Expanse raises $5.3m seed to predict AI workload compute needs
Expanse, an AI infrastructure startup co-founded by four University of Edinburgh engineers, has closed a $5.3 million seed round led by Crane Venture Partners, with PXN Ventures and a group of angel investors also participating. The angels include former DeepMind researchers and veterans of AI infrastructure teams, though Expanse did not name them individually.
The company's platform analyses an AI workload before execution begins and recommends the optimal GPU, CPU, memory and runtime configuration, rather than relying on the manual estimates that engineers currently make before committing cloud or on-premises resources. Expanse says its software installs within a customer's own environment, so code and telemetry never leave the organisation's perimeter, a design choice intended to address data-sovereignty concerns.
The problem being solved
The commercial rationale is straightforward. Industry estimates cited by Expanse suggest roughly 30 per cent of cloud spend is lost to over-allocation of compute. Microsoft Research has separately reported GPU utilisation of around 50 per cent across many internal deep learning workloads. Both figures are consistent with what infrastructure buyers tell analysts: provisioning AI jobs is as much art as science, and the cost of guessing wrong accumulates at scale.
Chief executive Ismaeel Bashir framed the problem directly: "Engineers shouldn't have to predict exactly how much compute their code will need before they press run. The machine should carry the uncertainty, not the person." Bashir developed a multimodal HPC resource prediction system at the Edinburgh Parallel Computing Centre that the company says set a new benchmark for predicting resource requirements across HPC workloads; that research underpins the commercial product.
In one production deployment, Expanse says it identified nearly $8 million of idle compute capacity in a single month. The company did not name the customer or provide an audited figure; the claim is unverified.
Market context and competitive landscape
Expanse is entering a crowded but genuinely under-served segment. Observability and monitoring tools from vendors such as Datadog, Grafana Labs and a range of cloud-native offerings tell operators what happened after a job ran. Workload schedulers such as Kubernetes and SLURM handle job queuing and resource allocation, but depend on resource requests that engineers submit manually. Expanse's stated differentiation is pre-execution prediction rather than post-execution analysis, a position that puts it closer to capacity-planning tools and to emerging AIOps platforms from hyperscalers, though none yet offers workload-level compute forecasting as a standalone product.
Scott Sage, co-founder and partner at Crane Venture Partners, described Expanse as representing "a new and important category in AI infrastructure," arguing that competitive advantage in the AI era will increasingly belong to organisations that use compute intelligently rather than those that simply acquire the most of it. That thesis is plausible given the procurement lead times for new GPU capacity, which can run to several months, and the physical constraints on data-centre power and cooling that are limiting the pace of hardware expansion across the industry.
GPU supply remains tight globally, and hyperscalers have been vocal about the capital expenditure required to expand capacity. For enterprises that cannot easily procure additional H100 or B200 clusters, software that recovers meaningful utilisation from existing infrastructure carries genuine economic appeal.
The seed proceeds will fund engineering headcount growth, product development and commercial expansion across AI infrastructure, quantitative finance, life sciences and high-performance computing. Expanse did not disclose a post-money valuation or a target for its next funding milestone.