Google will invest as much as $40 billion in Anthropic

This follows a similar, but smaller, investment by Amazon just days ago.

See full article...

 

[RZetopan]

So much demand that none of them can profit from their AI models. Crazy demand, guys.

Well, the bots and "real soon now" (this time for sure) the 10s of millions of Optimus robots will surely boost the user base. Bots are already shilling for bots. Makes one wonder if both Must and Altman are just bots or robots as well. We already know that Zuckerberg is just a robot, no human could ever act like that.

 

[RZetopan]

AI seems to be an MLM for tech bros.

I think that someone already identified it a fractal grifting, the more you zoom in on it the worse it becomes.

 

[RZetopan]

I’d like to coin “EI” for “enshittified intelligence”. This is the most honest abbreviation for such technology.

Excepting that there is no intelligence in LLM AI, only stupidity. Perhaps ES would be better.

 

[RZetopan]

An investment isn't a gift.

Just rephrase that to "why would they want to own a piece of it?" and you answered your own question.

It's much cheaper to acquire your competition if you get in early. Plus they can be each other's customers . . . which somehow isn't illegal.

SBF tried doing that with FTX and Alameda Research (which, of course, did no research).

 

[djgrowers]

Imagine the investment frenzy when something actually game changing comes along.

 

[grndkntrl]

I’d like to coin “EI” for “enshittified intelligence”. This is the most honest abbreviation for such technology.

For a long while now I've been referring to it as "Atrophied Intelligence", "Actually Idiotic", "Artificial Idiocy", "Advanced Incompetence"…

Take your pick; It's all brain-rot.

 

[sir_pings_alot]

The part I haven’t seen anyone here comment on yet is that we’re firmly in the “dealer giving you the drugs for free” phase of AI. Once you become dependent, that’s when they can jack up the price and have a life long customer.

 

[quamquam quid loquor]

Inference at API prices produce incredible marginal profit. Corporate users print cash. I wrote this analysis many times in many articles.

The drags on costs are free users, subsidized fixed monthly plans, and training costs. It’s unsustainable, something has to give.

 

[quamquam quid loquor]

The part I haven’t seen anyone here comment on yet is that we’re firmly in the “dealer giving you the drugs for free” phase of AI. Once you become dependent, that’s when they can jack up the price and have a life long customer.

The flip side is that open models are incredibly capable. Have you tried Gemma (Google) 31B? It’s not a frontier model, but it’s better than GPT4.

If they jack up the price on frontier models, everyone will just regress back a year.

 

[Geebs]

Meanwhile:

https://www.flyingpenguin.com/alisa-esage-throws-mythos-under-zero-day-bus/

Now, this isn’t my area of expertise so please feel free to jump in with additional context anyone…

…but it seems to me that the stink of circle investment has spiraled to near Madoff levels of grift.

(For those who need the reference:
https://www.britannica.com/biography/Bernie-Madoff )

A “game changing” new model from Anthropic doesn’t actually change the game? Must be a day ending in “y”.

 

[ClusteredIndex]

The part I haven’t seen anyone here comment on yet is that we’re firmly in the “dealer giving you the drugs for free” phase of AI. Once you become dependent, that’s when they can jack up the price and have a life long customer.

This is exactly where we are. And prices are being jacked up as we speak, since in the last few months we’re finally seeing useful and productive products. Expect new layoffs (especially in software engineering - for now) to actually be displacement due to AI. And regular home users will get priced out soon enough for things like Claude Code, GitHub Copilot, Claude Cowork, etc.

 

[StrangeOnion]

Maybe I'm dumb but doesn't Anthropic directly compete with Gemini? Why would Google inject so much money directly into a competing product?

It is indeed weird. I would have expected one of the conditions to be they move a chunk of their workload to Google Cloud but I haven't seen any mention of that.

 

[norton_I]

It is indeed weird. I would have expected one of the conditions to be they move a chunk of their workload to Google Cloud but I haven't seen any mention of that.

Google was already an early investor in Anthropic and Anthropic already uses google TPUs. They have had at least two announcements in the past half year about expanding their use of google TPUs and cloud computing.

 

[norton_I]

The part I haven’t seen anyone here comment on yet is that we’re firmly in the “dealer giving you the drugs for free” phase of AI. Once you become dependent, that’s when they can jack up the price and have a life long customer.

I think there is too much competition for that. Right now, Anthropic and OpenAI are spending money like crazy on training to stay 6 months to a year ahead of low cost providers and open weight models. But as soon as they stop that, everyone is going to catch up and they are going to be unable to command a large premium.

 

[clb2c4e]

If anyone needs more examples of why we don't need super rich and massive corporations, take a look at what they can do with huge amounts of money.

Then consider we could literally provide the entire world with clean drinking water for about the cost of this one deal.

 

[denemo]

So much demand that none of them can profit from their AI models. Crazy demand, guys.

Github Copilot is going to token based billings as I understood it.

https://github.blog/news-insights/company-news/changes-to-github-copilot-individual-plans/

They are even pausing sign-ups.

 

[Geebs]

Github Copilot is going to token based billings as I understood it.

https://github.blog/news-insights/company-news/changes-to-github-copilot-individual-plans/

They are even pausing sign-ups.

They specifically mention “agentic workflows” as an issue which I think is the we’re-contractually-not-allowed-to-ever-imply-anything-even-slightly-negative-about-ai way of saying “would you jerks kindly stop using our services to run openclaw”.

Microsoft could also probably free up some more capacity for coding services by not shoving copilot into every single corner of their software ecosystem regardless of whether or not it makes any sense.

 

[DarthSlack]

I think there is too much competition for that. Right now, Anthropic and OpenAI are spending money like crazy on training to stay 6 months to a year ahead of low cost providers and open weight models. But as soon as they stop that, everyone is going to catch up and they are going to be unable to command a large premium.

Sooo much this. Right now the big AI companies are spending money as fast and furiously as they possibly can to gain market share. When the investor cash runs out (and it will at some point), prices are going sky-high.

Which leaves all the customers either paying through the nose or looking for alternatives. And since very little the big AI companies are doing is unique/proprietary/not reproducible, a bunch of models that don't cost multiple firstborn children to run are going to become popular. DIY is probably going to become a major player.

This is going to be one impressive train wreck.

 

[honkafied]

Exactly, if Claude Code can help Google now, and if their special sauce keeps them in the lead, then better for Google to throw their weight behind the potential victor and benefit over the interim. What's $40 billion to Google anyways?

I think this is the right framing. Google has a $4T market cap.

 

[Glade9266]

They'll make it up on future revenue with a hooked customer base and long-term demand, paired with decreasing cost-per-token for existing models. A "frontier model" from 24 months ago has seen its cost/token drop 10x or more in the intervening two years, and there is still heavy demand for those models for certain workloads.

This is the Uber model at play (as one example out of many) and it has worked in the past. Subsidize a service, capture a large customer base through rapid expansion (also subsidized), raise prices to become profitable once the market stabilizes.

Not saying this is the best business model, and certainly not saying it benefits consumers, but it has been shown to work before and that's the bet being made here.

To my knowledge, that 10x price drop is for the API. I haven't seen any major AI company actually release their cost of inference per token.

There have also been a bunch of times where these companies adjust API pricing when compeditors come out with a new product to retain rough price parity. I personally don't think that they accurately reflect the cost of inference.

Also, other people have already mentioned this, but why pay OpenAI or Anthropic for their old models, when you can run equivalent open weight models on a bunch of different cloud providers, none of which need to include the model training costs in their pricing? Or, if I'm the right size of business, why wouldn't I just run the open weight model myself?

I don't think the Uber model works here, because Uber's costs stayed roughly the same regardless of how many new users they had. The drivers pay all the up front and operating costs, Uber mostly spends money on ads.

Even if the unprofitable growth to profit model is what these companies are trying to do, they are losing so much money on every new user, and betting on future AI demand in the form of data center payments. The cost to get to that turning point to flip the profit lever, even based on OpenAI's own projections, is hundreds of billions more dollars. Where are they still going to get all that money?

 

[postadelmaga]

finally some one give them some money so hopefully their product can run better for everyone

 

[Tactical Finesse]

Anthropic deliver values with their products, basically every developer may want to use their product at 100$ a month because it has more than that value.
Google has one of the best TPUs out there, they use Claude Code internally. Google owns about 14% of Anthopic (now valued 350+B $ ) so they are paying 40B for about 40B value + Anthropic gonna pay for their TPU.
So either you don't understand these fundamental and easy fact either you are just a parrot and unfortunately all those like confirm most of people cannot do 2+2 and they have some hidden retard

Before you get snide...you might want to reflect on the fact that Anthropic is having to shrinkflation its service--because people are using Claude Code, and Anthropic still isn't making money. As far as anyone can know with private companies--Anthropic is revenue negative, just like OpenAI. And it is getting worse not better--because of infra capex.

I do hope you enjoyed the sound of your keyboard, you really showed me. I feel so pwned.

 

[H2O Rip]

Is anyone trying to profit yet? Step 1 of the enshitification cycle is always sell at a loss to maximize market share and attract investment.

Copilot moving to per token cost vs basic (anthropic doing so as well for enterprise) is the start of it.
The players are all cross investing because the blitz scaling will end up collapsing to one or two winners. You need to cross bet or potentially be totally left out of a transformative tech evolution (to what degree is tbd, but its far more "introduction of PC" than "intro of crypto"). Individually they are all acting logically, but yeah it's very bubble inflating and volatile as hell.

As others have said, its acquisition now then assume you can decrease per token cost a ton down the road on new Nvidia or custom hardware to get 10x lower token cost. But the winners will need 100 or 1000x better tokenomics, just wait until this stuff gets properly enshitified, because it will have to for customers. And costs are going to go way up for employers. Simple capitalism logic - tech is going to price this for your company where bean counters say "this is just a bit more cost effective than hiring" ...of course not taking into account long term employee growth. That is quite a bit more expensive than where we are today.

 

[rojcowles]

For a long while now I've been referring to it as "Atrophied Intelligence", "Actually Idiotic", "Artificial Idiocy", "Advanced Incompetence"…

Take your pick; It's all brain-rot.

As someone quipped on one of the socials, LLMs are intelligent the way a pair of dice are intelligent

 

[rojcowles]

So has actual $CashMoney moved from Google to Anthropic as part of this deal or is this another in a long line of "letter of intent for enormous amount of dollars" deals that generate lots of headlines and once the hysteria and Line Go Up activity has faded nothing actually happens?

Tried reading the Bloomberg report but it was paywalled and in the Ars article "will" and "could" on the first line appear to be load bearing.

 

[UhtredsonofUhtred]

For once, a tech valuation that aligns with my own estimation and outlay. As a developer, I've paid this company thousands of dollars in the past few months and am happy with what I've received.

Enjoy the subsidized use while it lasts. Soon, you will either need to stop paying them because the company no longer exists, or start paying them a fee that will cover the actual cost of providing the service + the cost of managing their enormous debt + any profit.

Are you prepared to pay the 10's of thousands of dollars their service will actually cost you? If not, prepare to be sad.

 

[ChrisSD]

So has actual $CashMoney moved from Google to Anthropic as part of this deal or is this another in a long line of "letter of intent for enormous amount of dollars" deals that generate lots of headlines and once the hysteria and Line Go Up activity has faded nothing actually happens?

Tried reading the Bloomberg report but it was paywalled and in the Ars article "will" and "could" on the first line appear to be load bearing.

Bloomberg says:

Anthropic said that Google is committing to invest $10 billion now in cash at a $350 billion valuation, the same amount it was valued at in a funding round in February, not including the recent money raised.

Though it seems that Anthropic is spending a lot on Google's cloud compute so, reading between the lines, this is basically subsidising some of Anthropic's infrastructure costs.

 

[UhtredsonofUhtred]

They'll make it up on future revenue with a hooked customer base and long-term demand, paired with decreasing cost-per-token for existing models. A "frontier model" from 24 months ago has seen its cost/token drop 10x or more in the intervening two years, and there is still heavy demand for those models for certain workloads.

This is the Uber model at play (as one example out of many) and it has worked in the past. Subsidize a service, capture a large customer base through rapid expansion (also subsidized), raise prices to become profitable once the market stabilizes.

Not saying this is the best business model, and certainly not saying it benefits consumers, but it has been shown to work before and that's the bet being made here.

There is no "model" for what is occurring around AI. It's a singular clusterfuck of insanity.

If AI could ever become actually profitable...mashes knuckles on 10-key machine...it will be roughly 200 years in the future given the nearly $2 trillion that has already been thrown at it.

There is no future point of profitability, given the investments/costs trying to get there will never be recouped.

 

[icosapode]

AI seems to be an MLM for tech bros.

An LLMLM if you will?

 

[DarthSlack]

There is no "model" for what is occurring around AI. It's a singular clusterfuck of insanity.

If AI could ever become actually profitable...mashes knuckles on 10-key machine...it will be roughly 200 years in the future given the nearly $2 trillion that has already been thrown at it.

There is no future point of profitability, given the investments/costs trying to get there will never be recouped.

Gonna disagree, I think there are a couple of models/histories worth comparing:

1) The .com boom. Loads of money spent on dubious business models. The goal of startups was to sound all hip and semi-plausible and hope that a buyout and/or IPO happened before anyone figured out the business model was hot buttered garbage. Loads of companies went under, but e-commerce survived.

2) The Indian outsourcing boom. Companies threw megatons of good employees into the nearest trash can in a rush to hire remote Indian workers at a fraction of the wage. Turns out having an entirely remote workforce, particularly one in a significantly different time zone leads to all kinds of productivity problems. Companies beat a hasty retreat and while outsourcing does still happen, it's not management flavor of the month anymore.

 

[Mrbonk]

Just imagine how many employees they could have paid with that amount of money if you counted all the people they have laid off since Covid.

 

[VoterFrog]

Copilot moving to per token cost vs basic (anthropic doing so as well for enterprise) is the start of it.
The players are all cross investing because the blitz scaling will end up collapsing to one or two winners. You need to cross bet or potentially be totally left out of a transformative tech evolution (to what degree is tbd, but its far more "introduction of PC" than "intro of crypto"). Individually they are all acting logically, but yeah it's very bubble inflating and volatile as hell.

As others have said, its acquisition now then assume you can decrease per token cost a ton down the road on new Nvidia or custom hardware to get 10x lower token cost. But the winners will need 100 or 1000x better tokenomics, just wait until this stuff gets properly enshitified, because it will have to for customers. And costs are going to go way up for employers. Simple capitalism logic - tech is going to price this for your company where bean counters say "this is just a bit more cost effective than hiring" ...of course not taking into account long term employee growth. That is quite a bit more expensive than where we are today.

100x is really not as crazy as you think. The other Google news from a few days ago was that they had 2x hardware efficiency improvement over their last generation of TPUs. 6x more compute per power by including data center improvements and that's not even including model efficiency improvements.

It only takes two generations of 10x to reach 100x. Or 3 generations of 5x. Simple Moore's law scaling gets you there in about a decade, which is not that difficult to pull off for a large company running on debt and investment.

 

[no_great_name]

As someone who was deeply skeptical about LLM tooling for a long time, Claude has been a game changer at work. I don’t know exactly how to feel about that, but it’s inarguably true.

And the difference in capability between the smaller models and the larger premium ones like Opus are immediately noticeable. I could tell you just by watching it “think” if the agent has accidentally been set to one of the cheaper models.

I have seen the enemy and it is me

 

[ClickOnThisToBeHuman]

See this is why I love technology. Google loves space as much as I do. I have always wondered what the other side of a black hole would look like.

 

[LibraryCommoner]

AI seems to be an MLM for tech bros.

and before that it was (is) crypto!

 

[reyan]

You're not wrong about most of that, but characterising it as "a technology nobody wants" is hyperbole at best. There is clearly value in many aspects of "AI". It's like saying nobody wanted the web during the .com bubble.

I think it’s from living in an anti-AI bubble. anyone who says, basically “this thing hundreds of millions of people use daily… ya, no one wants it” isn’t really being a serious individual.

I’m not a fan of what AI is doing to society and the environment. But that statement also immediately jumped out to me as absurd.

 

[reyan]

For once, a tech valuation that aligns with my own estimation and outlay. As a developer, I've paid this company thousands of dollars in the past few months and am happy with what I've received.

You assign tech valuations based on your personal use of technology, without considering things like cost or profit margins of the service being provided to you?

 

[AI_Skeptic]

A "frontier model" from 24 months ago has seen its cost/token drop 10x or more in the intervening two years, and there is still heavy demand for those models for certain workloads.

That is news to me. Source please.

 

[AI_Skeptic]

I think it’s from living in an anti-AI bubble. anyone who says, basically “this thing hundreds of millions of people use daily… ya, no one wants it” isn’t really being a serious individual.

People like free. ChatGPT, Et.al, are currently "free" to the end user. Would people still use it if it's $1 a month?