I gave Replicate a shot but needed to run on my own GPUs, so I initially used Cog to port the workload.
I quickly realized Cog was an obstacle rather than an accelerator. I replaced it with a lightweight FastAPI layer, which immediately unblocked me:
1. Native I/O with Google Cloud Storage.
2. Freedom to use the latest Torch and Nvidia Docker images without abstraction overhead.
3. Running Torch and TensorFlow in parallel (legacy model constraints that Cog struggled with).
It forces the question: What is Replicate's value proposition for a startup where the founders are competent engineers? If you aren't afraid of a Dockerfile, the "ease of use" premium evaporates.
The answer to that question is likely this acquisition.
The standalone AI middleware market is precarious; the landscape shifts too fast and technical founders will eventually outgrow the training wheels.
Folding into Cloudflare gives the team a sustainable home to leverage the platform's scale, rather than competing solely on a container abstraction layer.
Wish them the best. Cloudflare’s infrastructure is likely the right environment to turn this into a high-leverage product
Me too. After trying it out I found Cog to be super frustrating to use and the only use case for me ended-up being trying out a new model through the web UI occasionally.
It's maybe obvious why Replicate might want to be part of Cloudflare.
It's less obvious why Cloudflare want Replicate.
As for the price:
> Replicate has raised $52.5M in funding from investors like Andreessen Horowitz, Y Combinator and Sequoia Capital, with last known valuation of $350M [2023]
It would be interesting to know how much hype there is in valuations since 2023. I assume it's mostly vesting options because I doubt Cloudflare has the cash to throw around. I would guess $500M valuation but I could be off by a lot.
Cloudflare has made considerable improvements in running small models in gpu's world-wide ( loading multiple small models).
I think they want to be the provider of inference for specialized slm's. Replicate is a perfect acquisition for that, they have a large catalog of smaller models.
With the acquisition they are saying that they have made enough improvements for the next step, earning money from those improvements.
I searched but could not find the "bought" or "money" or "dollar" or "stock" words in the marketing fluff piece, so it definitely does not answer the question in the title.
The term "joining" irritates me more than it should, because you're correct in asking "What is the value of the transaction?". My guess is that they aren't joining anything, CloudFlare bought the company and is keeping the team.
Sure, but Replicate will probably cease to exist in the near future. So a more accurate title could be: Cloudflare buys out Replicate and transfers staff to internal teams.
That's not a given as well. An acquisition usually involves restructuring the acquired company, sometimes in a way where the original team ceases to exist.
> An incredible journey is: One company buying another and closing its services down. This is a purchase of the second company’s staff, rather than their product. An acquihire.
> This is what is galling. A company that can afford to pay millions for some new staff but not for what those staff built. The people who used the service, and invested their belief and time in uploading photos, or forming friendships, or logging data, are left to find new virtual homes while their former hosts enjoy a nice (if possibly delayed) payday.
> This repeated pattern only encourages more people to create flashy services that have no hope of being sustainable businesses in their own right, but may survive long enough, with VC funding, to attract the attention of a large company eager for new ideas and staff.
The last paragraph is what gets me -- it makes sense to me found startups in hopes to be acquired (continue their work with the support of a big company), but founding with the intention to abandon your users? Yuck.
> We’ll be able to do things like run fast models on the edge, run model pipelines on instantly-booting Workers, stream model inputs and outputs with WebRTC, etc.
Benefit to 3rd party developers is reducing latency and improving robustness of AI pipeline. Instead of going back and forth with https request at each stage to do inference you could make all in one request, e.g. doing realtime, pipelined STT, text translation, some backend logic, TTS and back to user mobile device.
Depends on how much the latency matters to you and the customers. Most services realistically won't gain much at all. Even the latency of normal web requests is very rarely relevant. Only the business itself and answer that question though.
> "Even the latency of normal web requests is very rarely relevant."
Hard disagree. Performance is typically the most important feature for any website. User abandonment / bounce rate follows a predictable, steep, nonlinear curve based on latency.
I've changed the latency of actual services as well as core web vials many times and... no. Turns out the line is not that steep. For the range 200ms-1s, it's pretty much flat. Sure, you can start seeing issues for multi second requests, but that's terrible processing time. A change like eliminating intercontinental transfer latency - barely visible in results in ecommerce.
There's this old meme of Amazon seeing a difference for every 100ms latency and I've never seen it actually reproduced in a controlled way. Even when CF tries to advertise lower latency https://www.cloudflare.com/en-au/learning/performance/more/w... their data is companies reducing it by whole seconds. "Walmart found that for every 1 second improvement in page load time, conversions increased by 2%" - that's not steep. When there's a claim about improvements per 100ms, it's still based on averaging multi-second data like in https://auditzy.com/blog/impact-of-fast-load-times-on-user-e...
In short - if you have something extremely interactive, I'm sure it matters for experience. For a typical website loading in under 1s, edge will barely matter. If you have data proving otherwise, I'd genuinely love to see that. For websites loading in over 1s, it's likely much easier to improve the core experience than split thing out into edge.
Ok, I think we're actually in agreement -- given your all-important qualification "for the range 200ms-1s". Yes, ofc, that first part of the curve above the drop is quite flat; there's hardly time for the user to get impatient and bounce.
My point about the shape of the curve stands. 100ms can matter more on the steepest part of the slope than 2s does further to the right.
Tech is full of ironies. 5 years ago cloudflare was held as the savior of internet. People in HN and tech in general put them on pedestal. 1.1.1.1, generous ddos protection, cdn, adn to name a few.
Fastforward to today, they being hated foe bringing down the internet, compared to failing giants.
I think it is a reminder that evil and good recides within oneself.
your comment actually made me think about this, cloudflare hasn't actually turned full monopolist... yet, their generous free tier has led to monopoly like market share but the pain society feels is when they go down, they haven't turned the screw to make monopoly (or hyperscaler) profits
They and Prince may never go that way, as someone who occasionally picked SaaS/infra stocks, the ratio of their market share/customers metrics to revenues/profit vs other peers was always on the low side (haven't look at their numbers in a while tho)
Some parallels to gmail, which similarly rose to dominance by offering a free tier that was leagues above what everyone else was offering. They were also hailed as the savior, followed by a period of disillusionment. Now plenty of competitors have caught up with almost-as-good free offerings, but the world is still much more centralized than it was before
I have no comment on Cloudflare being evil, but if you actually try to use their hosting products which come with a generous free tier, you realize how bad the DX is:
- Their dashboard is next to GCP in terms of how bad it is.
- They ship like three different CLIs that'll often have overlapping functionality: wrangler, c3, cloudflared and flarectl. It feels like an organizationally confused tooling strategy dumped on the user.
- Docs are often out of date
They really need to learn a thing or two from Vercel on the DX
> I think it is a reminder that evil and good recides within oneself.
It's more like "organizations that attain monopoly position find themselves in a bubble that becomes disconnected from reality, regardless of the quality of their intentions".
Most recent example is Google. Cloudflare next, probably.
HN changed, is what happened. The tech startup crowd has sadly migrated to Twitter, and I guarantee you they still love Cloudflare.
HN has become a forum where old school enthusiasts complain loudly about modern tech while refusing to examine the fact that they’re doing it on a forum that’s inherently built to stimulate the very capitalism they decry.
I hate them a bit already, but beyond bringing down parts of the internet from time to time and occasional captchas, they are not an everyday annoyance like Google, Microsoft, and the rest.
They are already? See the football debacle in Spain. It's not their fault, legally and technically, but if everything wasn't centralized by them, it wouldn't happen.
Hmm, an AI company joining Cloudflare, a company's whose major strength is to take their customer's websites offline, selling it as saving the internet and such.
Maybe when the AI overlords take over Cloudflare will be our last bastion of defense. :D
I gave Replicate a shot but needed to run on my own GPUs, so I initially used Cog to port the workload.
I quickly realized Cog was an obstacle rather than an accelerator. I replaced it with a lightweight FastAPI layer, which immediately unblocked me:
It forces the question: What is Replicate's value proposition for a startup where the founders are competent engineers? If you aren't afraid of a Dockerfile, the "ease of use" premium evaporates.The answer to that question is likely this acquisition.
The standalone AI middleware market is precarious; the landscape shifts too fast and technical founders will eventually outgrow the training wheels.
Folding into Cloudflare gives the team a sustainable home to leverage the platform's scale, rather than competing solely on a container abstraction layer.
Wish them the best. Cloudflare’s infrastructure is likely the right environment to turn this into a high-leverage product
Me too. After trying it out I found Cog to be super frustrating to use and the only use case for me ended-up being trying out a new model through the web UI occasionally.
It's maybe obvious why Replicate might want to be part of Cloudflare.
It's less obvious why Cloudflare want Replicate.
As for the price:
> Replicate has raised $52.5M in funding from investors like Andreessen Horowitz, Y Combinator and Sequoia Capital, with last known valuation of $350M [2023]
It would be interesting to know how much hype there is in valuations since 2023. I assume it's mostly vesting options because I doubt Cloudflare has the cash to throw around. I would guess $500M valuation but I could be off by a lot.
Cloudflare has made considerable improvements in running small models in gpu's world-wide ( loading multiple small models).
I think they want to be the provider of inference for specialized slm's. Replicate is a perfect acquisition for that, they have a large catalog of smaller models.
With the acquisition they are saying that they have made enough improvements for the next step, earning money from those improvements.
500mil would be roughly 12.5% of a available liquid assets for cloudflare.
They are a huge, huge middleman of internet traffic and many, many services.
Recent and related:
Replicate is joining Cloudflare - https://news.ycombinator.com/item?id=45953702 - Nov 2025 (68 comments)
I searched but could not find the "bought" or "money" or "dollar" or "stock" words in the marketing fluff piece, so it definitely does not answer the question in the title.
What was the value of the transaction?
The term "joining" irritates me more than it should, because you're correct in asking "What is the value of the transaction?". My guess is that they aren't joining anything, CloudFlare bought the company and is keeping the team.
> CloudFlare bought the company and is keeping the team.
So the team is joining cloudflare...?
After an incredible journey, I’m excited to announce a case of beer is joining my fridge.
I think it's more like: after an incredible journey, I'm excited to announce this hamburger is joining my stomach
Sure, but Replicate will probably cease to exist in the near future. So a more accurate title could be: Cloudflare buys out Replicate and transfers staff to internal teams.
And the money is joining some bank accounts!
> So the team is joining cloudflare...?
That's not a given as well. An acquisition usually involves restructuring the acquired company, sometimes in a way where the original team ceases to exist.
No they will all get sacked next year.
It's a pretentious way of saying they acquired them. https://blog.cloudflare.com/tag/acquisitions/
Let us see if Replicate and Cog are shut down, and it becomes an Incredible Journey: https://ourincrediblejourney.tumblr.com/post/89180616013/wha...
> An incredible journey is: One company buying another and closing its services down. This is a purchase of the second company’s staff, rather than their product. An acquihire.
> This is what is galling. A company that can afford to pay millions for some new staff but not for what those staff built. The people who used the service, and invested their belief and time in uploading photos, or forming friendships, or logging data, are left to find new virtual homes while their former hosts enjoy a nice (if possibly delayed) payday.
> This repeated pattern only encourages more people to create flashy services that have no hope of being sustainable businesses in their own right, but may survive long enough, with VC funding, to attract the attention of a large company eager for new ideas and staff.
The last paragraph is what gets me -- it makes sense to me found startups in hopes to be acquired (continue their work with the support of a big company), but founding with the intention to abandon your users? Yuck.
> What was the value of the transaction?
I think this is being intentionally kept under wraps, so nobody who can say anything knows.
The did explain a little bit:
> We’ll be able to do things like run fast models on the edge, run model pipelines on instantly-booting Workers, stream model inputs and outputs with WebRTC, etc.
Benefit to 3rd party developers is reducing latency and improving robustness of AI pipeline. Instead of going back and forth with https request at each stage to do inference you could make all in one request, e.g. doing realtime, pipelined STT, text translation, some backend logic, TTS and back to user mobile device.
You are seemingly answering something that they did not ask at all
Does edge inference really solve the latency issue for most use cases? How does cost compare at scale?
Depends on how much the latency matters to you and the customers. Most services realistically won't gain much at all. Even the latency of normal web requests is very rarely relevant. Only the business itself and answer that question though.
> "Even the latency of normal web requests is very rarely relevant."
Hard disagree. Performance is typically the most important feature for any website. User abandonment / bounce rate follows a predictable, steep, nonlinear curve based on latency.
I've changed the latency of actual services as well as core web vials many times and... no. Turns out the line is not that steep. For the range 200ms-1s, it's pretty much flat. Sure, you can start seeing issues for multi second requests, but that's terrible processing time. A change like eliminating intercontinental transfer latency - barely visible in results in ecommerce.
There's this old meme of Amazon seeing a difference for every 100ms latency and I've never seen it actually reproduced in a controlled way. Even when CF tries to advertise lower latency https://www.cloudflare.com/en-au/learning/performance/more/w... their data is companies reducing it by whole seconds. "Walmart found that for every 1 second improvement in page load time, conversions increased by 2%" - that's not steep. When there's a claim about improvements per 100ms, it's still based on averaging multi-second data like in https://auditzy.com/blog/impact-of-fast-load-times-on-user-e...
In short - if you have something extremely interactive, I'm sure it matters for experience. For a typical website loading in under 1s, edge will barely matter. If you have data proving otherwise, I'd genuinely love to see that. For websites loading in over 1s, it's likely much easier to improve the core experience than split thing out into edge.
Ok, I think we're actually in agreement -- given your all-important qualification "for the range 200ms-1s". Yes, ofc, that first part of the curve above the drop is quite flat; there's hardly time for the user to get impatient and bounce.
My point about the shape of the curve stands. 100ms can matter more on the steepest part of the slope than 2s does further to the right.
So how long until Cloudflare joins the hated monopolies list?
Tech is full of ironies. 5 years ago cloudflare was held as the savior of internet. People in HN and tech in general put them on pedestal. 1.1.1.1, generous ddos protection, cdn, adn to name a few.
Fastforward to today, they being hated foe bringing down the internet, compared to failing giants.
I think it is a reminder that evil and good recides within oneself.
your comment actually made me think about this, cloudflare hasn't actually turned full monopolist... yet, their generous free tier has led to monopoly like market share but the pain society feels is when they go down, they haven't turned the screw to make monopoly (or hyperscaler) profits
They and Prince may never go that way, as someone who occasionally picked SaaS/infra stocks, the ratio of their market share/customers metrics to revenues/profit vs other peers was always on the low side (haven't look at their numbers in a while tho)
Some parallels to gmail, which similarly rose to dominance by offering a free tier that was leagues above what everyone else was offering. They were also hailed as the savior, followed by a period of disillusionment. Now plenty of competitors have caught up with almost-as-good free offerings, but the world is still much more centralized than it was before
People forget about the time when they stood up yo patent trolls. I'm no fan boy but it is just interesting...
I don't think Cloudflare is already widely regarded as "evil", but I'm personally moving to Bunny.net, so maybe...
I have no comment on Cloudflare being evil, but if you actually try to use their hosting products which come with a generous free tier, you realize how bad the DX is:
- Their dashboard is next to GCP in terms of how bad it is.
- They ship like three different CLIs that'll often have overlapping functionality: wrangler, c3, cloudflared and flarectl. It feels like an organizationally confused tooling strategy dumped on the user.
- Docs are often out of date
They really need to learn a thing or two from Vercel on the DX
> I think it is a reminder that evil and good recides within oneself.
It's more like "organizations that attain monopoly position find themselves in a bubble that becomes disconnected from reality, regardless of the quality of their intentions".
Most recent example is Google. Cloudflare next, probably.
Fair enough
HN changed, is what happened. The tech startup crowd has sadly migrated to Twitter, and I guarantee you they still love Cloudflare.
HN has become a forum where old school enthusiasts complain loudly about modern tech while refusing to examine the fact that they’re doing it on a forum that’s inherently built to stimulate the very capitalism they decry.
Judging by the downvotes that showed up over my EU night, they just read HN later than I post but are still here…
Their entire business model is effectively centralizing the web. The downtime over the last two weeks shows some of the problem with that.
I have used their products and have more favor toward them than I do for the corporations you're referring to, but ultimately my question is the same.
My guess, just a time question. Is Cloudflare publicly traded?
They are, on the NYSE: $NET
I thought they were already on that list.
I hate them a bit already, but beyond bringing down parts of the internet from time to time and occasional captchas, they are not an everyday annoyance like Google, Microsoft, and the rest.
To me they are, using IPv6 and VPN means CloudFlare has trouble automatically identifying me and pesters me with image captcha loops.
Funny, I use them to provide ipv6 to ipv4-only endpoints.
He's talking about tunnelbroker from HE being treated as malicious traffic by Cloudflare
Of course, soon only cloudflare traffic will be legit because you know, everyone uses it.
They are already? See the football debacle in Spain. It's not their fault, legally and technically, but if everything wasn't centralized by them, it wouldn't happen.
Related:
Replicate is joining Cloudflare
https://news.ycombinator.com/item?id=45953702
I'm surprised none of the reasons were financial
Because Replicate has no chance of beating Fal or Together?
Because capitalism structurally favors the concentration of power and wealth.
> Why Replicate is joining Cloudflare
Because a lot of money was transferred from Cloudflare's bank account to the bank accounts of the stockholders of Replicate?
I think the article is more about why Cloudflare thought that was a valuable proposition.
Why do I have a shovel? Because I transferred $48 to Home Depot's bank account.
It's more like
Home Depot sold you a shovel because you had $48 to spare. they don't care about the why.
"Why" is about motivation, not mechanics.
And that's the motivation ...
came here to put this comment exactly ;)
Okay, this is the kind of post I have not read and will never read but I will still comment.
Why Replicate is joining Cloudflare? Because you paid money to acquire it. Why the fuck else? Ffs.
Hmm, an AI company joining Cloudflare, a company's whose major strength is to take their customer's websites offline, selling it as saving the internet and such.
Maybe when the AI overlords take over Cloudflare will be our last bastion of defense. :D
Cloud flare taking a strong anti ai scraping stance. Then turning around and acquiring an AI hosting service also feels contradictory
Cloudflare literally offers a paid scraping API as one of its services.
https://developers.cloudflare.com/browser-rendering/rest-api...
Buying an AI company when you have a monopoly on scraping most of the internet seems like good business sense.
Cloudflare is also big on serverless on the edge (where the edge is their cdn), including AI inference