Liquid Compute, a New York-based startup building a marketplace platform for trading AI infrastructure, has raised $15 million in seed funding co-led by FirstMark and Chemistry, the company confirmed on September 16, 2026. The round also drew participation from K8 Capital, Night Capital, TrueBridge, Brainchild Holdings, UFO Holdings and investor Dmitry Balyasny, reflecting broad-based investor interest in a company targeting one of the most consequential bottlenecks in the current AI economy: the allocation and availability of computing infrastructure needed to train and run advanced AI models.
The scarcity and cost of AI computing infrastructure, particularly the specialised graphics processing units that underpin most large-scale AI model training and inference, has remained a defining constraint on the broader AI industry throughout 2026. Demand for this infrastructure has consistently outpaced supply, creating a market environment in which access to sufficient, reliably available compute capacity has become almost as strategically important to AI companies as the underlying models and algorithms they are developing. That dynamic has given rise to a wave of startups attempting to build more efficient markets and matching mechanisms around AI infrastructure availability, of which Liquid Compute represents a notable new entrant.
Liquid Compute's marketplace model aims to address a structural inefficiency in how AI infrastructure is currently allocated: computing capacity often sits idle at some providers while other companies face acute shortages, with limited efficient mechanisms for capacity to flow to where it is most needed on a dynamic, market-driven basis. By building infrastructure to facilitate more liquid trading of compute capacity — whether through spot availability, longer-term reservations, or more flexible capacity-sharing arrangements between infrastructure providers and AI companies — the company is positioning itself as a piece of financial and operational market infrastructure for an asset class, computing capacity, that has taken on outsised economic importance without a correspondingly mature secondary market to match it.
The involvement of FirstMark and Chemistry as co-lead investors brings considerable venture experience in both enterprise infrastructure and marketplace business models to Liquid Compute's cap table, while the participation of investors including Dmitry Balyasny, a prominent figure in quantitative and systematic trading, hints at the company's ambition to bring genuinely market-native mechanisms — the kind of efficient price discovery and liquidity provision familiar from financial trading markets — to a category of infrastructure that has historically been allocated through comparatively blunt, contract-based mechanisms between hyperscale cloud providers and their largest customers.
The broader AI infrastructure marketplace category has attracted growing venture attention over the past two years as the scale of global AI compute spending has continued to climb, with hyperscale cloud providers, specialised GPU cloud operators and increasingly, enterprises building their own dedicated AI infrastructure, all competing for finite manufacturing capacity from a small number of chip suppliers. That supply constraint has created genuine economic incentive for more efficient allocation mechanisms, since even modest efficiency improvements in how existing compute capacity is utilised can translate into meaningful cost savings for AI companies operating with famously thin margins on compute-intensive workloads.
For Liquid Compute, the central challenge ahead will be building sufficient trust and participation from both compute-capacity providers and AI companies seeking access to make its marketplace genuinely liquid — a chicken-and-egg problem common to marketplace businesses, where the platform's value to any single participant depends heavily on how many other participants are already actively trading on it. Overcoming that early liquidity challenge, likely through targeted partnerships with early infrastructure providers and design-partner AI companies willing to route meaningful compute demand through the platform, will determine whether Liquid Compute's seed round marks the beginning of a genuinely new category of AI infrastructure market-making or one of several ambitious attempts to solve a problem that may ultimately require broader industry standardisation to resolve at scale.
The timing of Liquid Compute's raise also reflects broader anxieties within the AI industry about compute concentration risk, as a small number of hyperscale cloud providers and specialised GPU cloud operators have come to control an outsized share of available AI training and inference capacity globally. That concentration has raised concerns among AI companies, investors and policymakers alike about single points of failure, pricing power imbalances, and the strategic vulnerability smaller AI companies face when their access to computing infrastructure depends heavily on maintaining favourable relationships with a limited number of large providers. A more liquid, transparent marketplace for AI infrastructure, of the kind Liquid Compute is attempting to build, could meaningfully reduce these structural risks if it achieves sufficient scale and participation across the broader AI infrastructure ecosystem.
Whether marketplace dynamics can genuinely take hold in a market as technically complex and capital-intensive as AI infrastructure remains an open question that will likely take several years to resolve, given the highly specialised, often bespoke nature of large-scale AI compute deployments compared with more standardised, fungible assets that have historically supported successful trading marketplaces in other industries.
Eclipse and other infrastructure-focused investors have increasingly argued that the physical and financial layers underpinning the AI boom — power, chips, and now compute-capacity trading — represent some of the most durable, if less glamorous, investment opportunities within the broader AI theme, since demand for these foundational inputs tends to persist regardless of which specific AI applications or model architectures ultimately win out in the more visible, faster-moving application layer.
For enterprises caught in the middle of the current compute crunch, even modest improvements in capacity utilisation and access flexibility can translate into meaningful cost savings and faster model-development timelines, giving Liquid Compute's marketplace thesis genuine near-term commercial relevance regardless of how the longer-term structural question of AI infrastructure ownership and concentration ultimately resolves across the broader industry.
The company's New York base also positions it advantageously to draw on the city's deep pool of quantitative-finance and market-microstructure talent, a resource that may prove just as important to building a genuinely liquid AI-compute marketplace as more conventional infrastructure or AI engineering expertise, given how closely the challenges of building efficient markets for a scarce, heterogeneous resource mirror problems long studied within financial-market design.
