A Mexican Standoff is a confrontation between three or more parties in which no single participant can advance, shoot, or retreat without ensuring their own destruction.
Unlike a classic two-person duel, where the fastest draw wins, a three-way standoff introduces a brutal paradox of game theory: the first person to shoot is almost guaranteed to lose. If Person A shoots Person B, Person A’s gun is instantly empty, or their attention is diverted, allowing Person C to shoot them completely unhindered. This creates a tense, frozen deadlock where everyone is highly incentivized not to make the first move.
The Ultimate Example: The Good, The Bad, and The Ugly
Director Sergio Leone put on a masterclass in this concept during the climax of his 1966 epic western, “The Good, The Bad, and The Ugly”. The three main characters, Blondie (The Good), Angel Eyes (The Bad), and Tuco (The Ugly), stand in a massive stone circle in Sad Hill Cemetery, competing for $200,000 in buried gold.
The classic three-way standoff in Sad Hill Cemetery. (The Good, The Bad and The Ugly)
The scene perfectly illustrates the tactical dilemma of the standoff:
The Mutual Threat: As they form a perfect triangle, each man evaluates his odds. If Angel Eyes shoots Tuco, Blondie (Clint Eastwood) will instantly kill Angel Eyes. If Tuco shoots Blondie, Angel Eyes kills Tuco.
The Psychological Torture: Because nobody wants to draw first, the tension stretches out for over five minutes. Leone uses extreme, rapid close-ups of shifting eyes, sweating faces, and hands hovering over holsters, backed by Ennio Morricone’s escalating score.
The Hidden Lever: Blondie secretly breaks the deadlock ahead of time. Knowing the math of a Mexican standoff is unwinnable by standard means, he secretly emptied Tuco’s gun the night before. Because Blondie knows Tuco is harmless, he can focus 100% of his speed on killing Angel Eyes the moment the shooting starts.
Ultimately, the Mexican standoff is a game of psychological chicken. It relies entirely on the premise that everyone has a loaded weapon and everyone knows that pulling the trigger is a form of mutual suicide.
The Three-Way Standoff: A Mexican Standoff in Tech
The concept of the Mexican Standoff has jumped from the cinematic dust of Sad Hill Cemetery straight into the modern global economy. A three-way, high-stakes deadlock has formed among Apple (the consumer-tech giant), Micron Technology (representing the memory-vendor oligopoly), and the US Government (wielding antitrust and regulatory oversight).
At the centre of this battle is an unprecedented surge in memory costs, dubbed “RAMageddon”, with DRAM and NAND flash prices climbing by 500%-700% over four years.
Just like the movie, the three players are locked in a triangle where nobody can move without taking heavy damage:
[ US GOVERNMENT ]
(Antitrust / CHIPS Act)
/ \
/ \
/ \
[ APPLE ] ————————————— [ MICRON ]
(The Uber-Buyer) (The Memory Oligopoly)Apple’s Dilemma: Apple has long used its massive purchasing power to bully memory suppliers into rock-bottom prices. Now, the tables have turned. Facing unavoidable memory costs, Apple raised MacBook and iPad prices by up to $300. If Apple tries to entirely bypass the big vendors by sourcing unapproved or sanctioned alternatives, it risks government retaliation and supply quality.
Micron’s Position: Micron is riding massive AI-driven tailwinds, but it is walking a fine line. It is aggressively squeezing customers to make up for years of terrible margins, but pushing too hard has triggered a massive regulatory backlash from the government and legal threats from consumers.
The Government’s Trap: The US government wants to punish memory price manipulation to protect consumers. However, under the CHIPS Act, the government is also heavily subsidizing Micron to build domestic fabs. If the government hits Micron too hard with antitrust penalties, it risks crippling the very domestic semiconductor supply chain it is trying to build.
The History of Collusion: Cycles of Revenge
The current standoff is deeply rooted in a toxic historical cycle of mutual retaliation between vendors and buyers.
Memory Vendors Colluding (The Sellers' Cartel)
The memory market is an oligopoly controlled by three players: Samsung, SK Hynix, and Micron. This isn't the first time they've been accused of rigging the game. Between 1998 and 2002, these exact companies engaged in a massive criminal conspiracy to fix DRAM prices. The US Department of Justice eventually cracked down, resulting in hundreds of millions of dollars in fines and actual prison sentences for several industry executives.
Memory Customers Colluding (The Buyers' Strike)
Conversely, giant tech buyers have historically banded together implicitly to squeeze memory vendors during economic downturns. During market gluts, dominant buyers like Apple, PC manufacturers, and auto companies would coordinate to withhold orders or relentlessly demand prices below the actual cost of manufacturing. By forcing memory vendors into negative margins, the buyers starved the chipmakers of capital, treating memory like a cheap commodity rather than a vital piece of infrastructure.
AI’s Role: The High-Bandwidth Memory (HBM) Squeeze
The explosive growth of AI infrastructure is the catalyst for this Mexican Standoff.
The Manufacturing Bottleneck: Producing HBM is highly complex and consumes three times the wafer capacity per bit compared to standard DRAM. As Micron, Samsung, and SK Hynix aggressively pivot their production lines to HBM to feed AI demand, the production capacity for traditional DDR4 and DDR5 memory used in consumer electronics has been severely cannibalized.
The Novelty of Long-Term Contracts: Historically, the memory sector operated on short-term, cyclical spot purchasing. In 2025 and 2026, the structural shortage forced a massive paradigm shift. Hyperscalers (like Microsoft and Google) began abandoning transactional buying to sign multi-billion-dollar, multi-year long-term contracts to lock up memory supply out to 2027 and 2030. These contracts have effectively boxed out consumer tech companies like Apple, leaving a shrinking pool of standard memory for the general market.
The Financial Cliff: Why Capacity Slipped
Micron's defensive stance stems from a severe financial crisis it faced during the 2023 semiconductor down-cycle.
Metric | Historical Crisis Era (2023) | AI Boom Era (2026) |
|---|---|---|
Gross & Operating Margins | Swung heavily into the negative | Record-breaking double-digit highs |
Operating Cash Flow | Severely depleted / bleeding cash | Massive, multi-billion dollar inflows |
Capital Expenditures (CAPEX) | Drastically Curtailed | Emergency expansion mode (Fabs building through 2030) |
Because aggressive price-cutting by buyers wiped out Micron's profitability in 2023, the industry was forced to restrict capacity buildout. Fabrication plants take 2 to 3 years to build; the under-investment forced upon Micron by its buyers in 2023 is the direct reason why there isn't enough memory capacity to meet the AI demand spike today.
The Voice from the Sides & Public Perception
The public standoff broke into the open through contrasting statements:
Tim Cook (CEO, Apple): The price hikes are "unavoidable." We have done our best to mitigate the enormous cost increases being passed on to us... but the situation has now become unsustainable. It is a "hundred-year flood" unlike anything I've seen in four decades.
Sumit Sadana (CBO, Micron): Excessive price pressure from customers has come back like a boomerang. We told a couple of the customers who were being very aggressive with pricing during the 2023 slump that this is not constructive. A lot of industry investments got shut down because of poor pricing.
The Public Perception Divide: General consumers and small businesses are furious, blaming "chipflation" for making electronics unaffordable. However, tech analysts note that Apple's own profit margins have safely outpaced inflation for a decade, viewing Apple's complaints as a sign that its long-standing "superior power" status over suppliers has completely shattered.
My Unfiltered View: This whole $AAPL and $MU saga reflects extremely poorly on Apple. I, for one, am glad the facts about how Apple abused the memory players over the years and overcharged customers are finally public knowledge. Apple needs to be investigated, not Micron.
The Breaking Point: The Federal Class-Action Lawsuit
The standoff hit a legal flashpoint on June 25, 2026, when individual consumers and small businesses filed a federal class-action lawsuit in California against Micron, Samsung, and SK Hynix.
The Core Allegation: The lawsuit alleges that the "Big Three" memory players have systematically used the AI boom and the transition to HBM as a "tactical excuse" to artificially throttle the supply of conventional memory (like DDR3 and DDR4), driving consumer RAM prices up 700%.
The Smoking Gun Claim: The plaintiffs point out that the companies cut legacy memory production concurrently, an action that defies standard economics, as prices were skyrocketing. They specifically cite Micron’s sudden 2025 termination of its highly profitable crucial consumer memory division during a time of record revenue as evidence of a coordinated effort to restrict consumer supply.
Represented by antitrust firm Bathaee Dunne LLP, the lawsuit seeks triple damages. It has created the final, tense element of the standoff: if the courts rule against Micron, the entire financial structure of the chip expansion could collapse, but if the government steps in to protect Micron, it alienates millions of angry tech consumers.
As one would have guessed, the whole AI space has been volatile recently. Trillion-dollar market cap companies are trading like penny or meme stocks.
AI Developments Roundup
Quite a lot is happening in the AI space on a daily basis. The headlines are coming in fast and furious. As I said a few weeks ago, the inflection point is here; something has to give. Now, this is my “I told you so” section.
Meta is limiting how engineers use Anthropic’s Claude Code and OpenAI’s Codex for AI model-building work, per The Information. Internal docs show Meta is concerned rival model outputs could leak into its own training data and trigger distillation issues.
OpenAI delays IPO to 2027.
Open Source AI models gaining traction:
Coinbase is burning more tokens but spending less on tokens by shifting to cheaper open-source models.
June 2026 data, including empirical token studies, confirms that open-weight models are massively capturing market share by taking over high-volume workflows. They are fundamentally compressing proprietary pricing power from below.
However, closed-source models still securely capture the most complex, high-value tasks. Because of this, developers are increasingly turning to dynamic multi-model routing to balance cost and capability.
NVIDIA CEO Jensen Huang views AI not as a single monolithic technology, but as a vast ecosystem of over 1.5 million highly specialized, targeted models built for specific industries. He advocates for an "open and proprietary" model approach, arguing that intelligence must be tailored to specific domains. (Got to applaud Jensen for having an open mind and figuring it out eventually.)
Microsoft CEO Satya Nadella is pushing the vision that every company should build and own AI models tailored to its own business context, data, and institutional knowledge. He argues against over-reliance on a few dominant AI giants, warning that outsourcing a firm’s learning to an external model risks hollowing out competitive advantage. (So, Satya finally seems to have woken up and decided to take a peek outside the tech bro echo chamber and actually talked to “rational“ Enterprise Customers).
Ford rehires more than 300 veteran human engineers after it says AI failed to deliver the same level of expertise.
Dario Hating on Open Source
Anthropic CEO Dario Amodei told lawmakers that open-source AI is moving down a “very dangerous path.” Oh, well!!! Reminds me of Microsoft CEO Steve Ballmer hating on Linux in the 2000s.
Steve Ballmer characterized Linux as "a cancer" that attached itself intellectually to everything it touched, and also likened it to "communism". Ballmer framed Linux as an unstable and legally dangerous choice for enterprise and government clients.
In 2003, Ballmer interrupted a skiing holiday to fly to Germany and plead with the mayor of Munich to abandon a large-scale government switch to Linux. Despite offering a 35% discount on Windows licenses, he was unsuccessful, and the city migrated its systems.
History doesn’t exactly repeat, but it sure does rhyme. Nothing against kids like Dario and a lot of AI executives repeating the mistakes of the past. It is just how humanity is.
Right at this moment, a parent somewhere in the world is teaching a child not to play with fire. The majority of children will still end up experimenting boldly with fire and quickly figure out how to be careful in the future. What all parents and, for that matter, experienced executives can do is provide enough creative leeway for learning, but, at the same time, ensure the damage incurred as part of the learning process is not beyond repair.
Unfortunately, with a lot of what is happening in the AI space, the adults seem to be missing in action or sidelined.
AWS CEO Discovers a New Religion
AWS CEO Matt Garman spoke highly critically of business leaders who suggest replacing entry-level developers entirely with AI, calling the idea "one of the dumbest things I've ever heard". He argues that ditching young talent is a massive misstep for long-term company health and innovation.
His argument for keeping junior staff comes down to three main points:
The Talent Pipeline: "Senior Engineers" can't be 3D printed. Garman notes that if you stop hiring recent college graduates and juniors today, you will have no experienced staff capable of tackling complex issues or reviewing AI-generated code in a decade.
The "Native" Advantage: Entry-level developers are usually the cheapest employees and are also the most enthusiastic and adept at leaning into AI coding tools.
Fundamental Learning: Instead of learning basic boilerplate, juniors can use AI to bypass tedious tasks, giving them more time to focus on problem-solving, critical thinking, and how the company's overarching systems actually fit together.
Well, should I say, I told you so? At least Matt is finally figuring it out. The good news is that we might be approaching the end of RetardMaxxing in the AI space. The first signs are definitely here, but that might not be good news for global markets.
Winners and Losers
On the winners and losers front, whatever I have said over the last few months still holds.
The SaaS ecosystem playbook remains intact as described on June 1.
Consulting companies will be winners in the long term as described two weeks ago, albeit they will remain volatile in the short term. Add IBM to the list.
As a lot of on-prem buildout is going to happen, DELL and HPE will be winners.
The seesaw between hyperscalers and semi players will continue as the headlines flow in. They will be quite volatile in the short term until all of them fall together in a generational flameout followed by a slow and steady long-term resurgence led by new winners.
Datacenter buildout beneficiaries other than semis will also remain under pressure in the short term as valuations reset and normalize.
AMZN and GOOGL remain a hold; however, any increase in CAPEX would lead to a sell-off.
Keep an eye on Mistral. They are the frontrunners in the Enterprise AI race.
I leave the portfolio management part to you. All I will say is: size it as per your risk profile.
Mohnish Pabrai, the famous value investor, had ~70% of his portfolio in Micron in 2023. He sold it all for a more than respectable gain of ~60%; however, he missed out on the current parabolic run. Mohnish has tried to rationalize the 2023 trade, but the key problem I believe was the size of Micron in his portfolio. He got shaken out. So, size it right so that you can hold through volatility and ride your conviction.
Global Macro Insights: Geopolitical Friction, Pipeline Pressures, and Corporate Resilience
This is the first time since I started this blog two years ago that I have waded into the macro debate. I guess, with all that is happening, the overall market will get swayed by macroeconomic headwinds, making it harder, but not impossible, to find alpha in individual names in the short term. Also, having been an engineer for more than half of my three-decade professional career makes me quite skeptical about macroeconomic pontifications, as there are ten different answers to the same problem. But then, I guess, the variety of it is what makes economics, for lack of a better word, spicy.
The Iran War Stand-off, Negotiations, and Supply-Side Shocks
Geopolitical friction in the Middle East remains the primary catalyst for global macroeconomic volatility. While the interim agreement brokered earlier this month provided temporary relief, ongoing skirmishes in the Strait of Hormuz, coupled with conflicting signals surrounding US-Iran technical talks in Doha, have kept the risk premium elevated.
The initial impact of the conflict has triggered significant supply-side distortions:
The Energy Security Challenge: The maritime blockades and disruptions in the Strait of Hormuz have impacted global crude and LNG flows. Though the U.S. remains insulated due to high domestic production, European and Asian economies face severe energy supply shocks, raising structural risks of stagflation in energy-dependent regions.
Second-Order Effects (Shortages & Logistics): The conflict has mutated from a pure energy crisis into a broader shipping and commodity crunch. The paralysis of key maritime corridors has severely impacted global logistics, causing a localized "grocery supply emergency" in parts of the Gulf and spiking transit insurance rates globally. Higher fuel and fertilizer costs are now projecting long-term structural increases in agricultural commodities, guaranteeing persistent non-core inflation.
The Inflation Pipeline: PPI Surge to Consumer Havoc
Recent data underscores a diverging but highly dangerous inflation narrative. While headline consumer numbers show some moderation from peak panic levels, the upstream production pipeline is flashing deep red.
Current Consumer Trends: CPI & PCE
CPI: US Headline CPI for May printed at 4.17% YoY (8.20% annualized over the past three months), proving that broad-based price pressures remain stubborn. Core CPI sits lower at 2.82% YoY, highlighting that the immediate problem is concentrated in food and energy.
PCE: The Federal Reserve’s preferred gauge, Headline PCE, hit 4.07% YoY in May, driven by energy inputs.
While consumer indexes reflect the lagging retail reality, the May Producer Price Index (PPI) advanced a staggering 1.1% MoM, pushing the unadjusted 12-month final demand index to 6.5% YoY, its highest level since late 2022.
More concerning is intermediate demand:
Processed goods for intermediate demand surged 13.3% YoY.
Unprocessed energy materials jumped 22.2% YoY.
The Bullwhip Transmission Delay
PPI acts as a leading indicator for consumer inflation. Historically, raw material and wholesale service increases take anywhere from 2 to 4 months to flow entirely into retail shelves. Because corporations cannot indefinitely absorb a 13.3% rise in intermediate input costs without decimating margins, a wave of delayed corporate pass-through is locked into the system. As these upstream costs clear the supply chain, expect a short-term wave of consumer price havoc that could delay central bank rate cuts across the globe.
Yes, rate cuts are coming in 2027 as demand destruction from a jump in electronic device prices and the eventual PPI transmission lead to a period of poor economic growth and low inflation. Eventually, AI-driven productivity gains and the resulting economic growth in 2028 and beyond would lead to inflation slowly ticking up, albeit moderately.
Corporate Earnings Cushion: AI CAPEX and Labor Resilience
Despite an aggressive inflation pipeline and geopolitical headwind, the equity market and corporate earnings are projected to remain remarkably healthy. The bear case of a classic stagflationary collapse is being aggressively countered by two structural pillars.
The Artificial Intelligence CAPEX Boom
We are witnessing an unprecedented secular capital expenditure cycle fueled by indiscriminate buyers that is insulated from short-term macroeconomic cyclicality. Big Tech and enterprise corporations are aggressively funding AI infrastructure, data centers, and advanced silicon.
This massive deployment of capital functions as an organic fiscal stimulus for the technology, industrial, and energy infrastructure sectors.
Unlike typical cyclical spending, many believe AI CAPEX is a structural arms race. Companies are prioritizing these investments even under high interest rates to avoid falling behind the technological curve, directly bolstering top and bottom lines for the B2B tech ecosystem. I believe the AI CAPEX cycle will eventually face headwinds as we head into the later part of the year or early 2027, driven by the reasons I have been highlighting.
A Steady and Resilient Job Market
The global consumer remains supported by a structurally resilient labor market. Wage growth, while moderating slightly, continues to track close to headline inflation, preventing a collapse in aggregate demand.
Low unemployment rates ensure that household balance sheets can absorb the upcoming PPI-driven retail price hikes without triggering a severe demand cliff; however, there will be short-term pain.
As long as employment holds steady, corporate revenues should withstand the margin pressure of rising input costs, allowing corporate earnings to sustain their positive momentum through the second half of the year.
A short-term carnage in the AI space wouldn’t impact other sectors as much, as other sectors are quite insulated. The stock market is another matter.
Playing Defense
As I said on June 1, the time has come to play defense. The following sectors should hold up well even as the markets work through the confluence of factors converging at the same time.
Healthcare, Biotech, and Consumer Staples: less crowded growth and defensive areas.
Regional banks: These have improved operationally significantly over the last year or so. In the short term, they should continue to do well. If you are interested in a particular bank and want to analyze it quickly, you may read my article “How to Analyze a Bank Quickly” from May last year.
Homebuilders: I discussed the housing space earlier this year, and as markets become convinced in 2027 that rate cuts are coming, homebuilders should do well.
Industrial Distributors: Industrial distributors such as GWW and FAST have pricing power, strong customer relationships, and should continue to do well.
Signals and Psychology
Cyclical stocks go parabolic from time to time because people feel this time is different. There is no point debating with people who feel this time is different. If they didn't exist, then we wouldn't have bubbles and crashes. Hence, instead of debating whether there is a bubble, it is best to evaluate incoming information on its merits, identify turning or inflection points, and play along, adjusting one’s portfolio.
Markets were never efficient. Reflexivity drives markets.
George Soros’s theory of reflexivity states that investors’ perceptions shape economic reality, and in turn, those altered realities influence investor perceptions. This continuous, two-way feedback loop means markets never reach true equilibrium, challenging the mainstream economic view that prices always accurately reflect market fundamentals.
A common question I get asked: Were there signals in the ‘90s and pre-GFC that trouble was brewing? Yes, there certainly were a boatload of them.
The ‘90s saw a lot of accounting sleight of hand and straight-up fraud. Every known trick to boost earnings was used, most of it legal but optically bad.
Waste Management was the biggest fraud before Enron. Among a slew of other things, they kept increasing the useful lives of their trucks and other assets to lower depreciation costs to boost earnings. Kept going on for years before a new CEO came and ordered a probe. (Hyperscalers increased the useful lives of servers and networking equipment from 3 to 6 years. There are parallels, but is it fraud? No.)
There was "Chainsaw" Al Dunlap. He was the rockstar restructuring CEO. Used to come in, slash costs, dress up the books, boost the stock price, and sell the company. His luck ran out with Sunbeam. (Constant restructurer currently? Meta).
AOL used to expense solicitation costs till 1994, which is the correct way to do it. However, in 1994, AOL started recording solicitation costs as assets on the Balance Sheet as "deferred membership acquisition costs". This sleight of hand was used to boost earnings to please Wall Street. Eventually, AOL got a slap on the wrist from the SEC.
IBM, the Blue Chip, was in on it too. In 1999, as it ran into a tough patch, gross margins began to fall, so it had to do something to boost operating income. It recorded the $4.1 billion sale of its Global Network business to AT&T as a reduction in SG&A.
Everybody knows about Enron. The thing with Enron was that it was flagged as fraud in 1998 by students at Cornell. Wall Street laughed at them. Another interesting story about Enron: Ken Lay was Bush's largest donor when he was running for Governor of Texas, and a big supporter during the Presidential Campaign. After Enron went bust, Bush was asked about Ken. Bush said he didn't know who Ken was.
Then there was Microstrategy. On March 20, 2000, MicroStrategy said they would have to restate earnings for 1997-1999. The stock fell from 226 to 86 in a single day. This was just a few weeks after PwC had blessed MSTR's 1999 financial reports.
To keep earnings boosted, WorldCom started recording a significant increase in line costs as assets. Essentially, capitalization of expenses. Not too many fans of FCF existed back then.
There were many such smaller cases, but as was the case back then, a lot of creative accounting has been happening recently. The frauds get highlighted and prosecuted when it is already too late for shareholders, and it can take years before anything happens. Only when the tide runs out do we know who is swimming naked.
CAPEX beneficiaries were making hay just like today. Meanwhile, the customer-facing companies, mostly the dot-coms, were not profitable. The dot-coms of today are the frontier labs.
The vanity metrics back then were DAUs and MAUs. Today, it is token usage. Slowly, being thrown into the dustbin. Once people realized the customer-facing companies had no path to profitability, overall demand from consumers and enterprises was overestimated, and that financing was running out before profitable demand would show up, the party was over.
Currently, we have a similar situation.
Coming to the pre-GFC period. There were early signals of big trouble in the housing space and risks to the financial system. Between February and March 2007, 25 subprime lending firms, including New Century Financial, declared bankruptcy. In June 2007, two Bear Stearns hedge funds collapsed, exposing the risky, intertwined nature of mortgage derivatives across global markets. However, markets continued to hold steady.
By mid 2008, most of the market was cheap; however, only after Lehman went kaput did we have the washout. Markets rallied from November to January, but started falling again until March 6, 2009. Obama said it was time to buy stocks that day, and the S&P 500 bottomed at 666.
As you can well imagine, that 666 number made for some rather interesting discussions.
That said, the one concerning thing about the whole AI bottleneck trade is that it is a bet against human ingenuity. Human ingenuity is the defining evolutionary superpower that allows our species to identify problems, invent unorthodox solutions, and reshape society. Betting against that has never ended well.
As I sign off today, I will pound the table and leave you with a recommendation, which I believe wouldn’t violate any SEC laws. If you have not, watch “The Good, The Bad, and The Ugly.”
