In the current climate limiting someone's use of AI might be expected to be about restricting access or restricting what someone can do with it, but the story here ostensibly seems to be about capacity constraints, not any limitation on what models or capabilities Google is giving Meta access to.
These kind of limits happen all the time for big clients.
Cloud services like to present the illusion of an infinite amount of compute available at a fixed price per unit, but the reality is if you try to use too much of any service you'll find you have a quota and requests to increase it will fall on deaf ears if the provider doesn't have more of that resource.
Too much of my working life has been spent shoehorning services into less space/compute/ram/spindles or migrations to other data centers to solve such issues.
If you allow me a bit of pedantry, it's infinite "for all intents and purposes". It doesn't mean you can request civilizational levels of compute, but for a blog, a crud, an ETL and such, that is regular use cases with sensible scale you can absorb any elastic demand.
Having said that, I agree with you. You have to request limit increases often and can't scale even in those instances if you don't plan ahead.
Yeah but you don't need cloud for a blog. Cloud was sold as effectively infinite resources - capacity isn't infinite, or effectively infinite, it's 20% more than you are currently using and you pay 300% more for that.
There has to be a name for this deceptive marketing tactic where you say something is unlimited and then it is only unlimited as long as you don't use very much.
It would be one thing if you occasionally got a "no more capacity" error when requesting large amounts of resources but it doesn't work that way. They confine you to a relatively small amount of resources the entire time you have an account. If you want more you have to request it.
A blog for your product, if your product is already on the cloud, is a very sensible use case for the cloud. Static one deployed to a bucket and a CDN, fast, cache on the edge, high availability.
The tiny blog sure isn't for the cloud, but also it's not the main client of the cloud.
> it's 20% more than you are currently using and you pay 300% more for that.
I'm assuming you are comparing to self hosting. Then you need to account for things that are difficult to put a price like your time maintaining a physical infrastructure and the lessons you will learn with it.
Sounds like I'm defending the big cloud, but there is a valid use that is disconsidered because it's trendy to hate on the cloud.
> They confine you to a relatively small amount of resources the entire time you have an account. If you want more you have to request it.
I compare cloud to non-cloud VPSes. If you compare them to self-hosting the price is even more biased against cloud, even with current RAM prices. Did you know you can get 40G or 100G dedicated internet to your colo rack for something like $2000 a month (prices vary greatly, YMMV)? Colo only makes sense if you need a fairly large quantity of compute resources, but the per-unit cost can be very good. Every other style of hosting is building on top of it with a profit margin, after all.
You can order bare metal servers delivery time in minutes from any number of hosting providers and the cost difference is so huge you can afford to keep excess capacity and still come out ahead.
I run the CI infra for our company, and our bare metal costs (sans my salary baked in), are one order of magnitude less than if using any other CI saas provider like github or others.
Like literally 10x times more expensive to do so, to run CI jobs...
I dont want to imagine the margin AWS has like generally, cause it can easily be a 90% too
Right? It's actually crazy how much they don't cost. Are you using it more than 10%? If so, you're saving money.
I assume you're using your owned server and not a provider like Hetzner? So you did have a substantial delivery time. Although in my city is a recycled that resells used servers, and I could show up there with a truck and get a server within hours if I'm not too picky. Or use some random desktop or laptop off the pile, short-term.
No we're on hetzner/ovh boxes, so delivery time really isnt an issue.
Right now the biggest issue is the vibe coded CI program is not really meant to be a distributed multi-node thing yet, so we're on the biggest machines (there's some newer bigger stuff we could migrate too) and the only issue is on peak hours queue can get a bit slow.. but that was also some other bugs etc making not ideal.
Tbh it works pretty well, we just need now to scale it to more than one node etc (which is not to say that is easy, but still, x10 headroom to work with)
Google makes claims here about high demand for Gemini - does anyone here have insight into how much of the load on Google is paid use vs the load from putting AI summaries into every web search?
Don't know, but Gemini 3.1 flash lite is available for free under relatively generous limits, and it had lots of random interruptions like when I was testing it. (Intermittently responding with errors due to high load.)
My curiosity is not the free AI summaries (which they can opaquely tune as necessary), but instead the renting of TPUs to Anthropic and OpenAI. Many of these contracts were announced last minute and seemed to involve a very desperate Anthropic. Based on the Anthropic/xAI data center contract, they’re willing to pay crazy markup to get immediate access to compute.
I want to know how impacted Gemini has been by that, because that will reveal a lot about their margins and revenue generating first party demand. Each MSFT earnings report they discuss the balance they’re dealing with between supplying GPUs to Azure customers and first party demand.
My pet theory is that Gemini is “losing” the LLM race because they’re preferentially selling the TPUs to competitors, while keeping just enough for themselves to stay competitive and build their own products.
Yeah I had a trial for AI Pro or whatever it's called and could never use Gemini CLI (when it still existed) because it was constantly "overloaded". Using the API directly (wihtout a subscription) sometimes works but the models are so buggy and the endpoints constantly spew errors that it's not usable. See this forum thread for example: https://discuss.ai.google.dev/t/frequent-503-errors-service-... it started with 503 errors since JANUARY and it's still not fixed. These are "stable" GA models!
I HIGHLY doubt that Gemini is overloaded, Google has been bullshitting with their crap models since release. Waste of everyone's time.
It's worse than OpenAI or Anthropic. However their lower tier consumer offerings can sometimes be had for <$10/mo on offer and come bundled with other Google services like cloud storage.
I do believe this will be the norm from now on to get access to top frontier model. Computing capacity plus state restrictions plus KYC will be imposed to organisations to get access, individuals will be served last on the queue with degraded performance. Once the Chinese models catch up, nobody (at least individuals) will turn back again to frontier labs.
This seems less about frontier models and restriction and more just lack of compute capacity to meet demand. This has always been an issue for large clients running on cloud, though not to this extent.
It's interesting that Meta is heavily using Google's models (as opposed to Anthropic or OpenAI) given that they are not SOTA for coding. I wonder if this for some strategic/competitive reason, or maybe for cost saving?
I would imagine there are many situations within Meta's applications where relatively small models can do a good job — sentiment analysis, abusive language detection, characterising users based on their posts, summarising a user's complaint so it can be ignored more efficiently, assessing whether ads are likely to be fraudulent so they can be run more often, etc.
Hmm ... I was assuming they were using these models for development, but I wonder if any of it might be for production instead - perhaps using vision models to analyze posted content? That would certainly be massive scale, but I'd have thought that scale would require them to be running in their own datacenters.
OTOH, if they are stressing Google's capacity then it seems it has to be for production use, which would relfect a massive failure on Meta's side given their investment in datacenters and AI. If they can't utilize their own models and datacenters, then maybe they should just rent the excess capacity to Google! :)
Image/video understanding still quite cost effective from the Gemini flash series models?
Image generation and veo models I’d imagine quite effective for creators; new Instagram accounts with AI content that are garnering millions of followers in spans of weeks are quite common now
Facebook is ethically challenged and that's putting it very very very mildly. Yes, they have unlimited money, but at a certain point, it comes across like a rich dude at a bar telling a beautiful woman that he'll buy her a diamond bracelet if she will just come over to his place right now. They make my skin crawl.
Must be to classify/moderate images for social media. They're pretty good at that. I can't imagine what else you'd want to use Gemini models for, certainly not coding.
I've criticized Antigravity in this same conversation, but Google Gemini is good at coding. Even Flash 3.5 low is good at coding. The problem is that Google isn't hungry anymore and it really really really shows in how much they've botched everything to do with Antigravity.
Misleading title on HN but an interesting article, a reminder of why the hyper scalers are investing heavily in infrastructure.
That said, I expect much of the AI bubble to pop. Google Gemini with Antigravity is a good product, as is a Claude Code subscription but I have switched to using DeepSeek v4 Pro with the Claude Code harness and DeepSeek v4 Flash with the OpenCode harness (when I am not using local models with little-coder/pi) and at least for the foreseeable future I don’t think I am going back. Fast APIs at low cost trumps having to spend a little more time to get the same quality of results.
I am a huge fan of Google Gemini, but Antigravity is not a good product. Just recently I've had issues with:
* Repeated instances of incorrect code insertion that the agent cannot clean up. Sure, version control, but this is often happening in new files that aren't even in version control yet.
* Lost chat history when I close and restart the app.
* Not being able to restore a chat from the history (just saw this last week).
* Overly broad searches that waste time and tokens.
* No vertical scroll bar arrows. WTF?? Doesn't the interface look "flat" enough already? This feels arbitrary and stupid.
* The previous chat prompt takes up a large portion of the vertical space of the chat window, even on a high res display.
When it works Antigravity is excellent. When it doesn't work, it's absolutely horrible. If you check the update history, there are usually just a few items and they're super generic things like "Fixed a bug with text entry.".
I don't see it improving at any kind of reasonable pace, even over the last 6 months As a result, I've mostly relegated Antigravity to a planning tool and it does an excellent job. Or I use it to write prompts that I give to Codex. It definitely can do an excellent job writing code sometimes, but sometimes it also does an absolutely horrible job with not breaking the code when it inserts it. It seems to be terrible at understanding C++ braces. How often? Way too often. I always know it's happening because it prompts me to run Git while it's doing something. LOL, that's how I know that it's broken something.
Codex is definitely way, way, way better. It's not even a contest at this point. Codex never breaks my code. It might not always do what I want, but it's just an order of magnitude better than Antigravity. Antigravity really feels like a comedy of errors at this point. ESPECIALLY from a company with Google's resources.
Using LLMs for development is not efficient. All of the problems these companies are having trying to provide enough compute and energy are proof.
Understanding the actual problems we are trying to solve with code and efficiently coming up with solutions (essentially, pre-LLM development) will always be better than wastefully brute forcing solutions with LLMs.
There can be a knitflex for Oculus Quest 3 as Valkyrie Emergate for 1080,pre order Meta yarn (requires Facebook login.). The cymflex renders at 720p in horizon.
This seems to be a bit of a misleading headline.
In the current climate limiting someone's use of AI might be expected to be about restricting access or restricting what someone can do with it, but the story here ostensibly seems to be about capacity constraints, not any limitation on what models or capabilities Google is giving Meta access to.
These kind of limits happen all the time for big clients.
Cloud services like to present the illusion of an infinite amount of compute available at a fixed price per unit, but the reality is if you try to use too much of any service you'll find you have a quota and requests to increase it will fall on deaf ears if the provider doesn't have more of that resource.
Too much of my working life has been spent shoehorning services into less space/compute/ram/spindles or migrations to other data centers to solve such issues.
If you allow me a bit of pedantry, it's infinite "for all intents and purposes". It doesn't mean you can request civilizational levels of compute, but for a blog, a crud, an ETL and such, that is regular use cases with sensible scale you can absorb any elastic demand.
Having said that, I agree with you. You have to request limit increases often and can't scale even in those instances if you don't plan ahead.
Yeah but you don't need cloud for a blog. Cloud was sold as effectively infinite resources - capacity isn't infinite, or effectively infinite, it's 20% more than you are currently using and you pay 300% more for that.
There has to be a name for this deceptive marketing tactic where you say something is unlimited and then it is only unlimited as long as you don't use very much.
It would be one thing if you occasionally got a "no more capacity" error when requesting large amounts of resources but it doesn't work that way. They confine you to a relatively small amount of resources the entire time you have an account. If you want more you have to request it.
It was sold as flexible, near instant provisioning of DC level resources. I don't recall having seen infinite anywhere.
It's not flexible if it only flexes 20% above your current usage
For 99+% of users it will flex 10000% above their current usage
Then 99+% of users could save three quarters of their costs by switching to a traditional VPS provider.
A blog for your product, if your product is already on the cloud, is a very sensible use case for the cloud. Static one deployed to a bucket and a CDN, fast, cache on the edge, high availability.
The tiny blog sure isn't for the cloud, but also it's not the main client of the cloud.
> it's 20% more than you are currently using and you pay 300% more for that.
I'm assuming you are comparing to self hosting. Then you need to account for things that are difficult to put a price like your time maintaining a physical infrastructure and the lessons you will learn with it.
Sounds like I'm defending the big cloud, but there is a valid use that is disconsidered because it's trendy to hate on the cloud.
> They confine you to a relatively small amount of resources the entire time you have an account. If you want more you have to request it.
It's a form of KYC, nothing wrong with that.
I compare cloud to non-cloud VPSes. If you compare them to self-hosting the price is even more biased against cloud, even with current RAM prices. Did you know you can get 40G or 100G dedicated internet to your colo rack for something like $2000 a month (prices vary greatly, YMMV)? Colo only makes sense if you need a fairly large quantity of compute resources, but the per-unit cost can be very good. Every other style of hosting is building on top of it with a profit margin, after all.
if im going to have to ask for capacity, why dont I just get my own bare metal servers then?
Because you don't have to wait for weeks just for delivery. And you pay for elastic usage.
You can order bare metal servers delivery time in minutes from any number of hosting providers and the cost difference is so huge you can afford to keep excess capacity and still come out ahead.
I run the CI infra for our company, and our bare metal costs (sans my salary baked in), are one order of magnitude less than if using any other CI saas provider like github or others.
Like literally 10x times more expensive to do so, to run CI jobs...
I dont want to imagine the margin AWS has like generally, cause it can easily be a 90% too
Right? It's actually crazy how much they don't cost. Are you using it more than 10%? If so, you're saving money.
I assume you're using your owned server and not a provider like Hetzner? So you did have a substantial delivery time. Although in my city is a recycled that resells used servers, and I could show up there with a truck and get a server within hours if I'm not too picky. Or use some random desktop or laptop off the pile, short-term.
No we're on hetzner/ovh boxes, so delivery time really isnt an issue.
Right now the biggest issue is the vibe coded CI program is not really meant to be a distributed multi-node thing yet, so we're on the biggest machines (there's some newer bigger stuff we could migrate too) and the only issue is on peak hours queue can get a bit slow.. but that was also some other bugs etc making not ideal.
Tbh it works pretty well, we just need now to scale it to more than one node etc (which is not to say that is easy, but still, x10 headroom to work with)
Even as a small customer it's easy to hit quotas or hit availablity constraints of more unusual instance types.
definitionally that's "for some intents and purposes" my man
For all intents and purposes is a figure of speech, meaning in every practical sense.
Given Meta’s current AI situation though, I wouldn’t be surprised if they were trying to do distillation and the capacity story is a cover
Google makes claims here about high demand for Gemini - does anyone here have insight into how much of the load on Google is paid use vs the load from putting AI summaries into every web search?
Don't know, but Gemini 3.1 flash lite is available for free under relatively generous limits, and it had lots of random interruptions like when I was testing it. (Intermittently responding with errors due to high load.)
My curiosity is not the free AI summaries (which they can opaquely tune as necessary), but instead the renting of TPUs to Anthropic and OpenAI. Many of these contracts were announced last minute and seemed to involve a very desperate Anthropic. Based on the Anthropic/xAI data center contract, they’re willing to pay crazy markup to get immediate access to compute.
I want to know how impacted Gemini has been by that, because that will reveal a lot about their margins and revenue generating first party demand. Each MSFT earnings report they discuss the balance they’re dealing with between supplying GPUs to Azure customers and first party demand.
My pet theory is that Gemini is “losing” the LLM race because they’re preferentially selling the TPUs to competitors, while keeping just enough for themselves to stay competitive and build their own products.
We use Gemini for some specific tasks. It is often unavailable due to capacity limits or other downtime.
It's probably the best multimodal model I've worked with (if somebody knows a better one for audio analysis, please let me know!)
I don't know numbers, but their APIs have a bad uptime in my experience for some models. Too often failure because of "traffic too high".
Yeah I had a trial for AI Pro or whatever it's called and could never use Gemini CLI (when it still existed) because it was constantly "overloaded". Using the API directly (wihtout a subscription) sometimes works but the models are so buggy and the endpoints constantly spew errors that it's not usable. See this forum thread for example: https://discuss.ai.google.dev/t/frequent-503-errors-service-... it started with 503 errors since JANUARY and it's still not fixed. These are "stable" GA models!
I HIGHLY doubt that Gemini is overloaded, Google has been bullshitting with their crap models since release. Waste of everyone's time.
It's worse than OpenAI or Anthropic. However their lower tier consumer offerings can sometimes be had for <$10/mo on offer and come bundled with other Google services like cloud storage.
I do believe this will be the norm from now on to get access to top frontier model. Computing capacity plus state restrictions plus KYC will be imposed to organisations to get access, individuals will be served last on the queue with degraded performance. Once the Chinese models catch up, nobody (at least individuals) will turn back again to frontier labs.
This seems less about frontier models and restriction and more just lack of compute capacity to meet demand. This has always been an issue for large clients running on cloud, though not to this extent.
It's interesting that Meta is heavily using Google's models (as opposed to Anthropic or OpenAI) given that they are not SOTA for coding. I wonder if this for some strategic/competitive reason, or maybe for cost saving?
I would imagine there are many situations within Meta's applications where relatively small models can do a good job — sentiment analysis, abusive language detection, characterising users based on their posts, summarising a user's complaint so it can be ignored more efficiently, assessing whether ads are likely to be fraudulent so they can be run more often, etc.
Google tends to be very good at vision and smaller/ edge
Hmm ... I was assuming they were using these models for development, but I wonder if any of it might be for production instead - perhaps using vision models to analyze posted content? That would certainly be massive scale, but I'd have thought that scale would require them to be running in their own datacenters.
OTOH, if they are stressing Google's capacity then it seems it has to be for production use, which would relfect a massive failure on Meta's side given their investment in datacenters and AI. If they can't utilize their own models and datacenters, then maybe they should just rent the excess capacity to Google! :)
I double check with Gemini anything ML/AI related, anecdotal but I feel like it's much more solid explaining things and pointing out pitfalls.
Not really, especially recent gemini's tend to hallucinate unbelievably much especially with visual input.
And their safety tuning is neither effective nor precise on edge models.
> It's interesting that Meta is heavily using Google's models (as opposed to Anthropic or OpenAI)
Who says they aren't? Could be using all of them for "research".
Google is the only LLM frontier that can supply huge enterprise grade AI, yet still struggle, the other one is spacex but their LLM is Grok
also the only cloud platform, the only workspace, the only cloud drive... it's just standard Google fare
Image/video understanding still quite cost effective from the Gemini flash series models?
Image generation and veo models I’d imagine quite effective for creators; new Instagram accounts with AI content that are garnering millions of followers in spans of weeks are quite common now
Facebook does seem to be falling behind. Does anyone here use Llama over more recent options for any technical reasons?
Facebook is ethically challenged and that's putting it very very very mildly. Yes, they have unlimited money, but at a certain point, it comes across like a rich dude at a bar telling a beautiful woman that he'll buy her a diamond bracelet if she will just come over to his place right now. They make my skin crawl.
if you use this as a rough gauge: https://openrouter.ai/models?order=top-weekly
Llama Meta 70b is 50th or so down the list of popular models.
It has 24.1b tokens used in 7 days vs the top models that have trillions or hundreds of billions of tokens.
So practically dead!
Is that biased towards code generation? As opposed to application features using LLMs, which I think is more what we’re talking about.
Meta's latest model is Spark Muse and not available outside of its products.
https://ai.meta.com/blog/introducing-muse-spark-msl/
still waiting on that API launch which was supposed to happen very very soon
Meta builds its own models. How similar is this to a story with the headline “OpenAI limits Anthropic’s use of its ChatGPT AI models.”?
Not similar at all, as explained in the article below the headline.
How do they figure out it's being used by Meta?
... Because Meta have a contract with Google, are paying for the requests, and are supplying their API key with every request.
Must be to classify/moderate images for social media. They're pretty good at that. I can't imagine what else you'd want to use Gemini models for, certainly not coding.
I've criticized Antigravity in this same conversation, but Google Gemini is good at coding. Even Flash 3.5 low is good at coding. The problem is that Google isn't hungry anymore and it really really really shows in how much they've botched everything to do with Antigravity.
Demand for tokens raises exponentially, we are in the middle of a compute crisis, and people still think AI is a bubble...
Misleading title on HN but an interesting article, a reminder of why the hyper scalers are investing heavily in infrastructure.
That said, I expect much of the AI bubble to pop. Google Gemini with Antigravity is a good product, as is a Claude Code subscription but I have switched to using DeepSeek v4 Pro with the Claude Code harness and DeepSeek v4 Flash with the OpenCode harness (when I am not using local models with little-coder/pi) and at least for the foreseeable future I don’t think I am going back. Fast APIs at low cost trumps having to spend a little more time to get the same quality of results.
I am a huge fan of Google Gemini, but Antigravity is not a good product. Just recently I've had issues with:
* Repeated instances of incorrect code insertion that the agent cannot clean up. Sure, version control, but this is often happening in new files that aren't even in version control yet.
* Lost chat history when I close and restart the app.
* Not being able to restore a chat from the history (just saw this last week).
* Overly broad searches that waste time and tokens.
* No vertical scroll bar arrows. WTF?? Doesn't the interface look "flat" enough already? This feels arbitrary and stupid.
* The previous chat prompt takes up a large portion of the vertical space of the chat window, even on a high res display.
When it works Antigravity is excellent. When it doesn't work, it's absolutely horrible. If you check the update history, there are usually just a few items and they're super generic things like "Fixed a bug with text entry.".
I don't see it improving at any kind of reasonable pace, even over the last 6 months As a result, I've mostly relegated Antigravity to a planning tool and it does an excellent job. Or I use it to write prompts that I give to Codex. It definitely can do an excellent job writing code sometimes, but sometimes it also does an absolutely horrible job with not breaking the code when it inserts it. It seems to be terrible at understanding C++ braces. How often? Way too often. I always know it's happening because it prompts me to run Git while it's doing something. LOL, that's how I know that it's broken something.
Codex is definitely way, way, way better. It's not even a contest at this point. Codex never breaks my code. It might not always do what I want, but it's just an order of magnitude better than Antigravity. Antigravity really feels like a comedy of errors at this point. ESPECIALLY from a company with Google's resources.
Using LLMs for development is not efficient. All of the problems these companies are having trying to provide enough compute and energy are proof.
Understanding the actual problems we are trying to solve with code and efficiently coming up with solutions (essentially, pre-LLM development) will always be better than wastefully brute forcing solutions with LLMs.
Counterpoint, 80% of the code I write is not hard and I don't care about it past it being close to a reference implementation
There can be a knitflex for Oculus Quest 3 as Valkyrie Emergate for 1080,pre order Meta yarn (requires Facebook login.). The cymflex renders at 720p in horizon.