TLDR
- After selling its footwear business, Allbirds renamed itself Smartbird and is building a new business focused on managed AI infrastructure.
- Rather than build GPU capacity first and find customers later, Smartbird plans to design bespoke single-tenant systems around individual enterprises’ workloads and manage the infrastructure for them, according to CEO Nadia Carlsten
- But Smartbird faces established cloud and AI infrastructure providers that are better capitalized.
What do shoes have to do with artificial intelligence?
It’s a non sequitur that made perfect sense to the management of Allbirds, which sold its sneaker business in June after posting material losses and shuttering stores, according to filings with the U.S. Securities and Exchange Commission. With assets sold, the publicly traded company pivoted to a new business: AI.
The company began coalescing its business model around AI infrastructure after speaking with Nadia Carlsten, a former CEO of an AI infrastructure company called DCAI. She was named CEO of Smartbird in June.
“Smartbird wanted to do something in in AI, and I think for me more specifically, I really wanted to do something in AI infrastructure, but also do it in a way that was different than all of the other AI infrastructure providers,” Carlsten said in an interview with The AI Innovator.
Smartbird is betting there is room between the hyperscale cloud providers and do-it-yourself computing clusters for a new type of enterprise AI infrastructure company.
Rather than build large pools of GPUs and then rent the capacity to customers on demand, Smartbird plans to start with the customer’s AI workload and construct infrastructure around it.
“We start the other way around,” Carlsten said. “We start with the customer, their workload. Based on that workload, we estimate what it is that they need, and we build that for them.”
The strategy puts Smartbird into a crowded and capital-intensive market that includes Amazon Web Services, Microsoft Azure and Google Cloud, as well as AI infrastructure specialists such as CoreWeave, Crusoe, Lambda and Nebius.
Carlsten said what makes Smartbird different is that it will primarily build bespoke, single-tenant systems for individual customers that don’t want to maintain the clusters themselves.
Smartbird would design and procure the equipment, assemble and commission the cluster, test and benchmark it and then manage the system for the customer. Customers could locate those systems in their own facilities or in third-party data centers selected by Smartbird.
That means an enterprise could obtain dedicated computing infrastructure without becoming an AI infrastructure operator itself.
“The whole point of why we exist is to let people do their job instead of having to worry about optimizing GPUs, orchestration of containers, and so on,” Carlsten said.
Building from the workload backward
Smartbird’s pitch rests on the premise that enterprise AI infrastructure requirements increasingly differ from those of companies developing frontier models.
Large AI labs need enormous GPU clusters to train models. Enterprises are more likely to be running inference, AI agents, simulations, fine-tuning or combinations of those workloads, Carlsten said.
Smartbird therefore intends to ask customers what they are trying to accomplish before deciding what hardware they need. Its engineers would evaluate requirements involving GPUs and CPUs as well as memory, storage and networking. Security, compliance, cost and performance could also change the configuration.
The resulting system would generally be dedicated to one customer.
That puts Smartbird somewhere between public AI cloud services and an enterprise-owned cluster. Carlsten said the company does not view itself as a replacement for cloud providers. Companies experimenting with AI should often start in the cloud, she said, because pay-as-you-go infrastructure provides flexibility before workloads become predictable.
Smartbird is instead targeting persistent workloads for which an enterprise decides shared cloud infrastructure no longer meets its economics, performance, security, compliance or data-sovereignty requirements.
Smartbird is entering a market with established competitors, although their business models do not precisely match its own. Hyperscale cloud providers primarily offer computing capacity through their cloud platforms, while AI infrastructure specialists provide varying combinations of cloud capacity, dedicated infrastructure and managed services.
Some of those offerings overlap more directly with Smartbird’s proposed model: Lambda, for example, offers dedicated, single-tenant infrastructure built to customer specifications, while CoreWeave offers dedicated clusters that it operates for customers.
Smartbird argues its differentiation will come from starting with each enterprise’s workload instead of starting with prebuilt GPU capacity. Carlsten also said the company will avoid speculative infrastructure construction.
“Everything that we build is backed by a customer contract,” she said. “We’re not going to buy a ton of GPUs just for the sake of buying a ton of GPUs.”
A business still being built
For now, Smartbird remains at an early stage.
Carlsten said the company has several prospective customers in its pipeline but has not announced customers under the new managed-infrastructure strategy. Smartbird is hiring engineers and customer-facing personnel and qualifying data center locations where future clusters could be installed.
She added that Smartbird has roughly $200 million in potential capital resources, combining cash, a convertible-note facility and access to an at-the-market stock offering program.
As of June 30, the company’s balance sheet showed $37.4 million in cash and cash equivalents, according to its latest 10-Q filing with the SEC. A convertible-note facility was expanded to as much as $100 million, of which $8.25 million in principal had been issued as of June 30. The remaining $91.75 million is at the option of the noteholders. Smartbird also has an at-the-market equity program authorizing up to $98.1 million in stock sales. It had raised $15.4 million in net proceeds through that program by June 30.
The company acknowledges other risks. Its latest quarterly filing warns that it is entering an industry where it has limited operating history and faces larger, better-capitalized competitors. Rapid changes in chips, AI models and infrastructure requirements could also reduce the value of equipment it acquires.
Carlsten said Smartbird intends to limit some of that exposure by purchasing infrastructure against contracted demand rather than betting on future customers.
Ultimately, Smartbird wants enterprises to view it as a third choice: not in the camp of renting shared computing capacity nor building and managing everything themselves.
“The goal is to be firmly known as that third option,” Carlsten said.
Author
Get the latest insights about enterprise AI.
Subscribe to our newsletter. Thank you.
