Skip to playerSkip to main content
  • 9 hours ago
Transcript
00:00IRON is an AI cloud that is building infrastructure all over the world, not just the United States.
00:05In all these geographies, Dell is its key server partner, right?
00:09It wants the NVIDIA GPUs. Couldn't think of a better conversation to have.
00:13It's kind of kick off the day. Dan Roberts, co-founder, co-CEO.
00:16You've been a busy guy as well on the M&A side.
00:20You've acquired a marketing or creative agency.
00:24I mean, basically, why?
00:26But there's also this conversation. Caroline and I have it all the time.
00:30About the public's perception of what's really happening with data center and how it impacts them.
00:35Yeah, look, brand awareness. We're going global. We're growing our customer base.
00:40It's really important to us. But that's a small point in the grand scheme of things.
00:43It's all about power and time to compute.
00:46And the issue is you can't implement a software update to bring on power.
00:51The tech industry is very brilliant at going really fast.
00:54But you can't code your way to an AI gigawatt of compute.
00:58And that's what we're focused on.
00:59In the United States, at least, there is some sort of public backlash, right?
01:04In zip codes where data centers are getting built, the public feels the pressure of higher energy prices.
01:09What's that like for Iron? And what are you confronting in that respect?
01:13Yeah, eight years ago when we started this business, we had a differentiated strategy.
01:16We never went to metropolitan areas.
01:18We went to regional communities where there was an abundance of renewable energy that creates a cost, a structural advantage
01:25over time.
01:26And we've reinvigorated local communities across British Columbia, Canada, Texas.
01:31And it's an advantage for us because we are located where people want us to do business.
01:36Dan, I think I'm right in saying, just as with everyone else, you're in a position where demand for compute,
01:42your compute, is vastly outpacing your ability to supply it.
01:45What is the biggest barrier to you, getting more capacity online, getting more stuff built?
01:53Absolutely. I mean, we're still in the dial-up era of AI.
01:56When you step back, the amount of friction you've got, waiting for prompts, you've got to craft your prompt really
02:01carefully.
02:01You know, complex prompts take 15 to 40 seconds to actually spit out an answer.
02:06As that speeds up, demand accelerates, which drives more compute and more use cases.
02:11And for us, it's been a focus on what is the end choke point.
02:14And that's power. It's steel. It's concrete.
02:17It's actually building the physical infrastructure that takes years and years.
02:20If you wanted to start today and build a gigawatt AI factory, you are looking 2030s before you get the
02:27first compute online.
02:28Let's get a bit deeper on that.
02:29So if you wanted to start today, in other words, you choose a site, you get permissions, you're not going
02:35to get it online for six years, four years.
02:38It'll take you 18 to 24 months to get the attention of a utility to work out whether your block
02:43of dirt even has power or not.
02:45That's the starting point.
02:46Now we're getting to it. What are the utilities not doing that you need them to do?
02:50Well, utilities, they're doing the best they can.
02:52Like, they're running complex networks with a lot of generation on one side, a lot of power demand on the
02:57other.
02:57And at the end of the day, when you connect to them, this is why grid-connected capacity is so
03:02valuable.
03:02When you connect to them, they're underwriting 24-7 guaranteed power.
03:07So they're naturally risk-averse and they're naturally dealing with an inundation of requests at the moment.
03:12I'd love to go back to basics for a second.
03:14You build data centers and those data centers have servers inside them and those servers are for AI workloads.
03:21What's it like working with a Dell?
03:24What is it that you rely on Dell for?
03:26And how do you get hold of NVIDIA GPUs, right?
03:29There is this idea that you can just phone Jensen Wong and say, hey, I need some of those.
03:33But in practice, it probably doesn't really work like that.
03:36Yeah, look, we've got a great relationship with both Dell and NVIDIA.
03:39We announced a partnership with NVIDIA for five gigawatts of compute.
03:42They also made an investment.
03:44Yes, so they're exposed to upside in us as we grow and deploy their GPUs and we're working on their
03:51reference architecture for a gigawatt factory in Sweetwater.
03:55But the opportunity with those partners is to work closely and build up the whole ecosystem, all the way from
04:01land, steel, concrete, down to the end users of this compute.
04:05Is there a geography somewhere in the world that is the best place to build a data center right now?
04:09Look, every geography is slightly different in terms of how they handle things, but the majority of our capacity is
04:15in Texas and they've been pretty good.
04:18If you could ask Mr. Wong and Mr. Dell one thing right now or ask something of them, what would
04:25it be?
04:26Oh, look, I think the whole industry is moving from if silicon is no longer the choke point, how do
04:31we solve steel?
04:32How do we solve concrete?
04:34How do we solve the kilowatts?
04:36And this president and this administration, they have an answer for that?
04:40I think everyone's trying to work it out.
04:42It's been an interesting evolution for your company, right?
04:46Now public in the United States, Sydney, Australia is your home.
04:50What's that like trying to work across borders at the moment and trying to get projects done at the same
04:56time in different jurisdictions?
04:58Oh, look, the reality is we've got projects in North America.
05:00We've got projects in Europe.
05:02We've got projects in Australia.
05:03Yes, my family is in Sydney and I get back for the weekends, but, you know, when people ask where
05:08I live, my typical answer is on a plane.
Comments

Recommended