Soundtrack: Lynyrd Skynyrd — Free Bird
This piece is over 19,000 words, and took me a great deal of writing and research. If you liked it, please subscribe to my premium newsletter. It’s $70 a year, or $7 a month, and in return you get a weekly newsletter that’s usually anywhere from 5000 to 15,000 words, including vast, extremely detailed analyses of NVIDIA, Anthropic and OpenAI’s finances, and the AI bubble writ large. I am regularly several steps ahead in my coverage, and you get an absolute ton of value. In the bottom right hand corner of your screen you’ll see a red circle — click that and select either monthly or annual. Next year I expect to expand to other areas too. It’ll be great. You’re gonna love it.
If you have any issues signing up for premium, please email me at ez@betteroffline.com.
OpenAI told me opex keep eating his revenues, so I asked how many rounds of private equity he has burned and he said he just goes to the market and gets new equity afterwards so I said it sounds like he’s just feeding equity to opex and then Sam Altman started crying — @FraPippo428
One time, a good friend of mine told me that the more I learned about finance, the more pissed off I’d get.
He was right.
There is an echoing melancholy to this era, as we watch the end of Silicon Valley’s hypergrowth era, the horrifying result of 15+ years of steering the tech industry away from solving actual problems in pursuit of eternal growth. Everything is more expensive, and every tech product has gotten worse, all so that every company can “do AI,” whatever the fuck that means.
We are watching one of the greatest wastes of money in history, all as people are told that there “just isn’t the money” to build things like housing, or provide Americans with universal healthcare, or better schools, or create the means for the average person to accumulate wealth. The money does exist, it just exists for those who want to gamble — private equity firms, “business development companies” that exist to give money to other companies, venture capitalists, and banks that are getting desperate and need an overnight shot of capital from the Federal Reserve’s Overnight Repurchase Facility or Discount Window, two worrying indicators of bank stress I’ll get into later.
No, the money does not exist for you or me or a person. Money is for entities that could potentially funnel more money into the economy, even if the ways that these entities use the money are reckless and foolhardy, because the system’s intent on keeping entities alive incentivizes it. We are in an era where the average person is told to pull up their bootstraps, to work harder, to struggle more, because, as Martin Luther King Jr. once said, it’s socialism for the rich and rugged free market capitalism for the poor.
The “free market” is a fucking con. When you or I run out of money, our things are taken from us, we receive increasingly-panicked letters, we get phone calls and texts and emails and demands, we are told that all will be lost if we don’t “work it out,” because the financial system is not about an exchange of value but whether or not you can enter into the currently agreed-upon con.
By letting neoliberalism and the scourge of the free markets rule, modern society created the conditions for what I call The Enshittifinancial Crisis — the place at which my friend Cory Doctorow’s Enshittification Theory meets my own Rot Economy Thesis in a fourth stage of Enshittification.
Enshittification unfolds in three phases: first, a company is “good to users,” Doctorow writes, drawing people in droves, as funnel traps do Japanese beetles, with the promise of connection or convenience. Second, with that mass audience consolidated, the company is “good to business customers,” compromising some of its features so that the most lucrative clients, usually advertisers, can thrive on the platform. This second phase is the point at which, say, our Facebook feeds fill with ads and posts from brands. Third, the company turns the user experience into “a giant pile of shit,” making the platform worse for users and businesses alike in order to further enrich the company’s owners and executives.
I’ll walk you through it.
Facebook was a huge, free platform, much like Instagram, that offered fast and easy access to everybody you knew. It acquired Instagram in 2012 to kill off a likely competitor, and over time would start making both products worse — clickbait notifications, a mandatory algorithmic feed that deliberately emotionally manipulated people and stoked political division, eventually becoming full of AI slop and videos, all so that Meta could continue to sell billions of dollars of ads a quarter. Per Kyle Chayka of the New Yorker, “Facebook’s feed, now choked with A.I.-generated garbage and short-form videos, is well into the third act of enshittification.”
The third stage is critical, in that it’s when the company also turns on its business customers. A Marketing Brew story from September of last year told the tale of multiple advertisers who found their campaigns switching to different audiences, wasting their money and getting questionable results. A New York Times story from 2021 described companies losing upwards of 70% of their revenue during a Facebook ads outage, another from 2018 described how Meta (then Facebook) deliberately hid issues with its measurement of engagement on videos from advertisers for over a year, and more recently, Meta’s ads tools started switching out top-performing ads with AI-generated ones, in one case targeting men aged 30 to 45 with an AI-generated grandma, all without warning the advertiser.
Meta doesn’t give a shit, because investors and analysts don’t give a shit. I could say “sell-side analysts” here — the ones that are trying to get you to buy a stock — but based on every analyst report I’ve read from a major bank or hedge fund, I truly think everybody is complicit.
In November 2025, Reuters revealed that Meta projected in late 2024 that 10% of its annual revenue ($16 billion) would come from advertisements for scams or banned goods, mere weeks after Meta announced a ridiculous $27 billion data center debt package, one that used deep accountancy magic to keep it off of its balance sheet despite Meta guaranteeing the entirety of the loan.
One would think this would horrify investors for two reasons:
Meta’s business is both supporting and profiting from organized crime, and at 10% of its revenue, it’s also kind of dependent on it.Meta is using deliberate and insidious accounting tricks to act like a data center that it is paying to build and will be the sole tenant of is somehow an “off balance sheet” operation.
One would be wrong. Morgan Stanley said a few weeks ago that it is “one of the handful of companies that can leverage its leading data, distribution and investments in AI,” and raised its target to $750, with a $1000-a-share bull case. Wedbush raised Meta’s price to $920, and Bank of America staunchly held firm at…$810. I can find no analyst commentary on Meta making sixteen billion dollars on fraud, because it doesn’t matter to them, because this is the Rot Economy, and all that matters is number go up.
Reality — such as whether there’s any revenue in AI, or whether it’s a good idea that Meta is spending over $70 billion this year on capital expenditures on a product that has generated no revenue (and please, fucking spare me the bullshit around “Meta’s AI ads play,” that whole story is nonsense) — doesn’t matter to analysts, because stocks are thoroughly, inextricably enshittified, and analysts don’t even realize it’s happening.
The Great Enshittification Of The Stock Market
The stages of enshittification usually involve some sort of devil’s deal.
In Stage 1, things are good for users: the platform is free, things are easy-to-use, and thus it’s really simple for you and your friends to adopt *and become dependent on it.*In Stage 2, things become bad for consumers, but good for business customers: the platform begins forcing users to do “profitable” things — like show them more adverts by making search results worse — all while making it difficult to migrate to another one, either through locking in your data or the tacit knowledge that moving platforms is hard, and your friends are usually in one place. Businesses sink tons of money into the platform, knowing that users are unlikely to leave, and make good money buying ads against a populace that increasingly stays because it has to as there are *no other options.*In Stage 3, things become bad for consumers and businesses, but good for shareholders: the platforms begin to deteriorate to the point that usability is pushed to the brink, and businesses — who are now dependent on the platform because monopolies have pushed out every alternative platform to advertise or reach consumers — begin to see their product crumble, all in favour of shareholder capital, which only cares about stock value, net income and buybacks.
We have now entered Enshittification Stage 4, where businesses turn on shareholders.
Analysts and investors have become trapped in the same kind of loathsome platform play as consumers and businesses, and face exactly the same kinds of punishment through the devaluation of the stock itself. Where platforms have prioritized profits over the health and happiness of users or business customers, they are now prioritizing stock value over literally anything, and have — through the remarkable growth of tech stocks in particular — created a placated and thoroughly whipped investor and analyst sect that never asks questions and always celebrates whatever the next big thing is meant to be.
The value of a “stock” is not based on whether the business is healthy, or its future certain, but on its potential price to grow, and analysts have, thanks to an incredible bull run of tech stocks going on over a decade, been able to say “I bet software will be big” for most of the time, going on CNBC or Bloomberg and blandly repeating whatever it is that a tech CEO just said, all without any worries about “responsibility” or “the truth.”
This is because big tech stocks — and many other big stocks, if I’m honest — have made their lives easy as long as they don’t ask questions. Number always seems to be going up for software companies, and all you need to do is provide a vociferous defense of the “next big thing,” and come up with a smart-sounding model that justifies eternal growth.
This is entirely disconnected from the products themselves, which don’t matter as long as Number Go Up. If net income is high and the company estimates it will continue to grow, then the company can do whatever the fuck it want with the product it sells or the things that it buys. Software Has Eaten The World in the sense that Andreesen got his wish, with investors now caring more about the “intrinsic value” of software companies rather than the businesses or products themselves.
And because that’s happening, investors aren’t bothering to think too hard about the tech itself, or the deteriorating products underlying tech companies, because “these guys have always worked it out” and “these companies have always managed to keep growing.” As a result, nobody really looks too deep. Minute changes to accounting in earnings filings are ignored, egregious amounts of debt are waved off, and hundreds of billions of dollars of capital expenditures are seen as “the new AI revolution” versus “a huge waste of money.”
By incentivizing the Rot Economy — making stocks disconnected from the value of the company beyond net income and future earnings guidance — companies have found ways to enshittify their own stocks, and shareholders will be the ones who suffer, all thanks to the very downstream pressure that they’ve chosen to ignore for decades.
You see, while one might (correctly) see that the deterioration of products like Facebook and Google Search was a sign of desperation, it’s important to also see it as the companies themselves orienting around what they believe analysts and investors want to see.
You can also interpret this as weakness, but I see it another way: stock manipulation, and a deliberate attempt to reshape what “value” means in the eyes of customers and investors. If the true value of a stock is meant to be based on the value of its business, cash flow, earnings and future growth, a company deliberately changing its products is an intentional interference with value itself, as are any and all deceptive accounting practices used to boost valuations.
But the real problem is that analysts do not…well…analyze, not, at least, if it goes against the market consensus. That’s why Goldman Sachs and JP Morgan and Futurum and Gartner and Forrester and McKinsey and Morgan Stanley all said that the metaverse was inevitable — because they do not actually care about the underlying business itself, just its ability to grow on paper.
Need proof that none of these people give a fuck about actual value? Mark Zuckerberg burned $77 billion on the metaverse, creating little revenue or shareholder value and also burning all that money without any real explanation as to where it went.
The street didn’t give a shit because meta’s existent ads business continued to grow, same as it didn’t give a shit that Mark Zuckerberg burned $70 billion on capex, even though we also really don’t know where that went either.
In fact, we really have no idea where all this AI spending is going. These companies don’t tell us anything. They don’t tell us how many GPUs they have, or where those GPUs are, or how many of them are installed, or what their capacity is, or how much money they cost to run, or how much money they make. Why would we? Analysts don’t even look at earnings beyond making sure they beat on estimates. They’ve been trained for 20 years to take a puddle-deep look at the numbers to make sure things look okay, look around their peers and make sure nobody else is saying something bad, and go on and collect fees.
The same goes for hedge funds and banks propping up these stocks rather than asking meaningful questions or demanding meaningful answers. In the last two years, every major hyperscaler has extended the “useful life” of its servers from 3 years to either 5.5 or 6 years — and in simple terms, this allowed them to incur a smaller depreciation expense each quarter as a result, boosting net income. Those who are meant to be critical — analysts and investors sinking money into these stocks — had effectively no reaction, despite the fact that Meta used (per the Wall Street Journal) this adjustment to reduce its expenses by $2.3 billion in the first three quarters of this year.
This is quite literally disconnected from reality, and done based on internal accounting that we are not party to. Every single tech firm buying GPUs did this and benefited to the tune of billions of dollars in decreased revenues, and analysts thought it was fine and dandy because number went up.
Shareholders are now subordinate to the shares themselves, reacting in the way that the shares demand they do, being happy for what the companies behind the shares give them, and analysts, investors and even the media spend far more energy fighting the doubters than they do showing these companies scrutiny.
Much like a user of an enshittified platform, investors and analysts are frogs in a pot, the experience of owning a stock deteriorating since Jack Welch and GE taught corporations that the markets are run with the kind of simplistic mindset built for grifter exploitation.
And much like those platforms, corporations have found as many ways as possible to abuse shareholders, seeing what they can get away with, seeing how far they can push things as long as the numbers look right, because analysts are no longer looking for sensible ideas.
Let me give you an example I’ve used before. Back in November 1998, Winstar Communications signed a “$2 billion equipment and finance agreement with Lucent Technologies” where Winstar would borrow money from Lucent to buy stuff from Lucent, all to create $100 million in revenue over 5 years.
In December 1999, Barron’s wrote a piece called “In 1999 Tech Ruled”:
George Gilbert, who manages the Northern Technology Fund , predicts the Web-centric worlds of consumer services and software will fare well next year, too.“A lot of people are increasing their access to the Internet,” says Gilbert. "And e-commerce and business networking are very high priorities for the Fortune 100."Lawrence York, lead portfolio manager of the WWW Internet Fund, is bullish on semiconductors, telecommunications and business-to-business – or B2B – e-commerce software. But he’s wary of online retailers. “That model won’t work long term,” he asserts.His top B2B picks? Ariba and Official Payments. In wireless, he likes Winstar , Ciena and AirNet Communications , which went public earlier this month.
Airnet? Bankrupt. WinStar? Horribly bankrupt. While Ciena survived, it had spent over a billion dollars to acquire other companies (all stock, of course), only to see its revenue dwindle basically overnight from $1.6bn to $300 million as the optical cable industry collapsed.
One would have been able to work out that Winstar was a dog, or that all of these companies were dogs, if you were to look at the numbers, such as “how much they made versus how much they were spending.” Instead, analysts, the media and banks chose to pump up these stocks because the numbers kept getting bigger, and when the collapse happened, rationalizations were immediately created — there were a few bad apples (Enron, Winstar, WorldCom), “the fiber was useful” and thus laying it was worthwhile, and otherwise everything was fine.
The problem, in everybody else’s mind, was that everybody had got a bit distracted and some companies that weren’t good would die. All of that lost money was only a problem because it didn’t pay off. This was a misplaced gamble, and it taught tech executives one powerful lesson: earnings must be good, without fail, by any means necessary, and otherwise nothing else matters to Wall Street.
It’s all about incentives. A sell-side analyst that tells you not to buy something is a problem. A journalist that is skeptical or critical of an industry in the midst of a growth or hype cycle is considered a “hater” — don’t I fucking know it. Analysts that do not sing the same tune as everybody else are marginalized, mocked and aggressively policed.
And I don’t fucking care. Stop being fucking cowards. By not being skeptical or critical you are going to lead regular people into the jaws of another collapse.
The dot com bubble was actually a great time to start reevaluating how and why we value stocks — to say “hey, wait, that $2 billion deal will only make $100 million in revenue?” or “this company spends $5 for every $1 it makes!” — but nobody, it appears, remained particularly suspicious of the tech industry, or a stock market that was increasingly orienting itself around conning shareholders.
And because shareholders, analysts and the media alike refused to retain a single shred of suspicion leaving the dot com era, the mania never actually subsided. Financial publications still found themselves dedicated to explaining why the latest hype cycle was real. Journalists still found themselves told by editors that they had to cover the latest fad, even if it was nonsensical or clearly rotten. Analysts still grabbed their swords and rushed to protect the very companies that have spent decades misleading them.
Much like we spent years saying that Facebook was a “good deal” because it was free, analysts and investors say tech stocks are “great to hold” because they kept growing, even if the reason they “kept growing” was a series of interlocking monopolies, difficult-to-leave platforms and impossible-to-fight traction and pricing, all of which have an eventual sell-by date.
I realize I’m pearl-clutching over the amoral status of capitalism and the stock market, but hear me out: what if we’re actually in a 15-to-20-year-long knife-catching competition? What if all anybody has done is look at cashflow, net income, future growth guidance, and called it a day? A lack of scrutiny has allowed these companies to do effectively anything they want, bereft of worrisome questions like “will this ever make a profit?”
What if we basically don’t know what the fuck is going on? What if all of this was utterly senseless?
AI Accelerated The Enshittification Of The Stock Market
As I wrote last year, the tech industry has run out of hypergrowth ideas, facing something I call “the Rot Com bubble.” In simple terms, they’re only “doing AI” because there do not appear to be any other viable ideas to continue the Rot Economy’s eternal growth-at-all-costs dance.
Yet because growth hasn’t slowed yet, analysts, the media and other investors are quick to claim that AI is “paying off,” even if nobody has ever said how much AI revenue is being generated or, in the case of Salesforce, they can say “nearly $1.4 billion ARR,” which sounds really big until you realize a company with $10.9 billion in revenue is boasting about making less than $116 million in revenue in a month.
Nevertheless, because Salesforce set a new revenue target of $60 billion by 2030, the stock jumped 4%. It doesn’t matter that most Agentforce customers don’t pay for the service, or that AI isn’t really making much money, or really anything, other than Number Go Up.
The era we live in is one of abject desperation, to the point that analysts and investors — and shareholders by extension — will take any abuse from management. They will allow companies to spend as much money as they want in whatever ways they want, as long as it continues the charade of “number go up.”
Let me spell it out a little more, using the latest earnings of various hyperscalers as an example.
According to its latest quarterly filings, Microsoft spent $34.9 billion on capital expenditures, Amazon $34.2 billion, Meta $19.37 billion, and Google $24 billion.The common mantra is that these companies are “spending all this money on GPUs,” but that doesn’t match up with NVIDIA’s revenues. NVIDIA’s last quarterly earnings said that four direct customers made up more than 10% of revenue — 22% ($12.54bn), 15% ($8.55bn), 13% ($7.41bn) and 11% ($6.27bn) out of $57 billion. While this sort of lines up with capex spend, it doesn’t if you shift back a quarter, when Microsoft spent $21.4 billion, Meta $17.01 billion, Amazon $31.4 billion and Google $22.4 billion, with the vast majority on “technical infrastructure.” In the same quarter, NVIDIA had only two customers that accounted for more than 10% — one 23% ($10.7bn) and one 16% ($7.47bn) out of $46.7 billion.Another quarter back, and Microsoft spent $22.6 billion, Meta $13.69 billion, Google $17.2 billion and Amazon $22.4 billion. In the same quarter, NVIDIA had two customers accounting for more than 10% of revenue — 16% ($7.49bn) and 14% ($6.168bn).Where, exactly, is all this money going? In Microsoft’s latest earnings (Q1FY26), it said that $19.39 billion went to “additions to property and equipment,” with “roughly half of [its total capex] spend on short-lived assets, primarily GPUs and CPUs.” A quarter (Q4FY2025) back, additions to property and equipment were $16.74 billion, with “roughly half…[spent] on long-lived assets that will support monetization over the next 15 years and beyond.” Let’s assume that Microsoft is NVIDIA’s biggest customer every single quarter — customer A, spending $12.5 billion (out of $34.9 billion), $10.7 billion (out of $21.4 billion) and $7.049 billion (out of $22.6 billion) a quarter.Assuming that Microsoft is only buying NVIDIA’s Blackwell GPUs (forgive the model numbers, but it’s based on my own modeling. Let’s say 40% B200s, 30% GB200s, 10% B300s and 20% GB300s), that works out to about 457MW of IT load for Q1FY26, 391MW for Q4FY25 and (adjusting to include more H200s, as the B300/GB300s were not shipping yet) 263MW for Q3FY25.Has Microsoft built 1.11GW of data centers in that time? Apparently! It claims it added 2GW in the last year, but Satya Nadella claimed in November that Microsoft had chips in inventory it couldn’t install due to a lack of power.In any case, where did the remaining $22.4 billion, $11.9 billion and $15.5 billion in capex flow? We know there are finance leases. What for? More GPUs? What is the actual output of these expenditures?
We have no idea, because analysts and investors are in an abusive relationship with tech stocks. It is fundamentally insane that Microsoft, Meta, Amazon and Google have spent $776 billion in capital expenditures in the space of three years, and even more so that analysts and investors, when faced with such egregious numbers, simply sit back and say “they’re building the infrastructure of the future, baby!” Analysts and traders and investors and reporters do not think too hard about the underlying numbers, because doing so immediately makes you run head-first into a number of worrying questions such as “where did all that money go?” and “will any of this pay off?” and “how many GPUs do they actually own?”
Analysts have, on some level, become the fractional marketing team for the stocks they’re investing in. When Oracle announced its $300 billion deal with OpenAI in September — one that Oracle does not have the capacity to fill and OpenAI does not have the money to pay for – analysts heaved and stammered like horny teenagers seeing their first boob:
John DiFucci from Guggenheim Securities said he was “blown away.” TD Cowen’s Derrick Wood called it a “momentous quarter.” And Brad Zelnick of Deutsche Bank said, “We’re all kind of in shock, in a very good way.”“There’s no better evidence of a seismic shift happening in computing than these results that you just put up,” Zelnick said on the earnings call.
These are the same people that retail and institutional investors rely upon for advice on what stocks to buy, all acting with the disregard for the truth that comes from years of never facing a consequence. Three months later, and Oracle has lost basically all of the stock bump it saw from the OpenAI deal, meaning that any retail investor that YOLO’d into the trade because, say, analysts from major institutions said it was a good idea and news outlets acted like this deal was real, already got their ass kicked.
And please, spare me the “oh they shouldn’t trade off of analysts” bullshit. That’s the kind of victim-blaming that allows these revered fuckwits to continue farting out these meaningless calls.
In reality, we’re in an era of naked, blatant, shameless stock manipulation, both privately and publicly, because a “stock” no longer refers to a unit of ownership in a company so much as it is a chip at a casino where the house constantly changes the rules. Perhaps you’re able to occasionally catch the house showing its hand, and perhaps the house meant for you to see it. Either way, you are always behind, because the people responsible for buying and selling stocks at scale under the auspices of “knowing what’s going on” don’t seem to know what they’re talking about, or don’t care to find out.
Let’s walk through the latest surge of blatant stock manipulation, and how the media and analysts helped it happen.
Oracle, September 10 2025
Oracle announces its unfillable, unpayable $300 billion deal with OpenAI, leading to 30%+ bump in stock price. Analysts, who should ostensibly be able to count, call it “momentous” and say they’re “in shock.” On September 22 2025, CEO Safra Catz steps down, and nobody seems to think that’s weird or suspicious.
Two months later, Oracle’s stock is down 40%, with investors worried about Oracle’s growing capex, which is surprising I suppose if you didn’t think about how Oracle would build the fucking data centers.
Basically anyone who traded into this got burned.
NVIDIA, September 22 2025
NVIDIA announced a “strategic partnership” to invest “up to $100 billion” and build 10GW of data centers with OpenAI, with the first gigawatt to be deployed in the second half of 2026. Where would the data centers go? How would OpenAI afford to build them? How would OpenAI build a gigawatt in less than a year? Don’t ask questions, pig!
NVIDIA’s stock bumped from from $175.30 to $181 in the space of a day. The media wrote about the story as if the deal was done, with CNBC claiming that “the initial $10 billion tranche [was] expected to close within a month or so once the transaction has been finalized.” I read at least ten stories that said that “NVIDIA had invested $100 billion.”
Analysts would say that NVIDIA was “locking in OpenAI” to “remain the backbone of the next-gen AI infrastructure,” that “demand for NVIDIA GPUs is effectively baked into the development of frontier AI models,” that the deal “[strengthened] the partnership between the two companies…[and] validates NVIDIA’s long-term growth numbers with so much volume and compute capacity.” Others would say that NVIDIA was “enabling OpenAI to meet surging demand.”
Three analysts — Rasgon at Bernstein, Luria at D.A. Davidson and Wagner at Aptus Capital — all raised circular deal concerns, but they were the minority, and those concerns were still often buried under buoyant optimism about the prospects of the company.
One eensy weensy problem though, everyone! This was a “letter of intent” — it said so in the announcement! — and on NVIDIA’s November earnings, it said that it “entered into a letter of intent with an opportunity to invest in OpenAI.”
It turns out the deal didn’t exist and everybody fell for it! NVIDIA hasn’t sent a dime and likely won’t. A letter of intent is a “concept of a plan.”
SK Hynix, Samsung, October 1 2025
Back in October, Reuters reported that Samsung and SK Hynix had “signed letters of intent to supply memory chips for OpenAI’s data centers,” with South Korea’s presidential office saying that said chip demand was expected to reach “900,000 wafers a month,” with “much of that from Samsung and SK Hynix,” which was quickly extrapolated to mean around 40% of global DRAM output.
Stocks in both companies, to quote Reuters, “soared,” with Samsung climbing 4% and SK Hynix more than 12% to an all-time high. Analyst Jeff Kim of KB Securities said that “there have been worries about high bandwidth memory prices falling next year on intensifying competition, but such worries will be easily resolved by the strategic partnership,” adding that “Since Stargate is a key project led by President Trump, there also is a possibility the partnership will have a positive impact on South Korea’s trade negotiations with the U.S.”
Donald Trump is not “leading Stargate.” Stargate is a name used to refer to data centers built by OpenAI. KB Securities has around $43 billion of assets under management. This is the level of analysis you get from these analysts! This is how much they know!
On SK Hynix’s October 29 2025 earnings call, weeks after the announcement, its CEO, Kim Woo-Hyun, was asked a question about High Bandwidth Memory growth by SK Kim from Daiwa Securities:
Kim: Thank you very much for taking my question. It is on demand. Now, there have been a series of announcements of GPU and ASIC supply cooperation between Big Techs and AI companies, fueling expectations of further AI market growth. Then, against this backdrop, what is the company’s outlook on HBM demand growth, as well as a broadening of the customer base?SK Hynix: Thank you for the question. Now, with upward adjustment in Big Tech’s CapEx and increased investment by AI companies, the HBM market, even by a conservative estimate, will keep growing at an average of over 30% for the next five years.I will point to our recent LOI with OpenAI for large-scale DRAM supply as an example of the very strong market demand for AI, as well as the need to secure AI memory based on HBM more than anything else when developing AI technology.
This is the only mention of OpenAI. Otherwise, SK Hynix has not added any guidance that would suggest that its DRAM sales will spike beyond overall growth, other than mentioning it had “completed year 2026 supply discussions with key customers.” There is no mention of OpenAI in any earnings presentation.
On Samsung’s October 30 2025 earnings call, Samsung mentioned the term “DRAM” 18 times, and neither mentioned OpenAI nor any letters of intent.
In its Q3 2025 earnings presentation, Samsung mentions it will “prioritize the expansion of the HBM4 [high bandwidth memory 4] business with differentiated performance to address increasing AI demand.”
Analysts do not appear to have noticed a lack of revenue from an apparent deal for 40% of the world’s RAM! Oh well! Pobody’s nerfect!
Both Samsung and SK Hynix’s stocks have continued to rise since, and you’d be forgiven for thinking this deal was something to do with it, even though it wasn’t.
AMD, October 5, 2025
AMD announced that it had entered a “multi-year, multi-generation agreement” with OpenAI to build 6 GW of data centers, with “the first 1GW deployment set to begin in the second half of 2026,” calling the agreement “definitive” with terms that allowed OpenAI to buy up to 10% of AMD’s stock, vesting over “specific milestones” that started with the first gigawatt of data center development. Said data centers would also use AMD’s yet-to-be-released MI450 GPUs. The deal would, per Reuters, bring in “tens of billions of dollars of revenue.”
Where would those data centers go? How would OpenAI pay for them? Would the chips be ready in time? Silence, worm! How dare you ask questions? How dare you? Why are you asking questions? NUMBER GO UP!
[Content truncated due to length…]
You must log in or # to comment.