Allbirds (Nasdaq: BIRD), the footwear maker that listed at a roughly $2 billion valuation in November 2021, has disclosed plans to pivot into neocloud GPU-as-a-service, Data Center Dynamics reported. The move would retool a consumer apparel business into an AI infrastructure operator.

What does the pivot entail?

Allbirds would acquire or repurpose wholesale data-center capacity and lease GPU clusters to AI labs, model developers, and enterprise inference customers. Neocloud providers sell hourly or contract compute on top of colocation space rather than building first-party facilities.

Customers pick instances by GPU count and memory footprint. Rates are set by chip type and contract length, with H100 spot pricing ranging from roughly $2 to $4 per GPU-hour on long-term deals in major US regions.

The hardware in question is almost certainly NVIDIA H100 or H200 silicon. These two SKUs account for the bulk of current neocloud inventory. AMD Instinct MI300X and the upcoming Blackwell GB200 platform round out the second tier. NVIDIA holds an estimated 80–90% share of deployed AI training capacity.

What does Allbirds bring?

Founded in 2016 by Tim Brown and Joey Zwillinger, Allbirds built its brand around wool-based running shoes and a direct-to-consumer model. The company priced its November 2021 IPO at $15 a share, peaked above $4 billion in early trading, and has since seen quarterly revenue decline through 2023 and 2024. The stock trades more than 90% below its post-IPO high.

That trajectory is the likely motivation. A pivot to a high-growth segment offers a path off a maturing footwear category with limited operating leverage and persistent gross-margin pressure.

Who is already in the market?

The neocloud segment is dominated by specialists rather than hyperscalers. CoreWeave leads by deployed capacity — multiple reports place its late-2024 fleet above 250,000 GPUs and its valuation near $19 billion. Lambda Labs, Crusoe Energy Systems, Vast.ai, and RunPod fill out the top tier, each running four- to six-figure H100 inventories. Customers span OpenAI, Anthropic, Meta, and a long tail of enterprise AI teams.

Crusoe's history is instructive: it pivoted from Bitcoin mining to AI compute, and now operates one of the largest H200 fleets outside CoreWeave. Lambda runs both on-demand and reserved instances, with dedicated clusters for Anthropic and other frontier labs. Each operator sells on utilization, region, and SLAs ranging from 99.5% to multi-nine uptime guarantees.

What stands in the way?

Three obstacles separate announcement from execution.

• GPU allocation. NVIDIA H100 and H200 supply has been constrained since 2023. Operators hold multi-quarter allocation agreements with NVIDIA and its board partners, including Supermicro, Dell, and HPE.
• Customer credibility. Enterprise AI buyers vet operators on uptime SLAs, RDMA fabric expertise, and reference deployments. A footwear brand starts with none of that history.
• Capital intensity. A single 1,024-GPU H100 cluster carries a price tag in the low hundreds of millions of dollars once servers, InfiniBand fabric, and storage are added to bare chip costs.

What happens next?

The Data Center Dynamics report does not specify a launch date, capital structure, or initial fleet size. Investors should watch for an 8-K filing or company press release detailing the transaction framework, board approvals, and any GPU purchase orders. Until those land, the neocloud pivot is a stated direction rather than an operating business.