- Lambda is one of the fastest-growing neocloud players in the market
- The company is aiming to deploy 3 GW by 2030 and has already signed leases for 320 MW
- Exec Ken Patchett told Fierce its strategy is to play to its strengths and take a measured approach
It’s not unusual for a startup or disruptor to have lofty goals. But it is certainly less common for such a player to openly admit it can’t achieve the desired outcome alone. So when Lambda’s VP of Data Center Infrastructure Ken Patchett told Fierce the company doesn’t have the time or inclination to squabble with rivals, it caught our attention.
“One GPU, one person, that’s our goal,” Patchett said of Lambda’s vision for the future. But he noted, “There’s not even enough people in it now to actually move as fast as we need to move. So, we don’t have time for that fight club or [to] throw shade or anything else like that. What we have time to do is actually make sure that we’re all working to put technology out there.”
While CoreWeave leads the neocloud space with around 1% of the global cloud market share, Synergy Research Group highlighted Lambda as one of three smaller neocloud players (the others being Nebius and Crusoe) who are “growing extremely rapidly," the firm said.
For its part, Lambda currently leases and has infrastructure deployed in around 20 data centers across the U.S. Patchett told Fierce Lambda is working toward 3 GW of data center capacity by 2030, with “more than 320 MW worth of data center space fully leased, signed and committed” to date and “lots coming on the heels of that.”
This strategy of leasing rather than owning is intentional. Patchett noted that Lambda’s expertise is in running a GPU-based AI platform, not operating a data center. That said, he added it will likely own and operate its own data centers at some point in the future, but it has to build up to that.
“You need to grow and build in such a way that you don’t get so far over your skis that you put $1.5 billion at risk,” he said, referencing Lambda’s recent Series E funding.
“If you think you’re going to do it by yourself, you should probably rethink that because you are going to miss something,” Patchett continued. “You will miss your timelines. You will not have the speed to market that you require. You might build the wrong thing.”
Capacity from chaos
Patchett knows a thing or two about data centers. Back in 1988, he helped build (yes, literally) a data center for Microsoft as an ironworker.
Ten years later he was back at the same facility, this time as its manager. Over the past two decades, he’s also done stints with Amazon, Google, Facebook (now Meta) and Oracle in data center-related roles. Before joining Lambda in May 2023, he was Manager of Construction for AWS.
The executive acknowledged that the current data center landscape overall is “messy.” But Patchett said that a certain level of chaos is part and parcel of building something entirely new. He likened it to the disruption that occurs when a new road is built or an existing one widened.
“When you get on that brand new road and you don’t hear the road noise, it’s great. But to get there it’s disruptive,” he said.
Patchett said Lambda has been very careful to be both fiscally responsible and demand-driven in its rollouts. “We could go faster, we really could. But if you go faster, maybe you run into problems where you’re overcommitted or overextended,” he said of its measured approach.
He added it is also trying to ensure it builds future-proof infrastructure to the extent possible.
“One of the things that we are very conscious about is the changing landscape of hardware and hardware requirements,” Patchett said. “We have to make sure the data centers that we’re preparing are prepared to receive the newest and latest and greatest GPUs that are being created.”
There is a certain degree of flexibility in Lambda’s operating model. As Patchett explained, the company offers private cloud services for hyperscalers and other large customers as well as public cloud services for smaller businesses all the way down to individual developers. Thus, when GPUs roll off a private cloud contract, they have a second useful life on Lambda’s public cloud.
Still, getting it right matters and that can be hard when a business with a three to five year planning horizon intersects with a technology that’s moving as fast as AI is.
“I wake up thinking about the fact that we have to invent new things based on the power and the density [requirements of new GPUs] but what if some of those new things help us not need to be that dense anymore and I’ve already spent a bunch of money building something really, really dense and now I don’t need it,” he explained.
“I have to be intentional, I have to commit and I have to be right. And so that keeps me up a lot.”