2025-11-27 23:46:52
While today is Thanksgiving in America, I'm not in America, so I thought it might be a good day to write about a topic where I am: the UK. Since my family moved here about two and a half years ago, I've slowly but surely been getting up to speed on the local things that one must – or at the very least, one can't avoid: taxes and politics. Both were in the spotlight yesterday with the announcement of the budget.
And there's a tech angle here, I swear.
2025-11-26 22:17:05
You know how I know that NVIDIA is concerned about Google's encroachment on their market? They won't shut up about how they're not concerned about it. As relayed by Kif Leswing of CNBC:
Nvidia on Tuesday said its tech remains a generation ahead of the industry, in response to Wall Street’s concerns that the company’s dominance of AI infrastructure could be threatened by Google’s AI chips.
“We’re delighted by Google’s success — they’ve made great advances in AI and we continue to supply to Google,” Nvidia said in a post on X. “NVIDIA is a generation ahead of the industry — it’s the only platform that runs every AI model and does it everywhere computing is done.”
I feel like a psychoanalyst could work wonders on that statement – including the fact that it's seemingly meant to be the royal "we", as it was sent from the corporate account, but it sure feels like it was written by or for one person who happens to be in charge of said company. Anyway, I don't know Sigmund Freud or Carl Jung, but I'm pretty sure my friend ChatGPT does, so I asked them to analyze the statement:
That has a passive-aggressive flavor, even though it's wrapped in corporate PR language. Here’s why:
💬 What the message says (on the surface)
• “We’re delighted by Google’s success.”
• “We continue to supply to Google.”
• “We’re the only platform that runs every AI model.”
All of that is framed as positive, collaborative, upbeat.
😏 What the message implies (between the lines)
• “Congrats to Google… but remember, they still rely on us.”
• “Google is succeeding thanks in part to our hardware.”
• “Whatever Google built, we’re still a generation ahead.”
• “They’re great… but we’re better, and essential.”
That contrast — praising someone while subtly asserting superiority — is a classic form of corporate passive-aggression.
Why it reads passive-aggressive
• The praise feels performative, not genuine.
• The second sentence immediately pivots to asserting dominance (“NVIDIA is a generation ahead…”).
• The phrasing implies: “Let’s be clear: Google’s success doesn’t undermine us. In fact, it shows how central we are.”
Yeah, that's roughly my read. And the read of more or less everyone on the internet. And thus marks the second day in a row that NVIDIA has sort of stepped in it with an overly aggressive (even if passive) self-own.
If the first two rules of Bubble Club are not to talk about Enron, the third rule is probably not to so overtly betray your fears in public under the thin veneer of competitive praise.
Look, we get it. After a stock run the likes of which has never been seen at this scale, culminating in not just NVIDIA becoming the most valuable company in the world, but being the first stock to surpass both a $4T and a $5T market cap, the past couple of weeks have been rough. But "rough" is also relative, NVIDIA is still a $4.3T company, and still the most valuable company in the world. But yes, Apple and... Google are quickly catching up as NVIDIA falls. It's a CEO's job to worry about such things, but it must be especially worrying when your best mechanism to combat this – and by far the hardest to pull off – crushing your earnings expectations, didn't really work to combat the drop.1
And that's a problem because it suggests the market is shifting from exuberance – be it rational or irrational – to irrational fear. And if that takes hold... watch out below. Not just NVIDIA, but everyone. But especially NVIDIA.
So that's clearly why we're getting a response to the concerns by short-sellers and now to the market's reaction to Google's recent success. This is Jensen Huang trying to reassure everyone that everything is all good and quite rational. But especially NVIDIA.
And he might not be wrong! Back to Leswing:
In its post, Nvidia said its chips are more flexible and powerful compared with so-called ASIC chips — such as Google’s TPUs — which are designed for a single company or function. Nvidia’s latest generation of chips are known as Blackwell.
“NVIDIA offers greater performance, versatility, and fungibility than ASICs,” Nvidia said in its post.
There is certainly a world in which the flexibility that NVIDIA's GPUs offer over Google's TPUs ends up mattering. And perhaps a lot – especially if we're on the cusp of moving beyond, or even just extending from, being mainly focused on LLMs and instead are about to veer off into other types of AI models, such as "World Models".
(Though it's also clear that Demis Hassibas is himself focused on such models and as such, undoubtedly has Google's chips focused on how to train them as well.)
Google's TPUs look great right now because they seem to be rivaling what NVIDIA's GPUs can do in terms of model outputs – based solely on the latest versions of Gemini. And even if NVIDIA is "a generation ahead", the TPUs are likely quite a bit more efficient by their nature of being purpose-built. Another fear (and eye-opening moment) in this news cycle for NVIDIA is probably the references to the notion that Google may have cracked something new in pre-training whereas the other models seem sort of "stuck" on their current trajectories in that regard. And that could imply there's something about the TPUs that helped with that breakthrough, whatever it may be. Or it may not imply that at all! Nobody knows.
Well, someone does. But they're not talking. So the rest of us are left guessing. (Or are they talking in the form of passive aggressive tweeting? Just asking...)
NVIDIA, of course, has absolutely owned the pre-training portion of AI to date – so much so that the other would-be competitors are mainly focused on other parts of the build-out, such as inference. So much so that they have become a $5T company! And while Google is believed to have been using TPUs to train Gemini for at least a couple of generations now, the fact that they still are one of the largest buyers of NVIDIA chips suggests there's still something they felt the need to use GPUs for as well – as NVIDIA implies in their statement!
Though that's a bit gray as well since a lot of Google's NVIDIA bill is undoubtedly to be able to host their chips on Google Cloud for others to use...
“We are experiencing accelerating demand for both our custom TPUs and Nvidia GPUs,” a Google spokesperson said in a statement. “We are committed to supporting both, as we have for years.”
That's an interesting and seemingly very specific phrasing and framing of a PR statement in response to NVIDIA's own PR statement! Such wording seems to suggest that Google's need for NVIDIA's chips are not for their own purposes but rather for their customers (of Google Cloud). But again, it's sort of gray! Undoubtedly on purpose!
Anyway, the reports that not only is Meta potentially partnering with Google to use TPUs, but is said to want to acquire them to use them for training of new models... well, it certainly makes you at least think about NVIDIA's ongoing dominance there.
Unlike Nvidia, Google doesn’t sell its TPU chips to other companies, but it uses them for internal tasks and allows companies to rent them through Google Cloud.
But they're clearly about to... And we're quickly going to learn and hear a lot about price, performance, power consumption, etc. Any of which could be bad news for NVIDIA. And any combination of which could be even worse news.2 For the first time, there could be real competition in the market. Imagine that.
And you can probably just follow NVIDIA's actions here to get a sense of how they actually feel about Google entering the market. From The Information report by Amir Efrati, Erin Woo, and Anissa Gardizy:
Whether Google succeeds in its TPU efforts, the mere specter of a powerful alternative to Nvidia may have already yielded benefits for big Nvidia customers like Anthropic and OpenAI that don’t want to rely on a single AI chip provider.
Last month, following a Google deal to provide up to 1 million TPUs to Anthropic, Huang announced another deal to invest billions of dollars into Anthropic. He also got a commitment from the AI startup to use Nvidia GPUs.
Similarly, after it became publicly known OpenAI was planning to rent TPUs from Google Cloud, Huang struck a tentative deal to invest up to $100 billion in the ChatGPT maker so it could develop its own data centers, and has discussed leasing out Nvidia GPUs to the company.
Yeah, we're totally about to see some sort of massive NVIDIA/Meta deal, aren't we?
At the same time, the fact that the entire AI chip market is still supply constrained suggests NVIDIA's business probably doesn't have too much to worry about; certainly not in the short term. And if Google's pre-training breakthrough isn't something specific to their TPUs, it also suggests that LLM scaling, as we've known it to date, will likely continue for a while yet. And that's good news for the status quo, which is naturally good news for NVIDIA.
Even if Google is able to start selling TPUs to third-parties, it will undoubtedly take a long time to ramp such production – probably years.3 And in this market, no one can afford to wait, quite literally.
But all of the above is nuanced and hard to um, grok – certainly for most mom & pop investors, but even some Wall Street analysts, it seems. So NVIDIA is going down their own defensive path here. Which is misguided, but you get why they're doing it. The market has some jitters and the company atop the market is doing their best to try to calm those to keep the good times rolling.
So sure, NVIDIA is "delighted by Google's success" in many ways. But they're also less "delighted" in other ways, naturally. And that relationship is seemingly going to get more complicated going forward...
M.G. Siegler
M.G. Siegler
M.G. Siegler
M.G. Siegler
M.G. Siegler
1 One has to wonder if the true "break glass" mechanism NVIDIA tries to use here involves China. If Jensen Huang can somehow convince the powers that be to let him sell there again it will give the stock bulls a new lease and outlook on life. But clearly he's having trouble convincing both the US and China to allow for this, despite some extraordinary concessions... ↩
2 My mind for some reason drifts to when Apple Silicon entered the CPU arena with a purpose-built chip to take on Intel. The incumbent kept touting top-end performance without realizing that the market had shifted to prize efficiency more... I'm not saying this market will play out the same way, but it's certainly interesting to note. ↩
3 And I'm not even getting into NVIDIA's software advantage here thanks to CUDA. Yes, many are trying to bridge that moat – including Google – but it's still a huge advantage their chips have that will take time to shift, if nothing else. ↩
2025-11-26 01:23:32
It is wild how fast stocks can swing – but especially multi-trillion-dollar stocks. Case in point: with NVIDIA's fall from grace the past few sessions (maybe don't bring up Enron in your memos?), Apple is on the verge of overtaking them as the market cap leader once again. Yes, Apple. Not just the AI laggard stock, but the company which spent much of the last year as the AI laughingstock.
As it turns out, perhaps there was some upside in not committing to trillions of dollars in CapEx spend. At least right now.
But it's not just that because it's not just Apple. Google has rocketed higher than the rest of Big Tech over the past several months. Just this past June, under threat of antitrust breakup, I was writing about how badly the market was undervaluing the company.1 That wasn't really even about AI, but everything else they were doing, from Waymo on down. Not only were they a trillion behind Apple at the time, they were about to fall two trillion behind NVIDIA and Microsoft. They were behind Amazon. Hell, they were almost lapped by Meta!
Just a few months later and the tables have turned. Thanks undoubtedly to the notion that not only is Google no longer behind in AI, but they might all of a sudden be ahead, they're also pretty close to overtaking NVIDIA in market cap. And that also puts them close to Apple and thus, close to the crown. Again, in June they were in 4th place in the "race" – and much closer to 5th than 4th, less than $200B ahead of Meta. Now they're $2.3T ahead of Meta.
None of this really matters of course – and these massive swings show just how little all of it matters. (Though don't let these companies tell you they don't care about this at all – of course they do, from Steve Jobs on down, they're all competitive and this is a measuring stick, albeit a sometime arbitrary one.) But it is an interesting gauge to read investor sentiment of these companies over time.
A year ago, Apple was king of the world. After getting passed by not only their old foe Microsoft but also their sometimes foe NVIDIA earlier in the year, they came roaring back. In November 2024, the rankings looked like this:
Six months later, in May 2025, it was a different story:
But by July, just two months later, boom went Microsoft, but even more boom went NVIDIA – both crossed $4T that month, the first companies to ever do so (after Apple had been the first to $1T, $2T, and $3T):
By October, just three months after that, NVIDIA hit the mythical $5T and was running away:
And here we are now, mid-day on November 25, 2025:
In one year, that's NVIDIA gaining $900B in market cap – but also losing about $900B from their peak. Apple gained $500B after losing about $500B at one point. Microsoft gained $400B but are also about $600B off their highs for the year. Amazon gained $300B and has been the most stable of the group. Meta gained $100B and remained in the six spot the entire year after nearly catching Google just a few months ago, as noted.
Google hasn't held the top market cap spot since a couple brief moments in 2016 (I was there!), but it feels inevitable that they'll retake it again soon. That is unless NVIDIA can find a way to stop sending memos mentioning their business in the same breath as Enron. Stop doing that and they may just have another run or two in them...
M.G. Siegler
M.G. Siegler
M.G. Siegler
1 FWIW, my math at the time suggested Alphabet, then valued at $2T, should be valued closer to $3.5T, and probably higher, if they were giving all of the other businesses beyond Google more of a fair market value. And well, here we are! ↩
2025-11-25 18:31:29
The first rule of Bubble Club is you do not talk about Enron.
The second rule of Bubble Club is you do not talk about Enron.
NVIDIA, it seems, didn't get the memo, as they sent a memo to Wall Street analysts mentioning the-company-who-shall-not-be-named multiple times...
2025-11-24 23:51:23
After his last op-ed in The New York Times completely and utterly failed to sway the judge in Google's antitrust case to enact harsh remedies, Tim Wu is back with another essay, aiming to escalate the general argument to a higher level. Here, his target is Meta, which just won their own antitrust case, with a different judge ruling that the government failed to make the case that the company had an illegal monopoly in the social apps space. "How Can Anyone Seriously Doubt Meta Is a Monopoly?" is the headline I got served in my RSS feed, but it has seemingly since been tweaked to "The Bad Reasoning in the Meta Antitrust Ruling Isn’t Even the Worst Part". The first headline was clearly better – albeit silly, as a federal judge literally just did doubt that, legally – the second one is a mouthful, but again, aims to take the argument broader.
In this way, Wu's op-ed is similar to one that the NYT ran last February from Lina Khan. It was a remarkable read because nearly every point she made was wrong. And now Wu is trying to give her a run for her money.
But to dwell on the shortcomings of Judge Boasberg’s reasoning is to overlook something even worse: the message that his ruling sends to the country. The United States is in a precarious moment when it comes to the rule of law and democratic accountability. Unchecked wealth and corporate power have led many to doubt that our system is fair. Judge Boasberg’s ruling only fortifies the impression that the world’s wealthiest and most powerful corporations are above the law.
I mean, maybe the ruling gives that sense to others, maybe it doesn't. But obviously it's not the judge's job to make legal rulings so that the country will nebulously feel better about inequality. This is, basically the argument Wu attempts to make in the entire peace. Forget the law, judge, think about the vibes!
A chief goal of the Sherman Antitrust Act of 1890, which the government accused Meta of violating, is to make clear who is actually in charge of the county. As President Theodore Roosevelt explained in 1901, the “vital question” is whether “the government has the power to control the trusts.” He understood that unaccountable power is a threat to the Republic and that the spectacle of untouchable elites fosters widespread resentment.
Humorous typo aside – "county"? I thought we were thinking bigger Tim! – as the name itself highlights, this law is 135 years old. It pre-dated the Wright Brothers flight by just over a decade. Basically no piece of technology we use today had been invented yet, except perhaps the automobile, but only just barely. To say that the world is different today would be the understatement of the century – actually, of two centuries. We can all agree with President Roosevelt's general sentiment – outlined a mere 124 years ago – but that doesn't change the fact that the antitrust theories of not just the 19th century, but even the 20th century, are wildly outdated and unequipped to suit the 21st century. And yes, both the Meta and Google cases highlighted this reality in different ways.
Tuesday’s ruling is clearly wrong on the merits. The government charged that Meta, then called Facebook, broke the law when it bought up its competitors Instagram and WhatsApp in 2012 and 2014. Judge Boasberg threw out the case by concluding that Meta lacks monopoly power now, when the relevant question should have been whether it had monopoly power at the time.
No, it was not wrong on the merits. The argument against Meta was poor and misguided – to the point where it was initially thrown out before getting reworked with the bare minimum tweaks to get it to trial. One major issue with these antitrust cases is that they're being litigated in hindsight – which is both obviously but obviously an issue with tech in particular, as it moves so fast to as make most central arguments moot by the time of trial. So Wu's argument here is that they should punish Meta (then Facebook) for past rule-breaking even though the government obviously looked into and approved their deals at the time. Why? In part because both deals weren't taken all that seriously back then by many. The Instagram deal, in particular, was mocked. Not by all of us, mind you! But my blog post at the time clearly failed to convince anyone that perhaps in the future, these would grow into massive companies under Facebook.
Essentially, Wu would prefer the judge makes the ruling based on the success of the deals that most at the time didn't think would be successful – and probably no one did to the extent that they have! And he doesn't want the judge ruling on the case for how things stand now. So he would like the judge to rule on the case based on some random moment in time where Facebook held monopoly power over social networking. Okay.
But even to deny that Meta now holds a monopoly in personal social networking (sharing with friends and family) means ignoring a lot of direct evidence to the contrary. The government presented records of the company’s extraordinary and durable profits ($87.1 billion operating profit on $164.5 billion in revenue in 2024, for example), which is a textbook signal of monopoly power. The company has also subjected users to more and more ads, removed privacy protections and otherwise reduced the quality of its service without losing its user base, which is hard for a company facing competition to do.
Then perhaps the government should make the case about Meta's ads business – as they're also currently doing with Google – not with the "personal social networking" silly moment-in-time made up market definition. But again, that's not the case that the government made – and undoubtedly in part because of Google's position in said ad market.
Those failings aside, Judge Boasberg’s ruling is also flawed in a broader sense. Does anyone seriously doubt that Meta is the kind of company that antitrust laws were designed to restrain? In the 1980s and 1990s, judges held AT&T and Microsoft to account for antitrust violations, despite their protests. In the 1940s, when the federal judge Learned Hand pondered a similar case against the Alcoa company, which dominated the aluminum market in the United States, he granted that there were “legal niceties” in which one might get “entangled.” But he concluded that “if we hold that it is not a monopoly,” everyone not caught up in those niceties would “quite rightly I think, write us down as asses.”
Why yes, I do have some doubts that laws written 135 years ago would foresee the rise of a social feed ad network. The rest of Meta does not make any real money – famously so in the case of VR and now AI – so it's all just that one business. Also, the Microsoft "victory" was more similar to the Google one in that nothing really changed as a result. People like Wu love to argue that it changed everything, leaving out what happened after the victory that neutered the remedies. What disrupted Microsoft was natural market forces. At best, you can say they were slightly distracted by the case, but they would have been disrupted regardless.
Also, Microsoft is now a $4T company. What a great victory for antitrust!
I don't know enough about the 1940s aluminum market to have a strong opinion there. But it strikes me that the market and time were probably different from, say, mobile applications in 2025. At least the 1940s were closer to the 1890s than we are now...
Judge Boasberg’s ruling in the Meta case follows a ruling last year by the federal judge Amit Mehta, who found that Google is an illegal monopoly — yet this year could not bring himself to impose on Google a strong remedy. Such decisions effectively bless one of the major power imbalances of our age. They signal to companies that spending millions of dollars on lawyers will be repaid with billions of dollars in profit. And they reinforce the tech industry’s sense of impunity — its belief that it is an untouchable sovereign above the control of any mere human.
Project much, my guy? Again, it is not the judge's job to make a ruling that Tim Wu believes will better balance power inequality in the US. The job is to look at the case in front of you and rule on that. And Judge Mehta did rule that Google was a monopoly, but he also rightfully concluded that the remedies put forward wouldn't actually change where Google had the monopoly. And that the market was already doing its own thing to change that market dynamic.
I do think there was a better case to be made about not being able to leverage the Search monopoly going forward to unnaturally prop-up Google's other services – which is something Microsoft was also regularly found in violation of for years after their antitrust issues – like, say, AI.
Despite a decade of Meta-related scandals and whistle-blowers, no public authority in this country has been able to hold the company meaningfully accountable for its behavior — a striking failure of our political system. Unfortunately, this ineffectuality is part of a broader trend. Many policies enacted in the 20th century to maintain economic fairness in the United States have been undermined or gutted: labor law, financial regulation, telecommunications law and taxation, among others. For decades those laws forestalled a collapse into a two-class society, avoiding the extreme class tensions that have upended other republics.
This is obviously oversimplified, but I'm not even sure I really disagree with this at the highest level. But the issue here is that this is unrelated to this antitrust case, which was a weak case around a very specific set of circumstances. It was not the judges job to rule if Meta should be punished for, say, privacy scandals here.
When Roosevelt embarked on his antitrust campaign in 1902, he did so with the view that it was necessary to preserve democracy. His was a moment like ours, with a dangerously unbalanced economy, extremist ideologies vying for adherents and a near-revolt of American farmers in the Midwest. But the federal and state governments diffused that anger by taming the railroads, regulating the banks and breaking up most of the big trusts.
This time around we are on track to do nothing of the kind. Congress has done little to restore economic balance and courts are showing themselves to be out of touch with economic realities that any American could describe. It is no wonder the population is growing increasingly cynical — and more and more ready to look to ideologies that promise to upend the entire system.
Beyond the overtly political statement, here we go again with references that are over a century old. Even if you agree that there are some parallels with the United States as it was 123 years ago, obviously, the same laws and policies aren't going to make sense in 2025. Instead, what Wu should argue for here is a new Roosevelt, someone to rally the country to come up with and pass new laws to handle such issues. Laws written in the 21st century, to tackle challenges in the 21st century.
But that's a lot harder than bringing forward weak cases and making outdated arguments around old laws. And it's certainly harder than writing endless and pointless op-eds.
M.G. Siegler
M.G. Siegler
M.G. Siegler
M.G. Siegler
M.G. Siegler
2025-11-24 06:56:56
Let's talk about Pluribus. Admittedly, I was a bit late in starting to watch it, but after hearing non-stop buzz about how good it is, I gave in...
The first episode is good, but also a bit misleading for the rest of the show thus far. It's a full-on zombie apocalypse situation. Yes, people are raised from the dead Night King-style with smiles upon their faces, but still, that's what it is. And this leads you down the path of thinking this is setting up to be a social commentary on the "hive mind". That is also nothing new with it comes to science fiction, of course. But it does feel more timely than ever in our current culture.
But then it keeps going. And now I'm here wondering if this isn't in fact meant to be a commentary about AI? To be clear and fair, that's probably projecting a bit much just given what I tend to write about. But it also works! Almost perfectly. Large Language Models in particular. Watching Pluribus feels like a statement about using ChatGPT, or Gemini, or any of the AI "chatbots"!
Granted, the allegory may break down if you simply look at the timeline for when Vince Gilligan (Breaking Bad, Better Call Saul, etc) was writing this show. ChatGPT, of course, launched in November 2022 – happy 3 year birthday next week! – and that very tool tells me that Gilligan was writing the show in 2023 and into 2024. Now it's certainly possible that he was ChatGPT-pilled by then, but it also feels fairly unlikely as he's not a person who, say, lives and works in Silicon Valley. Still, it's possible!
Regardless, it doesn't really matter. Because at least through these first four episodes, Pluribus works as a pretty great analogy for LLM-based AI. To wit: All of humanity's knowledge is involuntarily uploaded to the collective "cloud", as it were. With this, any of the beings on Earth – aside from a handful that the virus couldn't infect for unknown reasons1 – have access to all of this information in real time. And any of the "unenlightened", when they ask any question of these connected beings, they get back a seemingly factually true, but often bland response in real time.
The consensus view, quite literally. Have I mentioned the responses are highly sycophantic? Always said with a smile on the face.
I'm watching this and as the story unfolds, I'm thinking at times that it's almost too much on the nose for our current situation. Again, I'm not sure Gilligan meant this as an AI allegory versus simply a "hive mind" one about broader culture – and I'm specifically not reading about such things so as not to spoil anything for me, or alter my own thoughts on the matter – but it feels almost too perfect.
Beyond just the high-level stuff, it actually goes deeper with power/resource conservation and, in the most recent episode, even prompt injecting! By literal injection!
How's that for on the nose?!
Again, maybe I'm reading too much into this. But I'm not sure that I am. And I'm going to keep going with it until proven otherwise in future episodes.
Overall, I like the show thus far, but don't love it in the way that others seem to. It's fun and interesting and at times good, but also slow. The same could be said of Gilligan's other shows at times, I suppose. But this one has the most clear Lost vibes. But you know, hand grenade aside, there's not a lot of action.
I'm assuming that changes as the season crosses the halfway point. Maybe we don't get smoke monsters, but instead some agentic beings meant to convert Carol at all costs – perhaps quite literally. Something to watch, quite literally.