- Cerebral Valley
- Posts
- IREN is revolutionizing data centers with renewable energy 🔋
IREN is revolutionizing data centers with renewable energy 🔋
Plus: CTO Denis Skrinnikoff on GPU cloud solutions for AI applications...
CV Deep Dive
Today, we’re talking with Denis Skrinnikoff, CTO of IREN.
IREN is revolutionizing data centers by focusing on power-dense workloads powered by renewable energy. Initially focused on Bitcoin mining to bootstrap the business, IREN has diversified its platform to cater to the burgeoning AI market. Denis, who joined IREN in 2021, brings extensive experience from his background in data centers at RackForce and TeraGo. His expertise and IREN's strategic vision have supported the company's rapid growth. IREN now boasts a power portfolio exceeding 3,000MW and continues to expand.
Founded by Dan and Will Roberts, IREN capitalizes on excess renewable energy and supports energy networks, blending their expertise in infrastructure, renewable energy, capital markets and real assets. Their vision is to provide next-generation data centers, powered by 100% renewable energy and specializing in power-dense compute including GPU cloud solutions for AI applications. IREN’s Cloud Services business, powered by Nvidia H100 GPUs, offers state-of-the-art quality and impressive speed, making them a formidable choice for AI development. IREN IPO’d in 2021.
In this conversation, Denis discusses the founding story of IREN, how they differ from other data centers, and how they leverage renewable energy to power their data centers.
Let’s dive in ⚡️
Read time: 8 mins
Our Chat with Denis 💬
Denis - welcome to Cerebral Valley! First off, give us a bit about your background and what led you to join IREN?
Hey there! My name is Denis Skrinnikoff. I'm the CTO of IREN. I joined IREN back in 2021, and ultimately I was led here simply due to their strategy. I was in communication with the founders at the time, and they had the concept of building data centers close to excess renewable energy. I came from a data center background, having previously worked at RackForce and TeraGo, which had multiple data centers across Canada. I ultimately ran the cloud and colocation business units for those companies, and just understanding IREN’s's vision and the potential for explosive growth made it a no-brainer. Since then, IREN has grown rapidly. We IPO'd a few months after I joined, and now it's grown to have a land and power portfolio well in excess of 3,000MW and is constantly growing.
Can you give us an overview of what IREN is and does?
Yeah, so IREN owns and operates next-generation data centers, specifically optimized for power-dense workloads, and all powered by 100% renewable energy. For our AI Cloud Services customers, our entire fleet comprises Nvidia H100 GPUs. We're looking at some of the next-generation alternate technologies and provide cloud services either on a reserved or on-demand basis. What sets us apart from our competitors is our expertise in managing power-dense workloads. We own and operate our own data centers, including the land, facilities, and substations. Because we are end-to-end owners and have the expertise, we can offer excellent customer support and a cost-effective, flexible, high-performing GPU cloud offering. Ultimately, it's my intent to grow our platform and product roadmap with a suite of tools specifically designed for AI developers to help them quickly and effectively run their jobs.
Who are your users today? Who’s finding the most value in what you’re building with IREN?
So right now, primarily our customers have been AI startups and scale-ups looking for GPU cloud services. All of our customers today have been in various stages of training their AI models and primarily come from other providers where they are either price-sensitive or performance-sensitive. Recently, we've seen a significant influx of customers looking to run inference workloads and start monetizing their training. Our GPU service is well-suited for both types of customers and flexible enough for anyone who needs access to GPUs.
Right now, given AI is still in its relatively early stages of adoption, our customers are primarily AI startups and scale-ups. But ultimately, the platform is built to be able to service enterprises as well, and is equipped for both training and inference workloads.
Realistically, a startup would come to us when you're at a point where they need compute resources, especially clustered resources. We specialize in clusters and clusters as a service, which is the most challenging aspect—creating proper working InfiniBand fabrics and having a stable environment. It has been a challenge in the space. We're ready to support anyone along that journey, whether they're at the POC and training stage, or they're ready to do long-term inference.
Any customer success stories you’d like to share? And how do you measure success for the customer?
Every customer is different, and no two customers have used the same training schedulers, models, or goals. Primarily, our customer base comes to us because they want to train their workloads and accelerate their time to market. Our customers have told us that we have some of the best uptime and best-performing machines relative to other providers they have experienced.
Ultimately, success for a customer is being able to maximize their uptime and performance, helping them achieve their goals. Everyone is in a sprint right now to produce the next generation model and provide something they can monetize and solve a business problem. We specialize in supporting those customers effectively.
IREN focuses on using excess renewable energy – is that an economic decision or a climate decision?
That's an excellent question. The political landscape is changing, and traditional data centers often rely on non-renewable energy sources. Now, with the growing power requirements for new models, leveraging renewables places us in a strong position and aligns with our core values. We expect this conversation around power consumption in AI to continue to be front of mind. It's also very economically beneficial. Where there is excess renewable energy, power prices tend to be lower due to the surplus energy. The grids we are connected into have high reliability, ensuring excellent uptime.
There are many teams sprinting to provide GPU compute to AI teams right now. What’s the competitive landscape like and what sets IREN apart from a developer's perspective?
Let's start with traditional data centers. The data center space that has existed up until today was primarily built around general purpose compute workloads. GPU workloads always had a place for gaming resources, rendering, scientific studies, and machine learning, but it was all fairly low density. With the advent of generative AI, the need for clusters has increased significantly. Clustering requires high density to minimize latency within the cluster. Traditional data centers were designed to support five to six kW racks, and high-density data centers targeted ten to twelve kW racks.
Nvidia's reference architecture now requires a minimum of 40kW rack density for an effective cluster. We have been building data centers that do high-density compute, leveraging ASICs, and even before our H100 AI deployments, we were supporting upwards of 70kW racks with air cooling.
Being end-to-end owners and operators of data centers, we have the expertise to cool and manage these workloads, which we leverage for the benefit of our customers. You are not logging tickets with a third party or waiting on third party party vendors for assistance. We have in-house expertise for that, end-to-end.
Additionally, our data centers are already capable of handling the more power-demanding workloads that Nvidia's new products require. Since we own, operate, and manage everything under one roof, the process is more streamlined. It is also worth noting that we use 100% renewables, which will become increasingly important as large enterprises prioritize green initiatives and seek providers like us.
For developers, there's a lack of a true on-demand platform in the industry. Large hyperscalers have many caveats, and there's no great alternative for building on-demand clusters. We are well-situated to become a leader in this space, providing customers the ability to do burst capacity and custom-sized clusters without sacrificing performance or dealing with all those caveats.
What’s the biggest technical challenge in building IREN?
The biggest challenges aren't really technical but more time-bound. The digital world scales exponentially, and the real-world infrastructure isn't built to accommodate that pace. For instance, look at the rapidly increasing number of parameters in AI models. The industry wasn't prepared for this tidal wave of demand, especially in terms of data center capacity and places to plug in GPUs.
Building a new data center and getting access to power can take four to seven years, including securing a grid connection and construction. This process requires significant planning, development, and engineering. Additionally, the lead times on some equipment are now multi-year. Since day one, we've been focused on high-density compute, and our approach has been disciplined and patient. We know what needs to happen, and we're working diligently to build the capacity to meet these upcoming workloads and commenced development at our sites years ago to support current demand.
How do you see IREN progressing in the next 6-12 months?
We're committed to growing our presence in the digital infrastructure sector. Currently, we have our data center facilities which we're using to offer our cloud services, but we're also looking to leverage these for colocation. By using our expertise and facilities, we aim to attract other customers while growing our AI cloud service.
On-demand service is very important in this space, and we believe we can build a platform that meets customer needs and becomes a game-changer for many companies. Until now, the process for acquiring our compute resources has not been instantaneous. Customers have had to reserve instances, as they typically need these clusters for extended periods. Due to high demand, as soon as we stand up a new cluster and have it ready, customers are interested. However, our goal is to leverage a large pool of resources and provide an on-demand option, which has always been our ultimate aim.
What trends in AI currently excite you the most? How do upcoming trends such as agentic and multimodal AI affect your roadmap for IREN?
Well, it's still very early days for AI, so there's still lots coming. One trend that I’m following closely is we're starting to see a path to AGI (Artificial General Intelligence). LLMs (Large Language Models) aren't AGI yet, but they're beginning to show us the underlying mechanisms of producing an AGI, a single model that could meet or exceed the capabilities of a human. I believe that in the next three to five years, we're going to see some major breakthroughs in this space that will require significant underlying infrastructure to enable.
Can you tell us about the culture at IREN? Are you hiring, and what do you look for in prospective team members?
The culture at IREN is extremely collaborative, innovative, and very customer-focused; cultural fit is the main thing I look for in hiring. We have a strong commitment to operational excellence and doing things right. We're a founder-led business, and the founders are from Australia, so we've adopted that culture. This means taking a long-term view of our business, ensuring success, maintaining low egos, not taking ourselves too seriously, and actually having fun and enjoying our jobs. Our team is growing quickly, especially on the technology side, and we're hiring for multiple roles. I'm always on the lookout for top talent, so our career page always has many tech roles posted.
Conclusion
Read our past few Deep Dives below:
If you would like us to ‘Deep Dive’ a founder, team or product launch, please reply to this email ([email protected]) or DM us on Twitter or LinkedIn.