> The Claude Platform on AWS is a first of its kind offering for Anthropic, giving you all native Claude API features from day one. Anthropic operates the service and data is processed outside the AWS boundary.
So it's not... On AWS... ?
This statement sounds.... Backwards?
I get they have another option that is in AWS, but this continues the cryptic naming problem AWS already is overloaded with
I think the idea is that you can launder your team or product AI spend through your AWS account. This matters in Enterprise. It looks like the difference with Bedrock is that you access more "Claude platform" stuff than just the model.
More charitably, this lets an org heavy on AWS use their existing IAM / SSO / Finops processes to manage Claude stuff, this is genuinely helpful when otherwise you have to go thru several teams and build out whole new rails to adopt.
I've always wondered how this plays out in practice. I might certify that I have signing authority but I most certainly do not. What happens in the US (in Delaware?) when there's a dispute?
We had a customer try to back out of a contract by claiming the person signing didn't have authority. It didn't work because the person's manager (who has authority) was included in all of the communication.
Legally it didn't matter whether the signer had authority because the way the signer's company behaved during the signing process implied that the signer had authority.
E.g. If the CTO at a company tells a vendor to "send the contract over to my product manager" then the CTO created the impression with the counterparty that the product manager has authority, and the company will be hound to the contract based on that fact regardless of whether the product manager actually has authority or not.
I'm sure it's more nuanced than this, but my understanding is actual authority is less relevant than implied authority. E.g. if you have your board of directors take away the CEO's authority to sign a contract, it doesn't automatically invalidate everything the CEO signs, since a counterparty can reasonably assume that the CEO has authority just based on their job title.
Generally any W-2 has authority to enter into contracts, strictly from the vendor’s POV. As a vendor you don’t need to get your customer’s publicly listed officer or director to sign off on contracts. The W-2 can also be fired for entering their employer into the contract, but that's not (directly) the vendor's problem.
Once a vendor has entered into a contract, that could change - e.g. "any change orders must be approved by $EMPLOYEE_SET".
It's absolutely wild that every W-2 employee can expose their employer to essentially unlimited liability, but AFAIK, that's the truth.
This is my day job. I couldn't get access to the Claude Platform even with a business goal justification because of the management overhead while having Anthropic model access with Bedrock.
Through AWS, assuming the underlying data governance is reasonable, this will be a much easier pill to swallow.
yes it sounds like a hack to get access to untracked spend in corporate accounts.
In my org, I have to file a form for reimbursement if I bought a pencil for $0.25 but in AWS? spend varies by +/- $5k per month and nobody even questions it. This will definitely make it trivially easy for me to build on Anthropic's services without even telling anybody vs the hoops I would have to jump to get it paid for another way.
Nah that's not what's happening here. This service is offered under AWS Marketplace. The only argument is actually probably a shared billing console, and that's where it ends. Won't matter for small companies, small fish, but the for the big pond this means new contracts to check, lawyers and so on. So not really a "revolution" happening. News for startups, yes, but not so much for the big corps or gov.
Another selling point has been a guarantee of 1:1 api feature and design parity between Anthropic and this Claude platform. Helps if you have workloads you want to balance between providers.
> Claude on Amazon Bedrock keeps AWS as the data processor and operates within the AWS boundary. This is a good fit for companies that have strict regional data residency requirements or need their data processed exclusively within AWS's infrastructure.
Yeah i think this could backfire. At the moment they have such a clear messsage with Bedrock about data governance. You now have to ask a question and probalby get approval where previously there was no question and hence no barriers.
At this point I can only assume that AWS wants to have this naming issue. It’s an issue they have everywhere. Sagemaker is the worst offender. Only a solution architect can guide through such confusion…
As a long time Amazonian I can tell you it's simply because UX designers basically don't exist in Amazon (in case that wasn't obvious), and the ones that do exist are extremely bad at their job.
there’s a top level feature in aws for investors to give out credits of like $120k of AWS spend during funding rounds. there’s min commits of spend for cheaper prices (RI). funneling costs and invoicing though aws has real benefits. aws spend monitoring is literally a sub industry with billion dollar players
The credits you get from aws in their startup program are typically not spendable on marketplace. At least what we got through YC we could not spend there. Not sure how claude is integrating, maybe it’s different here
Yeah, as someone with strict export compliance concerns which forces us to use Bedrock because its exclusively us-based inference in our AWS account, this does nothing for me. Frankly, nothing Anthropic has shipped over the past 6 months besides the models themselves has been useful to our company, despite running into the same problems they're trying to solve with all of those features (managed, remote agents). There's not really a good solution, as AgentCore runtime sucks and is expensive. You basically have to build this yourself because nobody is solving for self-hosted managed infra for agents, and we don't really have the time to build this sort of system on top of building our actual product. It's very frustrating for them to put this out as a win, when it doesn't help the people who are using AWS Bedrock to begin with.
A little startup I'm part of is on the AWS startup program giving us £10k in AWA credits so the more we can "proxy" through AWS the better.
We already heavily rely on anthropic models via Bedrock but I'll be interested to see if the tok/s throughout is better on this new service (or worse).
To be honest though after a quick skim, I'm unclear what other advantages this might offer over Bedrock where we can already access the models including vision etc. Will it be worth refactoring our services, all our terraform etc? Unclear at this stage, especially since Bedrock allows us to use more than just the anthropic models if needed
Are you sure those credits apply to this usage. Many forms of AWS credits dont cover usage of Claude models on Amazon Bedrock for instance, as all of them are billed as third party / marketplace consumption.
I dont think this is about billing as other comments mentioned. AWS bedrock already does that.
I think AWS proposes to host actual agents, meaning you can customize their MCP servers, have them fetch and issue arbitrary requests, etc. Essentially a hosted minimal harness.
This is very much needed, as a form of "hosted claude code", allowing you to actually have the agent code, push, test from e.g. your phone.
Bedrock provides model inference as well as a few hosted things like agent core. This certainly provides more than bedrock, but it implies inference while fronted by aws policy and management will be farmed out to spacex and other providers. Anthropic already announced managed agents prior to this.
I would note that I don’t think I would lean towards Anthropic (operating with one 9 currently) over aws (operating with an implied 5 9’s within a region and arbitrary 9’s by composing across regions). Anthropic makes good models. They’ve yet to prove they make good operations.
> The Claude Platform on AWS .. giving you all native Claude API features .. Anthropic operates the service and data is processed outside the AWS boundary. This is a good option for companies that want the full Claude Platform experience.
Does seem to be mostly about billing like others said. But it might mean cloudformation / terraform providers for claude-platform, guess that's nice.
It might make strict networking/firewall things slightly easier somehow. But for everyone who thinks the new offering is about jurisdictional matters, it's not, that's the old one:
> Claude on Amazon Bedrock keeps AWS as the data processor and operates within the AWS boundary. This is a good fit for companies that have strict regional data residency requirements or need their data processed exclusively within AWS's infrastructure.
So claude.bedrock is where you run if you want complete data privacy, this - claude.aws - is just claude on/in AWS - is that the right core difference?
Interesting timing - been building with Claude API locally and hit AWS infra questions.
Anyone know if this solves cross-region failover? Main pain point I have is US vs EU latency differences when running agents in parallel. Local orchestration helps but cloud fallback would be useful.
There's companies out there with mid 8, possibly low 9 figure monthly AWS bills. In that world, letting people pay through said AWS pipeline gets rid of so much red tape. Supporting Bedrock was already a big winner vs OpenAI in the race for big enterprise, and Anthropic desperately wants to find any way to be sticky, as opposed to risking companies flipping a switch and telling entire departments to switch to any future codex + GPT on Bedrock plans. The harder it is for you to migrate, the better for Anthropic.
I get this, but isn't this a complete compliance failure?
What's the point of having all those loops to onboard vendors if you can just buy from AWS marketplace (which AFIAK is not a particularly high bar to achieve for SaaS options)?
Like imagine $POOR_QUALITY_VENDOR. If they go through the normal channels they might get shot down. If they get procured on AWS Marketplace, then it feels to me in many organisations 'its fine', though AWS does minimal checking?
Yes. You can pay your Anthropic bill thru AWS, and use AWS access controls to manage access to Anthropic APIs. For companies all in on AWS this lowers the friction to adopt Anthropic tools significantly.
I use Claude Code, the max plan, for 200$ a month. Privately. So if I convince my peers at work to use this, will it cost significantly more? Because it's token usage based? Not flat rate? And if so, how much more? I think I burned through 400m tokens in two months in Codex. Can someone shed on light on this?
Is this a bunch of AI coded slop? I feel like everyone should be skeptical of how quickly Anthropic is throwing random things out there. Why would anyone use this instead of using Claude directly on AWS?
anthropic and aws have been pretty close from the start
i think the more interesting part is that anthropic is to aws what target was to amazon
this feels like there's a coming 5-10 year change from aws acting as a cloud, to being a cloud entrypoint/marketplace, separately from the cloud marketplace they already have as cfn snippets you can deploy
will claude platform on aws offer the same nerfed (dumb) opus 4.5/4.6/4.7 of the recent weeks, or will it be possible to use opus 4.5/4.6 from february/march?
this is a bad bad idea people, i highly recommend not falling for this one. You dont wanna see your production database get deleted by mistake or spawn 464135453452 ec2 instances due to a mistake in autoscaling configuration
> The Claude Platform on AWS is a first of its kind offering for Anthropic, giving you all native Claude API features from day one. Anthropic operates the service and data is processed outside the AWS boundary.
So it's not... On AWS... ?
This statement sounds.... Backwards?
I get they have another option that is in AWS, but this continues the cryptic naming problem AWS already is overloaded with
I think the idea is that you can launder your team or product AI spend through your AWS account. This matters in Enterprise. It looks like the difference with Bedrock is that you access more "Claude platform" stuff than just the model.
More charitably, this lets an org heavy on AWS use their existing IAM / SSO / Finops processes to manage Claude stuff, this is genuinely helpful when otherwise you have to go thru several teams and build out whole new rails to adopt.
> I think the idea is that you can launder your team or product AI spend through your AWS account.
This is exactly it. For any reasonably sized org, setting up new contracts with new vendors involves a lot of procurement, lawyers, negotiations, etc.
If a team can just click a button in AWS, there’s no issue.
This is a product / solution that solves an organizational problem, not a technical one.
I wouldn’t even call it a hack as much as extremely common a strategy.
Sadly it’s going to be more nuanced.
The Bedrock models, at least, have additional click through EULAs for Anthropic models. You’re going to need to review and agree to those as well.
Claude is going to be marketplace spend and that’s usually capped towards your PPA at 25%.
> click through EULAs
Every year "don't agree to things on behalf of the company"
Every day "click here to agree that ..."
I've always wondered how this plays out in practice. I might certify that I have signing authority but I most certainly do not. What happens in the US (in Delaware?) when there's a dispute?
We had a customer try to back out of a contract by claiming the person signing didn't have authority. It didn't work because the person's manager (who has authority) was included in all of the communication.
Legally it didn't matter whether the signer had authority because the way the signer's company behaved during the signing process implied that the signer had authority.
E.g. If the CTO at a company tells a vendor to "send the contract over to my product manager" then the CTO created the impression with the counterparty that the product manager has authority, and the company will be hound to the contract based on that fact regardless of whether the product manager actually has authority or not.
I'm sure it's more nuanced than this, but my understanding is actual authority is less relevant than implied authority. E.g. if you have your board of directors take away the CEO's authority to sign a contract, it doesn't automatically invalidate everything the CEO signs, since a counterparty can reasonably assume that the CEO has authority just based on their job title.
Generally any W-2 has authority to enter into contracts, strictly from the vendor’s POV. As a vendor you don’t need to get your customer’s publicly listed officer or director to sign off on contracts. The W-2 can also be fired for entering their employer into the contract, but that's not (directly) the vendor's problem.
Once a vendor has entered into a contract, that could change - e.g. "any change orders must be approved by $EMPLOYEE_SET".
It's absolutely wild that every W-2 employee can expose their employer to essentially unlimited liability, but AFAIK, that's the truth.
Well, you see, I had my cat click "submit", so we don't have to pay the bill!
"Practice is policy"
No, "This is exactly not it." They are buying your data on a cheap.
As someone who is dealing with the procurement of both in a medium sized it, finops and infosec are exactly it.
Do you have experience selling to fortune 100 sized organizations?
“I don’t have the budget for this but we have AWS credits” is something teams beg for all the time.
When people beg to give you money, you accept it. Why? It’s not some conspiracy theory. You accept the money because it’s money.
100% correct... Have EPD or PPA? Reduced spend because of reasons? Well now you can make it up in claude tokens.
This is my day job. I couldn't get access to the Claude Platform even with a business goal justification because of the management overhead while having Anthropic model access with Bedrock.
Through AWS, assuming the underlying data governance is reasonable, this will be a much easier pill to swallow.
yes it sounds like a hack to get access to untracked spend in corporate accounts.
In my org, I have to file a form for reimbursement if I bought a pencil for $0.25 but in AWS? spend varies by +/- $5k per month and nobody even questions it. This will definitely make it trivially easy for me to build on Anthropic's services without even telling anybody vs the hoops I would have to jump to get it paid for another way.
Nah that's not what's happening here. This service is offered under AWS Marketplace. The only argument is actually probably a shared billing console, and that's where it ends. Won't matter for small companies, small fish, but the for the big pond this means new contracts to check, lawyers and so on. So not really a "revolution" happening. News for startups, yes, but not so much for the big corps or gov.
Another selling point has been a guarantee of 1:1 api feature and design parity between Anthropic and this Claude platform. Helps if you have workloads you want to balance between providers.
It is basically like invoicing through AWS Marketplace.
> I think the idea is that you can launder your team or product AI spend through your AWS account.
Can confirm that this is the one and only reason that we use Claude through AWS
Also you can spend your commitment contracts :p
Isn’t that capped at 25%?
Something like that, but if your boss went too high in your commitment contract, it's nice to have different options to put there...
Yeah its a "marketplace private offer"
As other people have pointed out, it makes contract signing much easier.
THe other side effect is that it bumps up your spend, possibly to the point where you are eligible for "private pricing" ie global discount.
So its a win-win for most people.
> Claude on Amazon Bedrock keeps AWS as the data processor and operates within the AWS boundary. This is a good fit for companies that have strict regional data residency requirements or need their data processed exclusively within AWS's infrastructure.
Seems like there are two different options.
Yeah i think this could backfire. At the moment they have such a clear messsage with Bedrock about data governance. You now have to ask a question and probalby get approval where previously there was no question and hence no barriers.
At this point I can only assume that AWS wants to have this naming issue. It’s an issue they have everywhere. Sagemaker is the worst offender. Only a solution architect can guide through such confusion…
As a long time Amazonian I can tell you it's simply because UX designers basically don't exist in Amazon (in case that wasn't obvious), and the ones that do exist are extremely bad at their job.
there’s a top level feature in aws for investors to give out credits of like $120k of AWS spend during funding rounds. there’s min commits of spend for cheaper prices (RI). funneling costs and invoicing though aws has real benefits. aws spend monitoring is literally a sub industry with billion dollar players
The credits you get from aws in their startup program are typically not spendable on marketplace. At least what we got through YC we could not spend there. Not sure how claude is integrating, maybe it’s different here
Yeah, as someone with strict export compliance concerns which forces us to use Bedrock because its exclusively us-based inference in our AWS account, this does nothing for me. Frankly, nothing Anthropic has shipped over the past 6 months besides the models themselves has been useful to our company, despite running into the same problems they're trying to solve with all of those features (managed, remote agents). There's not really a good solution, as AgentCore runtime sucks and is expensive. You basically have to build this yourself because nobody is solving for self-hosted managed infra for agents, and we don't really have the time to build this sort of system on top of building our actual product. It's very frustrating for them to put this out as a win, when it doesn't help the people who are using AWS Bedrock to begin with.
Synthetiq offers self-hosted (local or in your cloud)
A little startup I'm part of is on the AWS startup program giving us £10k in AWA credits so the more we can "proxy" through AWS the better.
We already heavily rely on anthropic models via Bedrock but I'll be interested to see if the tok/s throughout is better on this new service (or worse).
To be honest though after a quick skim, I'm unclear what other advantages this might offer over Bedrock where we can already access the models including vision etc. Will it be worth refactoring our services, all our terraform etc? Unclear at this stage, especially since Bedrock allows us to use more than just the anthropic models if needed
Are you sure those credits apply to this usage. Many forms of AWS credits dont cover usage of Claude models on Amazon Bedrock for instance, as all of them are billed as third party / marketplace consumption.
It’s included for Claude models on Bedrock.
Initially that wasn’t the case, until April 2024.
https://aws.amazon.com/blogs/startups/aws-activate-credits-n...
I recently received AWS startup credits and claude models via AWS Bedrock are included.
Builder credits apply to every AWS service I've tried.
Ours have done in other similar bedrock programs.
Yeah it's included
I dont think this is about billing as other comments mentioned. AWS bedrock already does that.
I think AWS proposes to host actual agents, meaning you can customize their MCP servers, have them fetch and issue arbitrary requests, etc. Essentially a hosted minimal harness.
This is very much needed, as a form of "hosted claude code", allowing you to actually have the agent code, push, test from e.g. your phone.
Bedrock provides model inference as well as a few hosted things like agent core. This certainly provides more than bedrock, but it implies inference while fronted by aws policy and management will be farmed out to spacex and other providers. Anthropic already announced managed agents prior to this.
I would note that I don’t think I would lean towards Anthropic (operating with one 9 currently) over aws (operating with an implied 5 9’s within a region and arbitrary 9’s by composing across regions). Anthropic makes good models. They’ve yet to prove they make good operations.
> The Claude Platform on AWS .. giving you all native Claude API features .. Anthropic operates the service and data is processed outside the AWS boundary. This is a good option for companies that want the full Claude Platform experience.
Does seem to be mostly about billing like others said. But it might mean cloudformation / terraform providers for claude-platform, guess that's nice.
It might make strict networking/firewall things slightly easier somehow. But for everyone who thinks the new offering is about jurisdictional matters, it's not, that's the old one:
> Claude on Amazon Bedrock keeps AWS as the data processor and operates within the AWS boundary. This is a good fit for companies that have strict regional data residency requirements or need their data processed exclusively within AWS's infrastructure.
it's mostly about procurement. a lot of companies use AWS IAM for single sign on in a multitude of services.
"mostly about billing" NO. People, this is about your data.
What exactly do you think is happening here? The customers using this already have a trust relationship with both Amazon and Anthropic.
how so? the data stays with Anthropic
So is this effectively a way for aws customers to more easily access Claude code?
Claude itself was almost from the very beginning available in bedrock.
Might be for enterprise customers (with slow procurement departments) who already have AWS accounts but not Anthropic ones.
So then they could use Bedrock, no?
A lot of tools, community and Anthropic’s, have zero support for Bedrock.
Anthropic’s offerings for Bedrock lag behind their main platform by months, maybe up to a year or more.
Bedrock is both more expensive, less feature complete, and less reliable in terms of raw volume of 500 errors.
> Claude Platform on AWS will be available in most AWS commercial regions and support global and U.S. inference geographies.
Isn't there the possibility to have EU-based inference?
It allows one to conveniently hide your claude expense in your big ass AWS bill.
I tried that with Bedrock but Claude models are billed through Marketplace and generate a completely separate invoice.
Suspect this is probably the same.
Need to just pay all your other vendors through marketplace too!
So claude.bedrock is where you run if you want complete data privacy, this - claude.aws - is just claude on/in AWS - is that the right core difference?
I hope Azure is next. I would love to ask my employer to finally get this...
Interesting timing - been building with Claude API locally and hit AWS infra questions.
Anyone know if this solves cross-region failover? Main pain point I have is US vs EU latency differences when running agents in parallel. Local orchestration helps but cloud fallback would be useful.
So confusing. Are they with AWS for the checkout? Because none of sentences make sense to me
There's companies out there with mid 8, possibly low 9 figure monthly AWS bills. In that world, letting people pay through said AWS pipeline gets rid of so much red tape. Supporting Bedrock was already a big winner vs OpenAI in the race for big enterprise, and Anthropic desperately wants to find any way to be sticky, as opposed to risking companies flipping a switch and telling entire departments to switch to any future codex + GPT on Bedrock plans. The harder it is for you to migrate, the better for Anthropic.
I get this, but isn't this a complete compliance failure?
What's the point of having all those loops to onboard vendors if you can just buy from AWS marketplace (which AFIAK is not a particularly high bar to achieve for SaaS options)?
Like imagine $POOR_QUALITY_VENDOR. If they go through the normal channels they might get shot down. If they get procured on AWS Marketplace, then it feels to me in many organisations 'its fine', though AWS does minimal checking?
Yes. You can pay your Anthropic bill thru AWS, and use AWS access controls to manage access to Anthropic APIs. For companies all in on AWS this lowers the friction to adopt Anthropic tools significantly.
It sounds like an “official” Anthropic API billed from an AWS account?
It is
"Claude in the cloud" has a nice ring to it.
I wonder how Anthropic will adjust their models as local models compete more closely with API models.
Probably not for another 5-10 years that it matters.
Right now, local LLMs are too expensive to run and not smart enough to matter much.
How's this work? Like if I want to self host claude on my gpu's? Do I just pay a one time lease?
This offering is purely a billing formality, not a way to run inference on your hardware.
Interesting move in to fully managed service by Anthropic.
Anthropic operates the service and data is processed outside the AWS boundary.
Waiting for them to launch Google Claude Platform (GCP).
Nice, might give managed agents a try now.
How to obscure your wasteful LLM usage behind your already massive AWS expenditure with this one tip.
I use Claude Code, the max plan, for 200$ a month. Privately. So if I convince my peers at work to use this, will it cost significantly more? Because it's token usage based? Not flat rate? And if so, how much more? I think I burned through 400m tokens in two months in Codex. Can someone shed on light on this?
So what's the good point?
Hide some of your ballooning AI spend behind your ballooning AWS bill
But then how would I be able to brag about how much I'm doing the AI?
Don't worry, you can still get on linkedin and write a bragging post about how many tokens your engineers are spending.
Couldn’t you already do that with Claude via AWS?
But you couldn't do it with the Claude app or the recently announced Claude Managed Agents I think.
At least now we can expect more usage limits.
Might be good?
Umm, why could this not just be Claude on AWS marketplace?
Seems intentionally deceitful.
Is this a bunch of AI coded slop? I feel like everyone should be skeptical of how quickly Anthropic is throwing random things out there. Why would anyone use this instead of using Claude directly on AWS?
anthropic and aws have been pretty close from the start
i think the more interesting part is that anthropic is to aws what target was to amazon
this feels like there's a coming 5-10 year change from aws acting as a cloud, to being a cloud entrypoint/marketplace, separately from the cloud marketplace they already have as cfn snippets you can deploy
will claude platform on aws offer the same nerfed (dumb) opus 4.5/4.6/4.7 of the recent weeks, or will it be possible to use opus 4.5/4.6 from february/march?
this is a bad bad idea people, i highly recommend not falling for this one. You dont wanna see your production database get deleted by mistake or spawn 464135453452 ec2 instances due to a mistake in autoscaling configuration
I think you might have misunderstood the headline and maybe not read the article, because that's not really relevant to what's being announced here