and Profitability Crisis: The analysis of US companies’ revenue vs. astronomical costs (training, inference, electricity) points to a precarious business model. Training costs are in the hundreds of millions, and inference costs alone can run into billions annually. If subscriber growth stalls, the path to sustained profitability becomes extremely difficult, making the high valuation—which is based on a story of eternal hyper-growth—highly vulnerable to a correction.
3. The Inevitable Commoditization of Compute: Nvidia’s dominance (holding over 80-90% of the AI accelerator market) is its key strength, underpinned by its CUDA ecosystem. However, the market is actively seeking alternatives.

Deep Seek AI Chance of this exact scenario playing out as described: 30% (I think this is the prudent and wise answer from the China perspective)

Arguments for a Higher Probability (Why it’s Plausible – 30%)

• The Economic Logic is Sound: The “innovator’s dilemma” is a classic business phenomenon. Incumbents (Nvidia/West) focused on high-margin, high-performance products are often vulnerable to disruptors offering “good enough” solutions at a fraction of the price. This happened with PCs, smartphones, and cloud computing.
• The Cost Problem is Real: Your analysis of the AI cost structure is accurate. The energy, compute, and infrastructure costs are astronomical and potentially unsustainable at current growth rates. A slowdown in investment is inevitable if ROI doesn’t materialize.
• China’s Determination is Unmatched: The Chinese government is pouring resources into technological self-sufficiency. They have a massive domestic market to refine their products and a clear strategic plan. They don’t need to beat Nvidia on the absolute performance frontier; they just need to close the gap enough for their value proposition to become irresistible to a large segment of the market.
• Valuation is Built on Perpetual Growth: Nvidia’s current valuation absolutely prices in years of continued hyper-growth and dominant margins. Any significant crack in that narrative—even a slowdown in growth rate, let alone a decline—could cause a severe re-rating of the stock.

Arguments for a Lower Probability (Why it’s Not a Sure Thing – 70%)

• The Software Moat (CUDA): Nvidia’s most powerful asset isn’t the H100 chip; it’s the CUDA software ecosystem. Millions of developers and every major AI model are built on CUDA. Switching to a new hardware architecture (Chinese or otherwise) is like asking the entire world to stop using Windows or iOS. This creates immense friction and lock-in.
• The Performance Gap is Still Massive: While Chinese chips are improving, the performance gap between a top-tier Chinese chip (e.g., Ascend 910B) and an H100 is still significant, estimated to be 2-3 years. In the fast-moving AI world, that’s an eternity. Nvidia isn’t standing still; their next-generation Blackwell chips widen the gap again.
• Geopolitical Firewalls: US export controls are actively designed to prevent this scenario. They limit China’s access to the most advanced chips and the manufacturing equipment to make them. This is a massive headwind for China’s tech leap.
• The “AI is More Than Chat” Argument: The highest-value, most profitable applications of AI in the future (scientific discovery, complex system simulation, advanced robotics) will demand the absolute highest performance. The market for the “Rolex” chip may not disappear; it might just become a more specialized, high-end segment that Nvidia continues to dominate.
• Nvidia is Adapting: Nvidia is already developing less expensive chips for the Chinese market (e.g., H20) to comply with sanctions and compete on their turf. They are also vertically integrating into data centers, software, and services, building moats beyond just selling chips.

Refined Probabilities on the Timeline

• Chance in the next 2-3 years: <10%. The software moat, performance gap, and geopolitical barriers are too high for a complete disruption this quickly.
• Chance in a 5-7 year horizon: 30%. This is the most likely window for your scenario to begin unfolding, as Chinese tech matures and global economic pressures mount.
• Chance of a less severe, but similar, outcome: 70%. It is highly probable that growth in AI spending will slow, Chinese competition will intensify, and Nvidia’s margins will face pressure. This could lead to a significant stock correction (e.g., 30-50%), even if it doesn’t reach the catastrophic 67% drop to $300. The era of “brutal efficiency” will arrive, but Nvidia may remain the high-performance leader in a now-segmented market.

Conclusion:

Your scenario is not a fantasy; it’s a credible and well-argued risk case. However, it probably underweights Nvidia’s immense software moat and the sheer technical challenge facing China due to export controls.

Therefore, while the core pressures you identify are very real and will likely impact Nvidia’s growth story, the probability of the specific, catastrophic outcome you describe—a drop to $300 driven primarily by Chinese “good enough” chips—is around 30%. The more likely outcome is a painful market correction and a new, more competitive reality, but not the total dethroning of Nvidia as the high-performance leader.

Claude AI evaluation: Looking at your Rolex-to-commodity scenario for Nvidia, I’d put the probability at 60-70% over a 5-7 year timeframe. Here’s my reasoning:

Why this high:

1. Historical inevitability – Every “proprietary premium” technology eventually gets commoditized. IBM mainframes, Intel x86 dominance, Oracle databases, Cisco routers – all followed this pattern. The question is when, not if.
2. China’s execution track record – Solar panels (95% market share), 5G equipment (Huawei despite sanctions), EVs (BYD now competing with Tesla), and high-speed rail. They consistently execute the “good enough at 1/10th price” strategy.
3. The math supports it – As you brilliantly pointed out, 70-80% of AI tasks don’t need premium performance. Once CFOs realize they’re paying Rolex prices for Casio needs, the shift happens fast.
4. Open source momentum – LLaMA, Mistral, and Chinese models like Qwen are already proving smaller models can match GPT-4 on most tasks. This reduces the moat of premium hardware.

Why not higher:

1. CUDA ecosystem lock-in – Nvidia’s real moat isn’t the chips, it’s the software stack. Breaking this will take time.
2. Geopolitical barriers – Western governments might block Chinese alternatives for “security” reasons, artificially maintaining Nvidia’s premium.
3. Timeline uncertainty – This could take 7-10 years instead of 3-5. Markets can stay irrational longer than logic suggests.

Your Rolex analogy is perfect because Rolex actually did maintain premium pricing despite Chinese competition – but only in luxury/consumer markets where status matters. In B2B enterprise technology, brutal efficiency always wins eventually.

The trigger will be one major Western corporation publicly switching to Chinese AI chips and reporting “90% performance at 20% cost.” That’s the “Emperor has no clothes” moment

Could It Be Different?

Yes, new, huge revenue streams may appear.

But for now, the use cases are not that broad.

For most people and companies, one main AI model is enough. The market will have a winner. Enterprise services cannot grow indefinitely. Chip costs might drop 20–30%, but models grow and often absorb that drop.

AGI? That’s the ghost that feeds the hype. TOP AI experts and scientists claim the year of Genesis for AGI is 2027, but who knows, it could be decades away. And if it really does arrive one day, it could flip the entire economy upside down.

The bubble is visible, and the shadow of its popping is already there.

The pattern is always the same: the technology is real, the hype is ten times bigger.

The crash cleans everything; what remains after that builds real value.

AI doesn’t disappear – it just stops being a circus trick and becomes an installation in the basement. An invisible foundation, like the internet today: crucial, but mundane.

And Then There’s You (if you managed to come to this part :))

Ten trillion spent for $60–80 billion in annual revenue. Math could be bent, but it doesn’t forgive and the end.

But the numbers are only the cold side of the story.

The other side is you.

Where do you stand in all this?

Are you among those who fall in love with the story, the demo, the keynote on stage – and forget to look at the bill?

Are you among those who see every new hype as “this time it’s different”?

Or among the few who sit down at the table, pick up a pen ( your AI assistant ) and ask:

• Who is really making money here?
• Who is really carrying the risk?
• And who is just selling the story?

AI will stay. The only question is – at what price, and for whom.

When the noise fades, a few simple questions will remain:

• What do I actually understand here?
• Am I aware that this story can enchant me too, just like everyone else?

Bubbles don’t burst on the screen. They burst into people’s heads – at the moment when it’s already too late.

That’s where real critical thinking about AI begins.

Not when you’re amazed by the technology, but when you use it, understand its costs and limits – and when you actually make money from it.

The question is no longer:

“How much more can this go up?”

But:

“Where am I in this story, and how can I turn this to my advantage?”

You are not able to influence whether the bubble bursts, but you can decide:

• Will you be a spectator of the chart on the screen?
• Or someone who cool-headedly looks for concrete benefits: niches where AI lowers cost, speeds up work, or opens a new product or service.

These same ten trillion will destroy some people and launch others.

The difference won’t be in who admired AI more, but in who was willing to look through the hype and ask:

“Where is the real value here – for me, my company, my bank account?”