‘MinIO embraced the S3 API as the standard. And today, MinIO’s adoption is larger than Amazon S3’s. Customers are in the cloud, across the cloud, in private clouds, all the way to edge. Our user application ecosystem is much larger than Amazon’s. We are the single largest player. We made the S3 API an industry standard. Now all these AI applications … are largely built on MinIO’s API,’ says AB Periasamy, MinIO CEO and co-founder.
Indeed, according to Periasamy, MinIO is the reason the AWS S3 standard has been adopted industry-wide, outside of the competitive cloud hyperscalers.
“Our user application ecosystem is much larger than Amazon’s,” Periasamy said. “We are the single largest player. We made the S3 API an industry standard. Now all these AI applications, because they are modeled after the cloud infrastructure, are largely built on MinIO’s API. [For example,] underneath all the popular vector databases is MinIO.”
[Related: Storage 100: The Digital Bridge Between The Cloud And On-Premises Worlds]
MinIO is focused on developing storage technology for AI and GenAI, an area where consolidating data from multiple sources and multiple clouds has become a critical first step, Periasamy said.
“Customers see if data is in different storage technologies, some here, some there, they can’t bring it to an AI practice,” he said. “For generative AI, they realized that buying more GPUs without a coherent data strategy meant GPUs are going to idle out. So they are now starting to consolidate data from different teams, different technologies, under one AI data infrastructure.”
For that reason, MinIO has developed strategic relationships with all the key AI infrastructure players, including Nvidia, Intel, and AMD, Periasamy said. Most recently, the company has initiated a relationship with processor maker ARM, an up-and-coming AI infrastructure company, and has expanded its Intel relationship. Intel is even an investor in MinIO.
There’s a lot going on at MinIO. For more information, read CRN’s full Q&A with Periasamy, which has been lightly edited for clarity.
How do you describe MinIO?
MinIO is an object store. It’s a data store for storing AI data. Our closest alternative is Amazon S3, but that’s a service. If you have data and want to store it in AWS S3, you have to take it to the cloud and to AWS. We are software, and you take our software to wherever there is data and build your data infrastructure. MinIO grew to become the most popular object store of choice. Before the cloud, people were storing their data on SAN, NAS, and virtual machines. But the traditional enterprise architecture is on its way out. Cloud is the gold standard for infrastructure. AWS is the blueprint.
Increasingly, we are seeing enterprises move towards AI and data as the heart of the business. AI was born in the cloud and is built around the cloud. Our customer base is bringing us and Kubernetes in. We became the AWS S3, and Kubernetes became the AWS EC2. Customers are taking us to their colos and data centers and building their AI and data infrastructures on top of us.
You said ‘object store’ and ‘AI’ a lot. Is MinIO’s object store aimed specifically at AI?
The company grew as a general-purpose object store. For the last 10 years, all kinds of use cases came on top of MinIO, from simple web development to log data processing, cybersecurity, and all kinds of analytics and database workloads. In the last three or four years, we started seeing increasing pull towards database, analytics, ML (machine learning), and deep learning. In the last 12 months, we saw a significant pull towards GenAI. Now the commercial focus of the company is on the AI data market. We are quickly scaling our business. Pretty much now it’s a fight between MinIO versus the public cloud. When you store even just hundreds of petabytes of data, it just becomes unsustainable. As you cross 40 or 50 petabytes, public cloud becomes very expensive. And as the scale of the data grows, customers are leaving public clouds and coming to MinIO. Now we are seeing significant commercial traction in AI workloads. It allows us to focus only on AI data workloads.
We also happen to be open source. We have an open-source version of MinIO which we offer for general purpose use. All kinds of people run it in very small use cases. Even the guys who run MinIO with hundreds of petabytes also run MinIO as a home NAS. They literally put MinIO on Raspberry Pi and a home NAS system like Synology or Qnap. It just runs everywhere on all kinds of devices. …
The commercial version of MinIO is focused on AI data use cases simply because we’re seeing traction in the AI market and the scale has grown. That also leads to bigger check sizes, so it’s certainly lucrative.
When you say AI, you’re talking specifically about GenAI?
Customers see if data is in different storage technologies, some here, some there, they can’t bring it to an AI practice. For generative AI, they realized that buying more GPUs without a coherent data strategy meant GPUs are going to idle out. So they are now starting to consolidate data from different teams, different technologies, under one AI data infrastructure. And now the chief data office is becoming an important business priority because when you consolidate all your data in one place for AI, they need to make a choice: public cloud or private cloud. Public cloud is where they initiated this movement, and now they are bringing it to private cloud. The first step is, consolidate all the data that is private to the enterprise because that’s its core asset. When you do, it quickly grows to hundreds of petabytes. Some customers are now reaching the exascale range. Even when a customer starts at 100 or 200 petabytes, they are telling us, ‘I need technology that can scale beyond exabytes, because that’s what we need to do for our AI infrastructure.’ We see a mix of all kinds of data. Some is stable data, like Apache Iceberg-type semi-structured open table format. Some is log data, audio, video, call center logs, customer support, conversations, and documents. All these data types are starting to be consolidated under one common catalog. Once that happens, businesses can talk about (building a GenAI practice).
You need to have an AI data infrastructure at the heart of your business. When you do this, what would you build it on? Not SAN, NAS, or traditional appliance models. You need to build it modeled on AWS. And that’s where it’s us versus AWS. Customers come to us and think of themselves as hyperscalers. They think of servers and drives as commodities. You go to a colo because what you want to own is your data. With vendors like Red Hat or OpenShift for Kubernetes and MinIO for object store, they can go to a colo and build their AI practice.
Is MinIO’s technology aimed at specific applications, or is it for any application that needs AI data?
We could have introduced yet another API, but it would be another standard. We needed to pick one. When we started, we made a conscious bet on S3. Amazon was the gold standard in the cloud. Within AWS, S3 was the de facto standard. But S3 was unknown outside of the public cloud. No applications existed. MinIO embraced the S3 API as the standard. And today, MinIO’s adoption is larger than Amazon S3’s. Customers are in the cloud, across the cloud, in private clouds, all the way to edge. Our user application ecosystem is much larger than Amazon’s. We are the single largest player. We made the S3 API an industry standard. Now all these AI applications, because they are modeled after the cloud infrastructure, are largely built on MinIO’s API. [For example,] underneath all the popular vector databases is MinIO. We’ve become the de facto standard.
MinIO has a couple bits of news. First, you’re working with ARM. What are you doing here?
On the compute side, we’ve seen the industry shift towards GPUs and other accelerators because modern workloads use a lot of vector processing. … The CPU era is coming to an end, and GPUs are taking over. ARM is a CPU, but it is a different architecture. It’s not x86. We are starting to see ARM playing an important role on the data side as well. On the compute side, even Nvidia GPUs still have PyTorch and a bunch of code that needs to run on a CPU. They embraced ARM. And on the data side, we are starting to see ARM making inroads. It already happened in the cloud. Look at Amazon’s Graviton 3 ARM-based processor. It is a serious threat to the x86 architecture. Google Cloud and Azure also embrace ARM. We haven’t seen ARM adoption in the enterprise. That’s starting to happen. The problem is not a matter of technology. It’s the software ecosystem. We play an important role in the AI ecosystem. These are all new world applications which are already getting ARM compatibility. We played an important role on the data side, because we need to crunch data at ridiculous speeds. …
Talk to AI people, and what’s the number one concern? Power. We need to conserve every bit of power so we can give back to the CPUs, GPUs, and accelerators. ARM is power efficient, but at the same time, it is showing significant performance promise. ARM is also getting ready for 400-gigabit RDMA and other technologies to help keep GPUs busy. The improvements we made to ARM are not just a matter of software porting. ARM has certain functions useful for high-end data processing. That capability was added to MinIO.
MinIO recently expanded its Intel relationship. What’s going on there?
Intel has the Intel Tiber AI Cloud. The company recognizes the importance of getting into the cloud ecosystem and also enabling a new generation of cloud applications to support the Intel architecture. That cloud is built around MinIO for the object store side. An enterprise can go to the Tiber Cloud and build AI applications and data applications on the cloud, and when they mature, they can take that blueprint and build it for themselves. That’s what Intel is enabling. Intel is not trying to compete with the cloud, but give enterprises a private cloud and the blueprint so they can build it for themselves. Their own cloud is built on MinIO for the data.
Are there any similar special relationships between MinIO and AMD or Nvidia?
Clearly, Nvidia is the GPU leader of the pack. They have a huge advantage over the competition. We see the close number two is AMD, and then Intel Gaudi 3, which is showing promise. I think in about 12 to 18 months, AMD will have their software ecosystem ready. What I’m hearing from customers, the AMD Instinct MI300X accelerators are getting ready. The software ecosystem is catching up.
It’s good for the overall industry to have multiple options. Competition drives innovation. It will push Nvidia to the next level. It’s not like Nvidia is showing any signs of slowdown. … Intel is on our board and is one of our investors. But there are no strings attached with Intel. We are very closely working with Nvidia and AMD.
What’s the competitive environment for MinIO?
It’s public cloud versus us. If you have lots of data, are you going to put it in the public cloud, or are you going to take control of your data and build a private cloud? The fight comes down to public cloud versus us. We are the private cloud for your AI data. The public cloud has an answer. Public cloud works today, scales today. The problem comes down to economics. And if businesses go to legacy storage players, the wheels will come off the cart. They are not ready to handle AI scale.
Is MinIO a profitable company?
We’re cash flow positive. We haven’t touched the last round of funding we raised.
How much money does MinIO have on hand if needed?
I can’t disclose it completely. But I can tell you, our last B round of funding raised $103 million. It was supposed to be $300 million. We didn’t need the money, so we only took $103 million out of that, and that is collecting interest. We haven’t even touched that interest.
So you don’t expect to need more investment?
Our board meetings are often about our investors getting stressed that we are not spending the money. It’s a good meeting to have. I would rather have this problem than them asking me to downsize and do layoffs. There is so much interest from our investors. They want to pour money into the company. I didn’t want to take that money until we see how we could invest. I don’t want to buy growth. I want to invest.
We see that AI is super-charging everything. The scale has grown multiple times. Enterprises are now talking about exascale like they talked about one petabyte or two petabytes in the past. Because the check sizes are larger and there is so much money in the market, budgets have come, and customers are like, ‘What can I do in the AI space to modernize the business?’ It’s an exciting time for us now. We see a clear path for us to become a household name in the enterprise. Think of AI data? Think of MinIO.
You will see major investments from us on the marketing side. The product is very mature. On the sales side, there is clear inbound traction. These are large customers coming inbound because they love the product. There was some adoption in those companies already. It became a critical part of their business, like how Red Hat grew inside the enterprise. So they come to us. Marketing is the biggest investment that we would make now.
Does MinIO primarily work directly with customers or through indirect channels?
We have been building for the last year a very robust channel partnership network. We have hired some top talent in that space. We go to market both with resellers and through distributors. We have a relationship with Carahsoft, for example. We have several relationships with the big systems integrators around the globe, and partner on major deals with them. We are growing our relationship with the likes of WWT, for example. So it’s a very broad play. And because of the inbound nature of the business, coupled with the kind of very horizontal nature of what object store is, we have a very large ecosystem through those resellers and distributors.
So channels are a huge part of our business. We’re looking to build relationships with those partners that have the biggest mind share in major enterprises. And we want to do that because that’s where a lot of these AI conversations are initiated.
Any acquisition strategy?
We actually create technologies. For instance, we built a very powerful, large-scale vector database by ourselves. Then we didn’t want to compete with our own partners, so we killed it. We said, ‘OK, we’ll win the AI storage market and leave the vector database to our partners.’ We also built MinIO SQL. It’s a very powerful log processing technology like Elastic Search and Splunk. We didn’t release that to market. We built a video search engine that we didn’t release to the market. Building tech is the easiest part of the business. Building a business around it is a whole different game. It’s very difficult. Go-to-market is always difficult. Why do we keep building these products? If we don’t build them, we will go insane, right? If we go into M&A, it will be more about who can accelerate us on the customer side.