Replicate is joining Cloudflare

Congrats Ben and team!

I think this is Cloudflare's most notable acquisition yet? From Wikipedia it looks like they've previously mainly acquired smaller cybersecurity firms like https://en.wikipedia.org/wiki/Area_1_Security

Acquisition or acquihire? I don't see a price tag, which seems to imply the latter?

How many players are there in this space? Replicate, RunPod, Modal, Northflank, FAL, ... Who are the big ones? It's pretty crowded, right?

FAL was smart. They ditched the "run any model" to focus on just image and video, and now they dominate that space. They raised a pretty substantial round recently. Though I don't think there's any moat and they'll soon face competition too.

What about these vs. the routers like "Open"Router?

Cloudflare is one of those companies similar to palantir where the valuation just makes you scratch your head. But bringing in AI to try and fill out the AI valuation makes some sense I suppose

why sell if they were doing good?

Looks more of an aquisition for me. They are integrating the 50000 models into worker AI.

I don't know that this was an acquisition in the sense that the Replicate investors and team made bank. I don't see a price tag, and the market for these "run model" infra companies is pretty crowded.

Congrats Ben and team. Looking forward to seeing what you build with Cloudflare's backing!

Great match up of technologies. Congrats to Ben and the team! I had the privilege of working with Replicate when I was at AWS, such a great team.

A very smart move for Cloudflare and a huge missed opportunity for Render [0].

[0] https://news.ycombinator.com/item?id=42833414

> Replicate’s going to carry on as a distinct brand

Uh huh

Well, yeah, for a year or maybe two.

Really?

Cloudflare isn't solely a CDN anymore. CDN and DDOS-protection were the most logical "first" products to build based on their SDN ( Software Defined Networking).

A cloud is the next thing and there's a lot of money involved with the cloud. I see them as the only real competitor/challenger to Azure, AWS, GCE, ... because they aren't bound to regions ( less DevOps)

For example, what you might not know about Durable Objects => https://boristane.com/blog/what-are-cloudflare-durable-objec...

This feels the most like Cloudflare branching out to something closer to a traditional hosting model and further from their networking efforts (though things like workers and R2 certainly have blurred that line).

Very smart move - I assume the Workers will get tight integration with the Replicate APIs

I really like both these companies. Replicate just make it so easy to bundle a bunch of different AIs up into a prototype - but I think they have tonnes of untapped potential that Cloudflare could fulfil.

I'm not saying they're not an interesting company I'm just reflecting on their extraordinary valuation multiple which puts them in a unique league that are predominantly AI bets. Their founder is uniquely strong storyteller. If you look at their capital investment it's just not anywhere close to a top tier cloud player.. so while they offer some interesting building blocks it's just not clear that large workloads move there anytime soon

(https://substackcdn.com/image/fetch/$s_!-PwA!,f_auto,q_auto:...)

You don’t know how good they were really doing and you don’t know how good the offer was.

From what I know they were the most used inference provider by developers a few years ago, but since the Together AI and Fireworks only grew while Replicate seems to have stayed quiet. It’s a highly competitive low-margin business, so volume is critical and if you’re losing volume then you’re doomed.

I'd love to work on Cloudflare's underlying email platform... from pretty early on, I've felt like the workers model could be great for handling relatively short lived tcp connections, not just web requests... in this way, all the other facilities CF offers such as R2 and D1 could be used with an email platform that could scale incredibly well. With this acquisition, it's another natural fit into such a platform for email spam filtering and classification.

Definitely exciting to see.

Cloudflare went x 10 since their IPO 6 years ago ( stocks).

Revenue:

$ 85 million (2016)

$ 287 million (2019) IPO year

$ 1,670 million (2024)

$ 2,154 million (2025)

https://www.macrotrends.net/stocks/charts/NET/cloudflare/rev...

That's not a crazy valuation multiple considering their growth.

Palentir is a body shop. Cloudflare is infrastructure/cloud. Very big difference ( at least to me)

Stock ( Cloudflare - https://finance.yahoo.com/quote/NET/holders/ ): 90.87% % of Shares Held by Institutions

Stock ( Palantir - https://finance.yahoo.com/quote/PLTR/holders/): 60.09% % of Shares Held by Institutions

I love witnessing companies getting acquired to form monopolies.

I wrote this comment [0] very recently and when I wrote it had in mind that Cloudflare might very well end up being a key player in a more centralized Internet that has developed far away from its original architecture.

Defense against threats is a pretty strong centralization incentive in different kinds of networks - social, biological.

I could imagine that a lot of people are investing based on similar scenarios in their minds.

[0] https://news.ycombinator.com/item?id=45946365

> that has developed far away from its original architecture

This was always their architecture, if you watched closely.

My predication 5 years ago - https://news.ycombinator.com/item?id=26821438

> And cloudflare is self hosting it on more edge locations ( i think they could even join the big 3 soon)

It's not a question of crazy or not, it's just an extreme end of the multiple spectrum.. and the company is not in a positive operating margin position. So you're right they have a strong track record and there's optimism they turn a profit, but it's highly speculative. Nothing wrong with that. Just makes sense they are fleshing out their AI story while they're in a position to invest with equity IMO

Their not investing with equity. Cloudflare is cash flow positive and re-investing.

Joining the big three requires capital investments orders of magnitude beyond where they are. Nevertheless I'd like to see someone do it, and if not Equinix then it could be them. But sadly the #4 right now appears to be Oracle.

Probably because their last funding round was in 2023 and they may have been looking for another round because their revenue and cost were probably not really adding up to the valuation needed for another round. Running GPU hardware at scale is just expensive

Given the price was not announced, it seems investors decided to exit via acquisition instead of trying to raise a multi hundred million round consistent with a multi billion dollar valuation and a high level of ambition that investors are shooting for in this space.

If the amount was high, they'd be be bragging about it. Given that they aren't, it might be on the lower end of the 2023 valuation for the then 40M round.

Just speculating here; I don't have any more information. Basing this on my understanding of how this stuff works. This might actually be the opening round for a few more such acquisitions of the somewhat risky investments of a few years ago of companies that are probably not going to turn into trillion $ unicorns. There are lots of pretty well funded startups in this space converting investment capital into cloud GPU cost. I think some level of consolidation is overdue and might take away some building concern about over exposure in the market. Big banks and investors might be getting nervous.

replicate "only" raised $50m, so I'd wager the founders, investors and early employees did well here.

If they raised $500m... then yes, the wouldn't have done so well.

Probably, but you can check out a more robust list here: https://blog.cloudflare.com/tag/acquisitions/

* BastionZero

* Kivera

* Baselime

* PartyKit

* Area 1

* Vectrix

* Zaraz

* Linc

* S2 Systems Corporation

* Neumob

* Eager

* CryptoSeal

* StopTheHacker

Recently implemented it to only process emails where the sender is in the list of customers email and then extract info from email and do an api call.

Really easy to do it on the edge. No hassle with POP3, IMAP, ... Easy to use a subdomain of my email domain

Loc: ~40

Not everything seems to be tagged as acquisition. Dyte.io was acquired, and announced here: https://blog.cloudflare.com/introducing-cloudflare-realtime-...

I havent looked into the financials but the reasons I really like cloudflare is that its workers are free without card for 100k and even after that its ridiculously cheap and cloudflare tunnels and a lot of features are free and really appreciate it

I feel like people like to rub that one time cloudflare messed up when I mention it but it was a gambling website and I feel like cloudflare could've better communicated it but overall its got so much less drama than the other cloud providers and its genuinely being really nice imo

But imo, cloudflare is really dirt cheap for just starting out and at scale as well especially if using cf workers

I feel like cloudflare can make a bank in enterprice section but their pricing model also feels the most saner compared to the shady tactics used by google or others with our marketing and privacy

I know that the internet is getting centralized but I feel like there are some ways of de-centralizing it, (by example archiving web pages and then seeding them, helping on internet archive or something similar as well)

As an internet user, cf feels mid but sometimes as a guy who just wants to deploy shit or basic apis, I "vibe coded" a cloudflare worker api which I actively use so much for my own purposes setting up a custom redirector and everything without paying anything at all, I think I like it.

Honestly, nothing is as good or as bad as it seems except palantir's evaluation which makes me feel like 448 pe ratio or something similar drove me nuts the other day.

Did a CF worker for an email group list which worked well enough... having each email address target have to approve themselves was about the only hiccup. Definitely nice options all around.

Would be interested in seeing an Outlook/(o365/Exchange) alternative from Cloudflare with a bit more of a feature rich offering... or at least the ability to build something like that on CF.

I was wondering if/when cloudflare would jump into doing “more AI”. Of all the paths they could have taken, this seems like a reasonable one. Hope that the dev team made some money/cloudflare stock from this transaction.

> all that’ll happen is that it’s going to get way better

Every acquisition announcement says something like this. It is almost never true. In fact, the opposite often happens.

> Of all the paths they could have taken

They can't innovate themselves, and rather than try to fix the reasons why (change in leadership, corporate structure, etc), they just buy a competitor, which they will most likely run into the ground. Perfectly reasonable!

From an AI integration perspective, I am hopeful that Cloudflare may be able to improve "performance on the cheap" for replicate's models a little bit.

Replicate has had multiple ways to deploy for auto scaling and you can just keep running periodically to keep the system in a booted and warm state, but that has always seemed like it would be too expensive for a broke bootstrapper like me so I avoided it and model popularity was a big deciding factor. Also because of that and the potential for boot up, in general I avoided it for latency-sensitive things.

I guess there is a limit to what you can do. At some point someone has to spend the money to have the resources stay ready.

But with Cloudflare, theoretically the pool of potential users goes up, and it becomes more likely for someone to have already booted your model.

At the moment I am especially interested in performant and easy ways to run models like "sensefvg/InteractiveOmni-8B" or Qwen 2.5 Omni or models that are even more all in one than that like OpenAI Realtime or Gemini Live.

Now that Ernie 5 launched with (Omni) multimodality built in, I think within six months, developers are going to start to expect speech-to-speech capability from major AI lab releases or product line ups. I feel like eventually the spatial-temporal understanding of video models will be merged in too to make the models understand the world better. But speech in and speech out is closer to being a standard expectation.

Instead of running three models for STT->LLM->TTS with a bunch of tricks like eager end of turn or speculative decoding that basically mean you run the LLM twice or on two different models, and possibly getting shut down by API rate limits, the speech to speech models are a single model that both understands and generates audio as well as text such as for function calls.

This is probably an annoying comment because I am immediately trying to increase the requirements to not only being every model for cheap, but every model for cheap in in a low latency real time streaming way. I just happen to have a contract now that has shown me that multimodal like voice to voice is much more convenient but also much more expensive and fewer options.

Replicate has been so awesome though. Within like a day of me requesting InteractiveOmni, lucataco had it up. So another annoying comment, I sure hope he got paid.

They already did and they had Worker AI for quite a while [0]. But I think it didn't catch up and they have there very old models.

Overall I think it's a good acquisition for both of them. Replicate would have to have more competitive pricing to compete with fal.ai

[0] https://developers.cloudflare.com/workers-ai/models/

Don't pretend that 98% of startups don't actually dream of an exit or payday.

I wonder if they'll continue to support and develop Cog [0], which I've found quite useful for local deployments of models (as well as to Replicate's servers).

[0] https://github.com/replicate/cog

Well I'd assume the acquisition is with equity and that much of their employee compensation comes from equity, neither of which show up in a free cash flow view. Still you do make a good point. I'd have expected bigger capital costs that even if amortized would show up on cash flow some of the time.. but they appear to run a very tight capex ship. So more security SaaS than cloud provider IMO

The one time I’ve interacted with cog, predictably, it delivered none of its benefits compared to a vanilla pyproject.toml, and all of the downsides of being, ironically, not replicable.

its better for founder because they can have their exit payout

Replicate is my preferred go to for image models and easy proof of concepts for language models, so congratulations to them!

If you (or others here) are truly interested in email+tcp proxying+workers-like-computing, then I have work to offer you in the build out of my company’s network/computing platform.

This is something that you can or could already do with Envoy proxy with the right amount of polish, knowhow, and elbow grease (which we have).

What Cloudflare doesn’t have are good general computing products and multitenancy supprt. But we have those, something like D1, and something like a remote tunnel that runs in your browser. So users can launch Stalwart instances painlessly and use our secret magic to render it in actual browsers without special middleware or extra state (or we could offer shared/default Stalwart to do it completely out of the box). So that would take you the rest of the way from platform-level email to application-level and user-level/managed email.

What we don’t have is the time to spare in productizing that and handling the email-specific routing in envoy, or finding capable/knowledgeable people who know all the email-specific content and skills. So hit me up if that's you, and otherwise, feel free to run wild with the knowledge that you can configure Envoy Proxy with an L4 network filter or HTTP filter that delegates to dynamically loadable/configurable wasm if you want a hackable Cloudflare workers or even a platform-level alternative (hint: store wasm code in FUSE). The L4 filter should work for email filtering.

[0] https://www.envoyproxy.io/docs/envoy/latest/intro/arch_overv...

[1] https://github.com/proxy-wasm/spec/tree/main/abi-versions/v0... (where you'd start implementing email filtering logic)

[2] https://mail.mplode.dev (demo/test Stalwart instance we're running in our dev environment. Platform magic would allow us to render an email client directly embedded in https://brilliant.mplode.dev via remote IMAP/POP3)

thanks; excellent analysis and makes much more sense to me now!

Is this the platform they are offering on Openrouter?

https://openrouter.ai/provider/cloudflare

Replicate was far from a monopoly and cloudflare was arguably worse off in this space. Together they still aren’t anywhere near monopoly status for running models.

Their area of expertise is in running large scale infra, which they have innovated a lot to get. They’re acquiring a company that has expertise in an area they're not familiar with, which needs large scale compute infra. It seems like their needs are aligned in this case.

not sure but that's quite possible. I haven't used cloudflare workers ai myself because they had quite limited number of models and choices so far.

As a Replicate customer, I’m excited about this news. Cloudflare has the strategic incentive to invest in this line of business and it puts Replicate in the hands of a very good product marketing and sales team to get the volume it needs for sustainability.

Who are the other ones? I'm aware of fal apart from Replicate

for investors, it will highly depend on the valuation ($350m) and liquidation preferences. so, even if they "only" raised $50m, seeing almost or none return is plausible. this also includes employees who exercised their options. another story is for the founders

My perception (engineer at modal.com) is that Replicate lost out to Fal.ai more than they lost to Together AI and Fireworks.

Are there indicators in Cloudflare's culture or history to suggest that Replicate's strengths (docs, api, design) will remain in the long-term?

I spent a moment wondering why did Cloudflare buy an on-prem tooling startup https://www.replicated.com

The cloud is just someone else’s on-prem

you've not been paying attention if you think this is their first foray

I don't think that much will change regarding to Replicate's API because of Cloudflare. It's specifically mentioned in Replicate's blog post and it's also not in their best interest.

What's important to know ( I think). Recently, Cloudflare released a blog post of Omni for AI inference. I think they performance tuned it better than other providers. So their costs per inference drops down a lot ( https://blog.cloudflare.com/how-cloudflare-runs-more-ai-mode... ). Since the performance is OK, they now want to expand usage and their model catalog.

Replicate is a perfect fit. Model catalog, infrastructure for larger models, more specialised tools for fine-tuning, ...

Eg. For inference, Replicate is basically just a Worker AI endpoint and easy to maintain. Fine-tuning could probably be something similar.

But then again, that's my 2 cents. It was already mentioned that Replicate will stay as a distinct brand.

It feels like more related to this: https://blog.cloudflare.com/how-cloudflare-runs-more-ai-mode...

And doing it their way than the traditional way.

Note: Their innovation seems to lie in smaller and fine-tuned ( broader catalog) models than larger ones. So Replicate seems a perfect match.

The traditional way: Rent a GPU and run inference on an container.

Now you can see the great plan going on... First, positioning itself as the "gatekeeper" of AI services, then owning such a service to be able to abuse of your position...

Great move. Congratulations to the Replicate team!

Together Fireworks Hugging face Modal AWS Azure Google CoreWeave

To name a few but on hwy 101 today there were billboards for more

wowwww

>cloudflare including replicate goes down the very next day

Edit: Although I still stand with what I say, this comment has also aged like fine milk with cloudflare taking down aws,chatgpt and x and sooo many others that cloudflare outage had more upvotes on HN (if I am correct, I can be wrong tho, nothing new about it as well ) and thus more impact...

Honestly now the only stable provider which doesn't have an outage is google cloud in sick twisted fate

I am wishing when google cloud has its outage so I can recommend everybody to use hetzner (yes I know its not an apples to orange comparisons but hetzner has some crazy good uptime)

Never a dull day

They didn't.

https://www.replicated.com -> https://replicate.com

Workers is a very expensive computing facility. Yeah, it is cheaper than AWS lambda but people don’t use that for everything.