The Pattern of Greed

Greed never changes the script. It only changes the supply chain names. Here is a counterintuitive truth. When a technology becomes as indispensable as air and water, and as long as there is no monopoly, its commercial value will be exactly as cheap and unprofitable as air and water. Railways went through this. Electricity went through this. The Internet went through this. AI will not be the exception. The market is currently treating compute power, HBM, optical modules, transformers, and liquid cooling plates as the toll booths of a new era, paying a premium for permanent scarcity. Infrastructure history proves one thing over and over again. Excess premium itself is the most powerful bounty in the world. It mobilizes top tier intelligence and massive capital to smash the bottleneck. For two hundred years, this template only swaps the supply chain roster, never the ending.

Trains were getting heavier and faster. The tracks were made of wrought iron, which was too soft. On busy lines, the rails would deform and crack every few months and had to be scrapped and replaced. This was a physical material bottleneck that could not be fixed through process improvements. Capital immediately bought into a narrative. The entire society's iron smelting capacity would be used to repeatedly cast and repair these rails. Iron supply would be in chronic shortage. Iron factory order books stretched out for years. Iron prices nearly tripled in five years. On the London Stock Exchange, iron factory stock prices rose nearly 10x on average, with some leaders approaching 20x. Speculators did not even look at financial reports. As long as the company name contained the words iron or casting, they would buy it blindly. This is the exact same underlying logic as today's frenzy over HBM, optical modules, and transformers based on a supply chain list.

Then Bessemer arrived. In 1856, the acidic Bessemer process blew air into molten pig iron to rapidly burn off impurities. For the first time in human history, the problem of low-cost mass production of liquid steel was completely solved. The toughness and lifespan of steel rails were ten times that of wrought iron rails. The tracks were rapidly replaced with steel. The stock prices of the iron giants that were once valued at the sky were slashed, eventually falling back to where they started before the 1848 frenzy. 95% of traditional European and American wrought iron factories declared bankruptcy. The railway asset mania lasted four years, then it was crushed by technology.

Under Edison's DC system, long distance transmission required copper cables as thick as a waist to reduce resistance. Otherwise, a power station could only supply a radius of 1-2 kilometers. Newspaper headlines screamed that all the copper mines in the world combined would not be enough to build the American power grid. Whoever controlled copper controlled the power supply. Copper mining and cable stocks skyrocketed overnight. Capital grabbed a magnifying glass and walked down the power supply chain, checking for the next bottleneck. They quickly found it. There was no modern synthetic plastic at the time. High voltage cables, transformers, and generator coils all required high strength insulation. Otherwise, it meant short circuits at best and fires at worst. The best insulator humanity could find was wild natural Para rubber from the South American Amazon, which had extremely limited production capacity. The market firmly believed that without rubber, generators and power plants were just piles of scrap metal no matter how well they were built. Between 1895 and 1910, the price of insulating rubber skyrocketed by five hundred percent. The stock prices of some core companies in the insulation sector increased 20-30x in just 4-5 years. A dedicated stock trading market for insulation materials even emerged. The power stock boom lasted six years.

Then Tesla's AC current and transformers tore the copper cable bottleneck out by the roots. As voltage increased, resistance losses naturally decreased. Ordinary thin cables could transmit power over long distances. There was no longer any need to rely on thick copper cables. In 1907, Baekeland synthesized the first fully synthetic plastic, phenolic resin, using cheap phenol and formaldehyde. Combined with extremely cheap ceramic insulators, insulation performance improved more than tenfold, while the price was only one tenth of natural rubber. Once these bottlenecks were destroyed, laying power grids was no longer high tech. Excess capital flooded in. Electricity prices collapsed. The industry entered a wave of mass bankruptcies. The power infrastructure stocks that once looked down on everyone became the most boring dividend and old man stocks in the secondary market, burdened by massive equipment depreciation pressure and almost zero growth. But from this rubble, Ford used cheap electricity to create the modern industrial assembly line. White goods liberated household labor. Hollywood grew into a trillion dollar entertainment industry. Aluminum was once more expensive than gold. Napoleon III used aluminum forks to show off his status, while second-tier nobles had to use gold forks. Alcoa used cheap electricity for aluminum electrolysis, turning aluminum into cabbage price and spawning the modern aviation and construction industries. The road builders went bankrupt. The road users got rich.

Wall Street firmly believed that internet traffic would double every 100 days, and that bandwidth was king. Global capital poured 500 billion dollars into the telecom industry over five years, laying 39 million miles of fiber optic cable under the ocean and on land. The total length was enough to wrap around the Earth 1,500 times. To maintain the myth of bandwidth scarcity, telecom giants like Global Crossing and WorldCom played bandwidth swaps. Company A spent 100 million dollars buying capacity from Company B. Company B turned around and spent one 100 million dollars buying capacity from Company A. Not a single cent of actual cash flowed, yet both companies magically added 100 million dollars of revenue to their financial reports (sounds familiar?!). Enterprises went crazy buying switches. Order books for giants like Cisco and Lucent were pushed out over 10 months.

Cisco's stock price hit 80 dollars in March 2000, briefly surpassing Microsoft to become the most valuable company in human history at the time. Then Dense Wavelength Division Multiplexing arrived. Engineers just needed to install wavelength splitters at both ends of the fiber, allowing over 160 different colors of laser light to shoot into the same glass strand simultaneously. The communication capacity of each old fiber instantly matched that of over a 100 previous fibers. By 2002, 97.5% of the world's fiber optic cables were lying under the sea and underground with nobody using them. They became dark fiber with no signal. Transoceanic bandwidth prices plummeted by 90%. Over 60 telecom giants like Global Crossing and WorldCom lined up for bankruptcy. The entire industry evaporated 2 trillion dollars in market value. The expensive switches that were bought up turned into electronic waste within a few months. Discounted switch sale flyers were scattered everywhere in the US market. The internet infrastructure mania lasted five years. From the dark fiber rubble, Google, Netflix, Amazon AWS, along with China's Taobao, WeChat, and ByteDance, grew into absolute monsters using cabbage price bandwidth.

Everyone says that during the American Wild West gold rush, the safest and most profitable business was selling shovels. This is where Nvidia's AI shovel seller nickname comes from. But selling shovels only makes you fast money in the first few years of the gold rush. Once the mining area is fully explored or the tools iterate, a warehouse full of iron production will just become scrap metal. The people who truly built century old business empires from the gold rush were not the heavy asset road builders. They were the light asset ecosystem players who came later.

During the 19th century gold rush, Americans Henry Wells and William Fargo realized the US postal service was unreliable and could not safely send gold back home. So they provided armed escorts for gold and cargo, and directly bought gold in the mining areas to exchange for paper drafts. These two businesses attached to the gold miners erupted with massive potential, eventually incubating the famous premium credit card company American Express and Wells Fargo bank. Today, these two companies have a combined annual net profit of over 32 billion dollars, with a total market cap that once exceeded 410 billion dollars. Meanwhile, over 90% of the countless mining companies and iron producers that rose during the gold rush went bankrupt and disappeared. The road builders went bankrupt. The people selling water, credit, and ecosystems got rich.

Now it is AI's turn. If you strip down the massive AI rally of the past three years, you will realize something. Wall Street and global equity markets were never actually trading artificial intelligence itself. They were holding a magnifying glass and walking down the infrastructure assembly line, bidding up the bottlenecks. Whoever choked the expansion of compute power had their stock sent to the moon by capital. In 2023 when Large Language Models (LLM) exploded, chip packaging capacity could not keep up and graphics cards could not be delivered. Capital pumped Nvidia and TSMC.

Once packaging capacity doubled, you could not just leave the chips sitting on a table. Servers, racks, and switches had to catch up. So they pumped server and data center stocks. Once the machines were stacked up, they realized there was not enough power and cooling could not keep up. Then they pumped power, nuclear energy, and cooling equipment stocks. The phrase AI's limit is power even became popular. Later, as inference and agents consumed more, computing fast was not enough. You had to store it and transmit it. High-bandwidth memory, high-speed interconnect, custom chips, and even the once neglected CPU became capital's targets.

Once chips and memory were assembled, they hit the physical bottleneck of copper cables. Capital rushed to frantically pump 800G, 1.6T optical modules, and CPO co-packaged optics. Once optical modules were plugged in, they hit the power and cooling bottleneck again. Capital turned around and madly rushed to buy liquid cooling equipment, transformers, and nuclear power plants. This game of musical chairs on supply chain bottlenecks has reached the absolute end of the chain. When capital treats transformers and liquid cooling plates as core technological faith and pumps them to the sky, it means the game is over.

The debt market smelled it first. In October, Anthropic exposed a 63 billion dollar chip financing debt code-named Project Beijing. Of that, 43.5 billion dollars in senior debt had to be guaranteed by chip partner Broadcom. Out of 18 billion dollars in subordinated debt, Blackstone took on 9 billion dollars and was demanded a yield of over 10%. Keep in mind, companies with the same credit rating only pay around 7% interest in the public bond market. Anything over 10% is basically considered high-risk junk bond territory (trash). This means Wall Street is demanding a massive premium for the rapid depreciation and default risk of compute chips.

A top-tier global player like Anthropic, with quarterly revenue of 16.8 billion dollars, was not only unable to get a discounted rate to buy TPU chips from its major shareholder Google, but had to use future 5-year chip lease contracts as collateral just to pay over the 10% interest of its new debt. Adding the 35 billion dollars borrowed from Apollo in June, Anthropic has already taken on 95 billion dollars in chip debt and 518 billion dollars in long-term compute procurement commitments within a single year. Since Google went public in 2004, even through the 2008 subprime crisis and the 2020 COVID shock, it never experienced negative free cash flow.

Yet in the second quarter of this year, it reported negative 5.85 billion dollars in free cash flow for the first time. With no money, they have to go to the market to finance. But the Fed restarted rate hikes and Treasury yields are climbing. LLM companies are stuck between a rock and a hard place. Tech giants want to give money but do not have it. The market has money but it is very expensive. The equity market is still pricing compute hardware with a 5-10 year growth duration. The debt market is pricing it based on a 1-2 year chip collateral duration. This time mismatch is the greed template baring its teeth once again.

Remember this old adage on the Street: when the credit and equity markets are pricing two different things, the credit market is always right.

Right in the middle of this duration mismatch, the tape just served up a textbook high-level shakeout. Heavyweights like NVDA and MSFT ripped up and down, and the semiconductor sector got absolutely obliterated. The extreme whipsaws pushed holders to their breaking point. A close friend asked me the other day: among the retail crowd, how many actually know the difference between de-grossing and de-risking? My answer? Almost none. They think cutting size is the same thing as managing risk.

What is de-grossing? It’s just shrinking your total gross exposure. AI stocks dump, fear kicks in, you liquidate the heavyweights and hide in cash. Your gross exposure drops. Great. But your portfolio's risk profile hasn't changed one bit. You just went from full-throttle volatility to zero exposure. At least the guys who know how to cut size are doing something right. The real idiots are the bagholders who refuse to budge, waiting for the cavalry to rescue them. That is no different from a degenerate gambler going all-in on every hand. It’s pure, fear-driven capitulation. Volatility spikes, panic sets in, you de-gross. Then AI stocks bounce, the index hits new highs, FOMO strikes, and you chase it right back up. Your portfolio risk was never managed. You just ping-ponged between fear and greed. That’s not risk control. That’s a gambler’s suicide squeeze.

So what is de-risking? It’s about swapping out risk factors. Your gross exposure can stay exactly the same; you are adjusting the portfolio's risk characteristics, not just blindly cutting size. When real institutions face volatility, they don't hit the panic button and liquidate everything. They dump high-beta, high-vol retail AI names and rotate into low-vol defensive sectors, utilities, or treasuries. Or, they keep their core AI positions intact but buy puts or short index futures to hedge. Under a proper de-risking framework, your gross exposure might even increase because of the hedges, but your overall portfolio volatility and VaR will drop. The core of de-risking is factor substitution. It lets you sleep at night while carrying the same, or even more, risk capital.

Let’s pull up the chart. Look at the move from July to October 2026, with all values normalized to 100. The first line is 100% SMH, the fully exposed bagholder. When the semis sell-off hit, this red dotted line got its face ripped off, dropping to around 85. The second line is the retail de-gross: 70% SMH plus 30% SGOV, parking 30% in cash. Sounds smart, right? Cut 30% and hide in cash. Look at that yellow dashed line. When the panic hit in August and October, the retail crowd sold their chips for cash out of fear. The result? Once SMH bounced, that 30% cash became a massive cash drag. Panic selling didn't just make you dump your chips at the absolute bottom; it completely killed your upside capture on the rebound. The retail de-grossing strategy ended up hovering below 95, getting crushed by the broad tape.

How do the pros do it? Look at the green solid line. That’s real de-risking: 70% SMH plus 30% XLF, financials. The institutions didn't run to cash. Gross exposure stayed roughly the same. They simply swapped high-beta semis risk for low-correlation defensive assets. This line shows you exactly what it means to smooth drawdowns via uncorrelated assets while keeping your skin in the game. When SMH crashed, XLF’s resilience provided a buffer. The green line barely bled compared to the red line. And when semis rebounded, the green line rallied right back, completely dodging the cash drag that destroyed the yellow line. That’s what de-risking is. You sleep like a baby, and you don't get shaken out of your core book during the washout.

It is mind-boggling that in 2026, people are still stupid enough to think long VIX structures is a real hedge. Look at the tape. This entire sell-off is largely a semis dump. The SPX is just sitting there, propped up by defensive rotation.

Here is the basic math many seem to be missing: VIX tracks the S&P 500. It does not track the semiconductor index. When that pink VXN/VIX ratio rips, it is screaming at you that Nasdaq 100 vol is blowing out relative to the broad market. It tells you the blood is in tech and semis. You are buying broad market panic to hedge a sector-specific fire. That is not a hedge. That is a factor mismatch.

When you buy VIX structures, you are buying overall market volatility. But what you actually have is semis concentration risk. These are not the same factor. Semiconductors are getting slaughtered, SPX is held up by defensive sectors, front-end VIX moves lower. Your long VIX structure is not only directionally wrong, but you are bleeding carry into the contango roll. Your semis position is puking, and you’re getting double-fucked by a VIX curve steepening in your face. This is almost a Texas hedge. Using long VIX structures to hedge semis concentration risk is like bringing an umbrella to a gunfight. The umbrella will break, and the bullets will still go through.

How to hedge properly? Hedging, like macro trading, is a form of art at times. In case of concentration risk, whether it’s semis or precious metals, you need to find an asset with low volatility-adjusted correlation and the right instrument. Well, this part is not free.

Put history and the current tape side by side, and the verdict is brutal. The AI era has just finished its opening act of infrastructure buildout. The truly game-changing applications are nowhere near here. We cannot even conceptualize them yet. Long term, the value creation and the bull market are far from over. But the path there will be paved with a gut-wrenching depression, or an outright crash.

The railway mania ran for four years before the Bessemer process slaughtered the myth of permanent rail demand. The early power stock boom lasted six years before Tesla's AC and synthetic plastics scaled up. The telecom and fiber boom ran for five years before the dot-com wreckage hit the floor. This AI infrastructure rally, propped up entirely by bottleneck speculation, won't run much longer either. The technology that will obliterate the compute bottleneck isn't here yet (or perhaps it is already?), but when it hits, it will be a devastating wave of creative destruction.

When this AI infrastructure bubble finally pops, the future giants of the AI application layer will likely be bootstrapping right now in some unremarkable garage. Here is the iron law of the last two centuries: A technology so expensive that only the giants can afford it is nothing but a casino for capital. It only becomes a new era for humanity when it is so cheap that a street corner shop treats it like a worthless utility.

The people who bet on the internet changing the world watched their portfolios get chopped by 90%, yet they were dead right about the industry. The road builders went bankrupt. The guys selling cars got rich. Do not confuse the two. Being proven right by the market eventually and surviving the market's attempt to kill you halfway are two entirely different games. The technological revolution so far has not failed humanity. But it always requires a fresh batch of fanatical gamblers to eat the losses first.

This time, the bagholders are lining up, and the bill is denominated in compute. And yet, most of them still can't distinguish de-grossing from de-risking. They are using long VIX structures to hedge semis concentration risk. That is not a hedge. That is bringing an umbrella to a gunfight. The umbrella will break, and the bullets will still go through.

Don’t be the exit liquidity when shit hits the fan.

All rights pertaining to the above content are reserved. Any unauthorized reproduction, citation, or distribution in any form without the express prior written consent of the rights holder is strictly prohibited and will be pursued to the fullest extent under applicable law.

Disclaimer: This report represents solely a macro‑strategic analysis predicated upon information available as of the date hereof. It does not guarantee the accuracy of the data and simulation results, does not constitute specific investment advice, nor does it amount to an offer to buy or sell, or a solicitation of any transaction order. Considerable uncertainty attends the outlook, and actual developments may diverge materially from the trajectories delineated herein. Investment decisions should be based upon independent judgement and consultation with professional advisers.