저자: 미상 · 외국인
요약 보기
마이클 베리는 엔비디아 중심의 AI 버블이 형성 중이며 붕괴 시 금융시스템 전반에 파급될 것이라고 주장한다. 핵심 근거는 GPU 감가상각 기간을 3~4년에서 6년으로 늘려 이익이 인위적으로 부풀려졌다는 점이다. 엔비디아는 2년 주기 신제품 출시로 구형 칩이 빠르게 경쟁력을 잃는다는 점을 근거로 들며, 엔비디아 측은 6년 감가상각이 적절하다고 반박했다. 또한 AI 인프라 투자가 약 3조 달러 규모지만 수익화는 미흡하고 대규모 차입이 동원되고 있다고 지적한다. 2026년경 촉발 사건이 있을 것이며 연준의 대응이 다음 버블의 씨앗이 될 수 있다고 경고한다.
다듬은 글
Last time when Michael Bur warned the market was delusional, the world lost trillions of dollars. Now he says the biggest bubble in financial history is forming again, but this time it is around Nvidia. He believes AI profits are fake, valuations are insane, and Wall Street is blind to what's coming.
If he's right again, the collapse will hit not only tech stocks, but it will also wipe out retirement funds, destroy credit markets, and take the entire economy down with it. So why is Nvidia at the center of a bubble that could break the world again? The big AI companies like Meta, Google, and he is saying that they have extreme valuations and unrealistic growth expectations and investors are convinced that this time it is different.
The danger of an AI bubble begins with accounting because the profits you're seeing are not real. What Bur is talking about is wrong asset depreciation for increased profits. In simple language, asset depreciation for increased profits means when a company buys machines for production, they have to consider the depreciation value of that machine each year.
For example, if a company buys an expensive machine worth $100,000, which has a life of 5 years, with 5 years of life, the company can depreciate the machine each year by $20,000. But if the company says that the machine has a life of 10 years, then each year the machine loses value of $10,000. The company's profits are magically increased by $10,000 each year.
Nothing actually improved in real life. They just changed the accounting estimate. No extra revenue, no extra efficiency, just a trick. Bur says this is exactly what is happening with AI companies. The most important equipment in the AI revolution is Nvidia's GPUs. These GPUs are incredibly powerful, but they become outdated quickly because Nvidia releases a better version of these GPUs almost every 18 to 24 months.
The older chip might still work, but it becomes slow, inefficient, and expensive to operate compared to the new one. And AI companies have changed their accounting rules to show that these chips will stay useful for 6 years. Before, they used to depreciate GPU hardware over 3 or 4 years. Now they're spreading the cost over 6 years.
That means yearly expenses go down and reported profits go up artificially. Many investors do not realize this. They only look at headlines. They see record profits. They see skyrocketing revenue. They assume AI is already paying off. But Bur is paying close attention and he's noticing that the profitability is not coming from better business performance.
It is coming from extending depreciation life. Even many of us common people change our phones or laptops after 3 to 4 years because new ones come with better processors, new features and better efficiency. So if normal people are changing their devices that frequently, do you think these AI companies which primarily depend on fast calculations and efficiency of these GPUs won't change their GPUs every year?
On top of that, Nvidia itself kills its own old chips every 2 years. In 2023, Nvidia's Hopper series's H100 GPUs were the most advanced. Then in late 2023, Nvidia launched the H200 GPU, which was better than the H100. Then in 2024, Nvidia launched the Blackwell series B100 and B200 GPUs, which are next generation and much more efficient.
Next year in 2026, Nvidia is launching its Reuben series GPUs, which will be more powerful. Did you notice the pattern? Each new GPU is multiple times more powerful than the previous one. Bur is saying that the Blackwell is almost 40 times more powerful than the H100. So these older GPUs become economically useless long before 6 years.
They still function, but functionally useless means companies using them lose money because they consume more power and produce slower results than newer chips. So the older chip becomes a liability, not an asset. Bur says if a chip becomes uncompetitive in just 2 years, then a six-year depreciation schedule is a lie.
And this lie boosts profits. Those boosted profits drive stock prices higher. Those stock prices make investors believe everything is fine when everything is not fine. Nvidia quickly fired back at Bur's analysis, distributing a private memo to Wall Street analysts that directly refuted his claims. Their central counterargument was that the 4 to6-year depreciation schedules are indeed appropriate because their chips have a longer functional life in a real world data center environment than critics acknowledge.
They cited older chips like the A100 GPU released in 2020 as evidence that their hardware is still running at high utilization rates today, justifying the long-term accounting treatment. Essentially, Nvidia is telling investors that what Bur warned about is not economic reality and that the long life of their chips supported by their powerful CUDA software ecosystem that validates the current profit reporting.
Bur stood by his analysis saying that there also comes the capital expenditure problem. The world is spending around $3 trillion to build AI infrastructure, new data centers, more GPUs, faster networks, cooling systems, electricity supplies. It is the largest technology spending boom in human history.
But the painful question is where is the return on this investment. Are companies making enough money from AI right now? No. Are AI tools profitable? Most are losing money. Are customers paying high prices? No. Companies are giving away AI for free to attract users, hoping monetization will magically appear later.
The money being spent is real. The profits are not real yet, but valuations assume massive profits will come soon. Bur is saying history shows that this usually ends badly. We've seen this movie before during the dot and housing crisis. During the dot bubble, the internet absolutely changed the world, but most internet companies died because they could not generate profits.
Back then, investors were betting on future potential without real revenue to support the price. And today, investors are betting on AI potential without real profits to support the price. Many of these huge data center infrastructure costs are being financed with loans and bonds. That means companies are taking huge debts, assuming future profits will pay them back.
But if real profits do not show up, these debts become impossible to handle. That is when a tech bubble turns into a credit crisis. In my previous video, I explained in detail how debt connects one company to another company and affects the market. If one company fails to repay its huge debt, it could trigger panic in the entire market.
Exactly like in 2008 when the housing market collapsed. There is also a factor of energy. AI data centers consume insane levels of electricity. They're consuming power equal to small cities. If old chips consume more electricity to produce the same calculation, the electricity cost alone can make them economically worthless.
This will force these companies to retire older GPUs faster than accounting statements claim, which means the six-year depreciation schedules are even more unrealistic. Now, combine all these dangers where hardware is aging rapidly, profits are inflated, and companies are spending huge capital. They're also taking on huge debts and energy costs.
And the most important one is the weak monetization. Bur is saying this is the exact pattern he saw before the housing crash. Not the same product, not the same industry, but the same financial blindness. Everyone is caught up in the narrative. The narrative is AI will make everything more efficient or AI will transform every business.
That might be true in the long term, but investors assume those gains will arrive immediately and pay for today's spending. Meanwhile, the spending is real and happening now, and the returns are not. This is usually the psychology of bubbles. People confuse hopes and dreams with business fundamentals.
They assume price momentum means they're right. They see Nvidia stock going up every week and believe it will never fall. They ignore warning signs because accepting them would mean admitting the possibility of losses. Also, these big tech companies are spending billions not because the returns are guaranteed, but because they're afraid of missing out.
Microsoft spends because Google spends. Google spends because Amazon spends. Meta spends because everyone else spends. When everyone does the same thing for the same emotional reason, the market becomes fragile. If one major player suddenly announces a slowdown in AI spending, the illusion breaks. Bur believes that AI spending will remain strong this year and next year.
Nvidia might continue showing great results through 2025. But somewhere around 2026, a trigger will arrive. Whatever that trigger will be, once the first crack appears, more cracks will follow. Everyone suddenly realizes the valuations were based on assumptions and not on reality. One more thing is that as the Nvidia stock is booming, pension funds and mutual funds own huge quantities of Nvidia stocks.
Hedge funds are betting heavily on AI. Banks are lending to infrastructure projects. Private equity firms have bought data center companies. So if the AI bubble cracks, losses will spread across the financial system. It will not be limited to tech. It could affect the whole economy. When the crash begins, the Federal Reserve will be under pressure to save markets again.
There will be demands for rate cuts, liquidity injections, and government support. Policymakers will try to prevent a deep crisis, but by doing so, they may plant the seeds for the next bubble. This is exactly what happened after the housing crisis. The policies that saved the system created a decade of near zero interest rates, which inflated asset prices and led to the current bubble.
I've explained that in detail in this video. Do check that out. All Michael Bur is saying and warning is that the depreciation method is wrong. Profits are inflated, valuations are delusional, the capital structure is fragile, and the energy economics is brutal. The bubble will eventually burst. He doesn't know exactly when, but he knows the end of the story.
And the endings of bubbles never change. The terrifying part is that if he is right, then we're looking at a much bigger crisis than we have ever seen before.
Comments 0