Yes and to make matters worse some competitors (DuckDuckGo for one) have no-AI versions of their search engines on offer. I moved away from Google aeons ago and off AI as soon as possible. It is a necessity for me because I need to be able to cite actual sources for the data I use.
Well said, Stephen. I left Google long, long ago ~ and I'm currently using DDG's no-AI version now.
Reminds me of the very best comment I've ever heard about using AI for "search":
"The difference between AI and pre-AI. . . . Google gave you 4 million web sites that were the answer to . . . [your] question. . . . Today, ChatGPT will give one very articulate answer . . . and it will position that answer as the truth. If you are not the most intelligent man alive, or woman alive. , you'll believe it. . . . Very few of us go back to ChatGPT and say, "Where the f**k did you get that from?"
$132 billion isn't enough money to run a company on. At this point, Google is going to have to go to the US government and ask for a multi-trillion dollar bailout to stay afloat.
Because Google is one $5 mistake away from declaring bankruptcy, they can't even afford H1-B labor at this point.
I think it's time to short GOOG all the way down to $0.
Shorting a company that just pulled $120B in a single quarter, sits on $240B+ in cash, expanded operating margins to 34%, and holds a $500B+ Cloud backlog because they borrowed cheap long-term money to build AI centers is... certainly a choice.
You live under the delusion that a paltry $132 billion is enough to run a company on; did you not read the line in this post that said that it's worrisome that el Goog has went to the bond market looking for cash?
Calling $162+ billion in liquid cash 'paltry' while completely misunderstanding why AA+ rated mega-caps issue debt is wild.
Google doesn't go to the bond market because it's broke; it goes because institutions line up with $100B+ in orders to lend them money at low fixed rates. S&P literally gives them an adjusted net leverage of 0.0x.
If you want to short a company generating $40B in operational profit a quarter because they know how to utilize corporate debt markets, go for it lol the market will appreciate the liquidity.
Lol free cash flow and cash reserves are two completely different financial metrics tho...
FCF dipped negative in Q2 because they spent $45b in a single quarter on data centers and AI hardware. Operating Cash Flow on the other hand was still positive at $39b+.
OCF is just the incoming side, so to say, so with a negative cash flow, they spent more cash on their business than they earned. Does not sound great to me, to say the least.
I actually look for inconvenient truths because it usually indicates that there's an opportunity... your Google to 0 thesis however is void of any substance and is frankly laughable.
Let’s not forget that many bonds in 2007-2008 were rated AAA but were ultimately failed during that financial crisis. Ergo, your bond rating argument is not persuasive.
If Google can’t deliver orders on $40 billion in profit with $100 billion in outstanding orders as you claim, it’s got a much bigger problem as a company and is worthy of shorting.
We’ll review your comment in a year and see if Google is still in business OR if it has run to the government for a bail out. I can’t wait to read your public apology for being so wrong.
Comparing subprime mortgage-backed structured CDOs from 2008 to simple corporate bonds backed by $40B in quarterly operational earnings shows a fundamental misunderstanding of corporate finance.
Google isn't a bank with hidden toxic subprime assets. It's an enterprise tech giant with $120B in quarterly revenue, $160B in cash, and 34% profit margins.
Are you doing some kind of troll engagement strategy you spout ridiculous comments and when people reply, you restack them with your comment?
>Google isn't a bank with hidden toxic subprime assets
Of course they are: they think there insulated from all the dept of their SPVs, but I wouldn't be suprised if they get a haircut when that AI idiocy blows up.
Patrick Boyle has a good episode why these SPVs all the Mag7 are using aren't illegal, and that they are listed in their reports: As a foot note on page 72 of the 83 pages report. True "Yes, Minister"-Sir Humphrey Appleby-style :-)
I think you hit the right issue in terms of a whether a race to dominate the LLM market is even the end goal. That makes sense if people think of AI as competition for search and other administrative and organizational tasks that can be automated (at least partially). The bigger question is how society will develop and use the formative elements of AI (data base analysis, algorithms, software, communications, etc.) in a wider range of research and other learning activities. This is much bigger than Google. It like thinking about the beginning of electric power and asking what people would do with it. Most of the gains will come from the end-user applications.
Roger that. I think it is perhaps better to consider Alphabet. YouTube is a monster. I have personally found Google Gemini to be much more reliable than OpenAI, although I use both. I think that they have effectively co-opted AI in addition to their search business, and Gemini is informed by literally decades of search, so it has a much better sense of what is reliable and what is not. It is much superior to OpenAI as far as I am concerned for fact-checking, although OpenAI is getting better as time goes by. I think that they will be able to monetize the AI search as well as the traditional SERP pages. They are sort of like Elon Musk. They do not just sit there and cling to an outmoded business model. They are always looking for the new new thing. As you say, their revenue is the kind of problem I think almost any company would like to have.
Not really. They all have their attributes and weaknesses
Gemini is like a young engineer that thinks it knows everything and gets out over it's skis often
Grok is like a super research assistant. Let me at'em. Let me do the searches boss and show you what I know. Let me show you all I know about Math and Physics. But then it stops short of brilliant ideas.
Claude is an old curmudgeonly engineer that's seen pretty much everything once or twice. It is super conservative, want's to base every thing on facts that it can verify, but doesn't have the research skills that Grok has. To Gemini it says "hold my beer, son".
That's s very nice uninformed emotional non-technical response. I can hold two things in my mind concurrently: Elmo is an asshole and the people who work at his companies are good engineers and create good producers. Read F. Scott Fitzgerald
The real moat is distribution, not the model. Google ships the product to a billion people who never asked for a model; that reach is the scarce asset, and it doesn't erode just because the frontier keeps moving. The technology races ahead, but ownership of the default is what pays the abundance tax.
And who knows, Google's huge, maybe they'll start a neurosymbolic program because, why not? they can. In which case they stand to be way ahead of the others in a decade or so.
I agree Marcus. And if you ask me if I wanted to broadly deploy Gemini vs ms copilot in an enterprise, I would choose Gemini. It's great for slides and images, copilot not. Notebook lm a hidden gem. Now which model to call for deep analytics, varies.
If this shake-up is real, the interesting part isn’t one leader stepping aside, it’s whether DeepMind can keep its research culture intact while the talent market gets this aggressive. In AI, the org chart matters less than where the best researchers decide they can do their best work.
Anthropic's conflict with the administration is arguably an asset if the tides change and the pendulum swings again; they may believe they're feeling the pain now but if they're smart it could also be used as positioning. The administration is also really its own worst enemy and I wouldn't place great bets on it being a stable entity.
Counting them out of what? reaching AGI/ASI or having a profitable LLM chatbot?
Realistically isn’t Deepmind still the only player (out of the major ones you listed) that has a shot to AGI? as they are the only people who aren’t putting all the eggs in transformers and scaling laws, the only one that is doing science, and not all product marketing?
Anthropic has the same, arguably worse, declining reputation by continuing to pretend to be the good guys while stealing and destroying books and dreaming about achieving ASI via RSI (Claude Code) in 2-5 years. Claude Code is neat, not RSI neat though. Most of their “breakthroughs” so far seem to be direct applications of academic researches on their proprietary models, not anything developed in house.
"There’s also a huge group of AI academics who don’t work with LLMs at all. Many of them are scientists who build specialized AI models that can analyze data, make useful predictions, or even simulate entire physical systems. Those researchers aren’t necessarily competing with the frontier labs—Google DeepMind’s AlphaFold team, which built a Nobel Prize–winning model that predicts the structures of proteins, was disbanded last month. But they face plenty of their own challenges. At the convening, several voiced concerns about how the widespread ignorance of non-LLM AI was affecting their work. Researchers who build specialized AI tools to help address climate change, for example, sometimes struggle to advocate for their work when so many people believe that “AI” means “energy-guzzling LLMs.”
Do people really understand that LLMs and OpenAI/Anthropic are not all inclusive of AI? Clearly the researchers know this but perhaps the general public does not understand this.
The recent announcement on Science Friday/NPR of the work being done at Cal Tech on enzymes by the 2024 Nobel winner in biochemistry and her AI team is one of biggest stories of the year.
I’m not sure what to make of it. Gemini was leading major LLM benchmarks as recently as a year ago, yet Google has a well-earned reputation for shelving projects that fail to remain strategically central or clearly ahead of the field
Google's negative cash flow and the fact that they included the entirely theoretical value of OpenAI as "earnings" (net income) to inflate their numbers, would emphasize your last sentence also from a business perspective. See Ed Elson at Prof G Markets for a detailed analysis.
The line worth pushing on is reason #3 — $402B revenue and $132B profit is trailing performance, and you flagged the real tell yourself: Google going back to the bond market despite that cash pile suggests the TPU buildout and Gemini training runs are burning faster than the balance sheet lets on. Meta and OpenAI are both funding compute with even more leverage, so this isn't a "who has more cash" race anymore, it's a "who can service the debt if your predicted multi-way tie actually happens" race. Koray inheriting DeepMind without Hassabis's name recognition looks like the bigger near-term risk than any compute or talent gap.
Google is already losing search by converting it to AI.
Why isn't this already another serious red flag?
Yes and to make matters worse some competitors (DuckDuckGo for one) have no-AI versions of their search engines on offer. I moved away from Google aeons ago and off AI as soon as possible. It is a necessity for me because I need to be able to cite actual sources for the data I use.
Well said, Stephen. I left Google long, long ago ~ and I'm currently using DDG's no-AI version now.
Reminds me of the very best comment I've ever heard about using AI for "search":
"The difference between AI and pre-AI. . . . Google gave you 4 million web sites that were the answer to . . . [your] question. . . . Today, ChatGPT will give one very articulate answer . . . and it will position that answer as the truth. If you are not the most intelligent man alive, or woman alive. , you'll believe it. . . . Very few of us go back to ChatGPT and say, "Where the f**k did you get that from?"
~ Mo Gawdat | https://bra.in/3joWP5
“I’m afraid I can’t help you with that.” - TruthministerGPT
🎯💯~ Exactly!
$132 billion isn't enough money to run a company on. At this point, Google is going to have to go to the US government and ask for a multi-trillion dollar bailout to stay afloat.
Because Google is one $5 mistake away from declaring bankruptcy, they can't even afford H1-B labor at this point.
I think it's time to short GOOG all the way down to $0.
This is probably the worst take I’ve read on Substack. Seriously impressive work.
Thank you.
I had to work hard to earn this reputation.
Shorting a company that just pulled $120B in a single quarter, sits on $240B+ in cash, expanded operating margins to 34%, and holds a $500B+ Cloud backlog because they borrowed cheap long-term money to build AI centers is... certainly a choice.
You live under the delusion that a paltry $132 billion is enough to run a company on; did you not read the line in this post that said that it's worrisome that el Goog has went to the bond market looking for cash?
I'll pray for you.
Calling $162+ billion in liquid cash 'paltry' while completely misunderstanding why AA+ rated mega-caps issue debt is wild.
Google doesn't go to the bond market because it's broke; it goes because institutions line up with $100B+ in orders to lend them money at low fixed rates. S&P literally gives them an adjusted net leverage of 0.0x.
If you want to short a company generating $40B in operational profit a quarter because they know how to utilize corporate debt markets, go for it lol the market will appreciate the liquidity.
"Calling $162+ billion in liquid cash" What liquid cash ? They are cash flow negative, aren't they ? First time ever.
Lol free cash flow and cash reserves are two completely different financial metrics tho...
FCF dipped negative in Q2 because they spent $45b in a single quarter on data centers and AI hardware. Operating Cash Flow on the other hand was still positive at $39b+.
OCF is just the incoming side, so to say, so with a negative cash flow, they spent more cash on their business than they earned. Does not sound great to me, to say the least.
Still, you can’t run a company on a meager $39 billion
Sssh! These people don’t want to read inconvenient truths.
I actually look for inconvenient truths because it usually indicates that there's an opportunity... your Google to 0 thesis however is void of any substance and is frankly laughable.
Let’s not forget that many bonds in 2007-2008 were rated AAA but were ultimately failed during that financial crisis. Ergo, your bond rating argument is not persuasive.
If Google can’t deliver orders on $40 billion in profit with $100 billion in outstanding orders as you claim, it’s got a much bigger problem as a company and is worthy of shorting.
We’ll review your comment in a year and see if Google is still in business OR if it has run to the government for a bail out. I can’t wait to read your public apology for being so wrong.
Comparing subprime mortgage-backed structured CDOs from 2008 to simple corporate bonds backed by $40B in quarterly operational earnings shows a fundamental misunderstanding of corporate finance.
Google isn't a bank with hidden toxic subprime assets. It's an enterprise tech giant with $120B in quarterly revenue, $160B in cash, and 34% profit margins.
Are you doing some kind of troll engagement strategy you spout ridiculous comments and when people reply, you restack them with your comment?
>Google isn't a bank with hidden toxic subprime assets
Of course they are: they think there insulated from all the dept of their SPVs, but I wouldn't be suprised if they get a haircut when that AI idiocy blows up.
Patrick Boyle has a good episode why these SPVs all the Mag7 are using aren't illegal, and that they are listed in their reports: As a foot note on page 72 of the 83 pages report. True "Yes, Minister"-Sir Humphrey Appleby-style :-)
> Google isn't a bank with hidden toxic subprime assets.
Citation Needed
I especially appreciate the 007 reference, regarding the bonds. Well played!
That was only coincidental.
I think, "I think it's time to short GOOG all the way down to bankruptcy" would be better advice.
That way, you're covered even if it's your own bankruptcy. 😉
Nicely played, Sir.
This is ridiculous.
This commented added a lot to the conversation.
Irrational at best, how can you be taken seriously with comments like that. I would definitely edit with more substance.
the same way that people take you seriously: it's called "the internet".
You forgot the /s
Not sarcasm, ergo no `/s`
I think Google (in Hassabis) also has a better overall strategy, focusing on hybrid AI instead of grinding more data through their transformers.
I think you hit the right issue in terms of a whether a race to dominate the LLM market is even the end goal. That makes sense if people think of AI as competition for search and other administrative and organizational tasks that can be automated (at least partially). The bigger question is how society will develop and use the formative elements of AI (data base analysis, algorithms, software, communications, etc.) in a wider range of research and other learning activities. This is much bigger than Google. It like thinking about the beginning of electric power and asking what people would do with it. Most of the gains will come from the end-user applications.
Roger that. I think it is perhaps better to consider Alphabet. YouTube is a monster. I have personally found Google Gemini to be much more reliable than OpenAI, although I use both. I think that they have effectively co-opted AI in addition to their search business, and Gemini is informed by literally decades of search, so it has a much better sense of what is reliable and what is not. It is much superior to OpenAI as far as I am concerned for fact-checking, although OpenAI is getting better as time goes by. I think that they will be able to monetize the AI search as well as the traditional SERP pages. They are sort of like Elon Musk. They do not just sit there and cling to an outmoded business model. They are always looking for the new new thing. As you say, their revenue is the kind of problem I think almost any company would like to have.
OpenAI aka ChatGPT is the IE 5.5 of AI.
Saying that Gemini is better than ChatGPT is like claiming you've picked up a turd by the clean end: it's just not possible.
If you want an AI that works, use Grok. It cost $8/mo to use the browser on X.com's website. (https://x.com/i/grok)
Not really. They all have their attributes and weaknesses
Gemini is like a young engineer that thinks it knows everything and gets out over it's skis often
Grok is like a super research assistant. Let me at'em. Let me do the searches boss and show you what I know. Let me show you all I know about Math and Physics. But then it stops short of brilliant ideas.
Claude is an old curmudgeonly engineer that's seen pretty much everything once or twice. It is super conservative, want's to base every thing on facts that it can verify, but doesn't have the research skills that Grok has. To Gemini it says "hold my beer, son".
I wouldn't touch grok with a ten-foot pole. MechaHitler wasn't enough for you to discredit that garbage propaganda machine?
That's s very nice uninformed emotional non-technical response. I can hold two things in my mind concurrently: Elmo is an asshole and the people who work at his companies are good engineers and create good producers. Read F. Scott Fitzgerald
Facts don’t care about your quotes.
Thanks for playing.
The real moat is distribution, not the model. Google ships the product to a billion people who never asked for a model; that reach is the scarce asset, and it doesn't erode just because the frontier keeps moving. The technology races ahead, but ownership of the default is what pays the abundance tax.
And who knows, Google's huge, maybe they'll start a neurosymbolic program because, why not? they can. In which case they stand to be way ahead of the others in a decade or so.
I agree Marcus. And if you ask me if I wanted to broadly deploy Gemini vs ms copilot in an enterprise, I would choose Gemini. It's great for slides and images, copilot not. Notebook lm a hidden gem. Now which model to call for deep analytics, varies.
If this shake-up is real, the interesting part isn’t one leader stepping aside, it’s whether DeepMind can keep its research culture intact while the talent market gets this aggressive. In AI, the org chart matters less than where the best researchers decide they can do their best work.
Anthropic's conflict with the administration is arguably an asset if the tides change and the pendulum swings again; they may believe they're feeling the pain now but if they're smart it could also be used as positioning. The administration is also really its own worst enemy and I wouldn't place great bets on it being a stable entity.
Counting them out of what? reaching AGI/ASI or having a profitable LLM chatbot?
Realistically isn’t Deepmind still the only player (out of the major ones you listed) that has a shot to AGI? as they are the only people who aren’t putting all the eggs in transformers and scaling laws, the only one that is doing science, and not all product marketing?
Anthropic has the same, arguably worse, declining reputation by continuing to pretend to be the good guys while stealing and destroying books and dreaming about achieving ASI via RSI (Claude Code) in 2-5 years. Claude Code is neat, not RSI neat though. Most of their “breakthroughs” so far seem to be direct applications of academic researches on their proprietary models, not anything developed in house.
8. Demis is a chess master. Everyone else is playing checkers.
This from MIT Tech Report:
"There’s also a huge group of AI academics who don’t work with LLMs at all. Many of them are scientists who build specialized AI models that can analyze data, make useful predictions, or even simulate entire physical systems. Those researchers aren’t necessarily competing with the frontier labs—Google DeepMind’s AlphaFold team, which built a Nobel Prize–winning model that predicts the structures of proteins, was disbanded last month. But they face plenty of their own challenges. At the convening, several voiced concerns about how the widespread ignorance of non-LLM AI was affecting their work. Researchers who build specialized AI tools to help address climate change, for example, sometimes struggle to advocate for their work when so many people believe that “AI” means “energy-guzzling LLMs.”
Do people really understand that LLMs and OpenAI/Anthropic are not all inclusive of AI? Clearly the researchers know this but perhaps the general public does not understand this.
The recent announcement on Science Friday/NPR of the work being done at Cal Tech on enzymes by the 2024 Nobel winner in biochemistry and her AI team is one of biggest stories of the year.
I’m not sure what to make of it. Gemini was leading major LLM benchmarks as recently as a year ago, yet Google has a well-earned reputation for shelving projects that fail to remain strategically central or clearly ahead of the field
Google's negative cash flow and the fact that they included the entirely theoretical value of OpenAI as "earnings" (net income) to inflate their numbers, would emphasize your last sentence also from a business perspective. See Ed Elson at Prof G Markets for a detailed analysis.
The line worth pushing on is reason #3 — $402B revenue and $132B profit is trailing performance, and you flagged the real tell yourself: Google going back to the bond market despite that cash pile suggests the TPU buildout and Gemini training runs are burning faster than the balance sheet lets on. Meta and OpenAI are both funding compute with even more leverage, so this isn't a "who has more cash" race anymore, it's a "who can service the debt if your predicted multi-way tie actually happens" race. Koray inheriting DeepMind without Hassabis's name recognition looks like the bigger near-term risk than any compute or talent gap.