Q2 2026 Call for Founders

Each quarter, our investment team goes deep on the areas where we think the next generation of B2B companies will be built. These are the spaces we're actively researching, debating, and looking to fund. If you're building here, or thinking about it, we want to meet you.

01. AI-Native Health Insurance Brokerage

Pulse of the Market

Employer healthcare costs in 2026 are expected to see their largest increase in 15 years, with employer sponsored health insurance spending projected to surpass $1.7 trillion. Insurance distribution is roughly 20 years behind other industries in technology adoption. Despite recent activity in AI native commercial brokerages (Harper, Gyde, Casey) and benefits platforms (Angle Health, Nava Benefits), the core brokerage function- plan selection, carrier negotiation, renewal strategy, compliance, remains largely human driven.

We’d love to meet with founders who are thinking about:

• SMB health insurance brokerage: An AI-native broker who is fiduciarily aligned with the employer, rather than the carrier, and can model self funded scenarios in real time, surface claims data, handle TPA selection and stop loss placement is the distribution unlock the market is missing.
• Self funded/ alternative funding: SMBs are increasingly embracing captives, cash prices, and AI to bypass traditional fully insured plans. A brokerage that can model and place these structures with AI would be differentiated.
• Benefits navigation and advocacy: 95% of employers are taking steps to improve open enrollment, deploying AI-driven recommendation engines to help employees choose appropriate coverage. A brokerage that owns the employee relationship (not just the employer) could build deep retention.

02. The AI Compute Migration: From Cloud to On-Prem and Edge

The cloud giants spent two decades convincing enterprises to move everything off-prem. AI is reversing that for a significant share of compute. Data sovereignty laws, classified workloads, latency requirements, regulatory constraints, and the simple math that owning GPUs is cheaper than renting them after 18 months are all pushing AI workloads back on-premises. But enterprises don't want to go back to managing their own infrastructure- they want the cloud experience running on hardware they control. Meanwhile, many inference workloads don't need a data center at all: a $2,000 edge device running quantized open source models can serve inference that would cost $50K+ annually in cloud API calls. The enterprise AI infrastructure market is projected at $300B+ by 2030, and the companies that own the on-prem and edge layers capture the massive middle ground between public cloud and bare metal DIY.

We'd love to meet with founders who are thinking about:

• Distributed edge inference as a service: Many inference workloads don't need a data center at all. A $2,000 edge device running open source models can serve inference that would cost $50K+ annually in cloud API calls. The opportunity is deploying and managing a distributed fleet of inference nodes at customer locations, handling all the operational complexity, and pricing it as a simple per inference fee. CoreWeave for on-prem edge.
• The global inference network: Thousands of inference nodes at the edge of every network, dynamically routing requests to the nearest capable node, running optimized models across mixed hardware. Every autonomous vehicle, industrial robot, smart building, and military installation needs low latency inference that centralized data centers can't serve.‍
• The managed on-prem AI platform: A turnkey platform that gives enterprises the cloud experience on hardware sitting in their facility. Elastic, managed, API driven, always updated. The platform provider sells, deploys, and manages the full stack so the enterprise just consumes AI compute the same way they would from the cloud

03. The CFO Stack for the AI Era

Pulse of the Market

AI is creating an entirely new cost structure that enterprise finance teams have no tooling to manage. Compute is now a major variable OpEx line that fluctuates with workload, pricing model, and vendor. AI agents are performing work previously billed as labor, with no payroll or tax framework to account for it. Enterprises are acquiring GPU infrastructure with no standardized approach to depreciation or valuation. And model failures and agent errors are creating real liability exposure with no viable insurance products. This is where cloud cost management was in 2012 before Apptio and CloudHealth built what became a $4B+ category. 

We’d love to meet with founders who are thinking about:

• Compute cost forecasting for finance: FP&A tools built specifically for AI compute as variable OpEx, including token attribution, workload level cost allocation, and forecast modeling for usage based and outcome based pricing. The cloud FinOps category doesn’t cover AI inference, and none of it touches on-prem chip assets.
• On-prem AI asset management: As enterprises move inference on-prem, CFOs are inheriting GPU rack assets with no tooling for depreciation scheduling, capital allocation, insurance valuation, or collateral assessment. 
• AI labor accounting and attribution: Enterprises running mixed human and AI teams have no way to track, report, or optimize cost attribution between them. As AI takes on more tasks previously performed by people, finance teams need a way to account for that shift, and regulatory and tax frameworks around AI labor are not far behind.
• AI enterprise risk and insurance: Model failure, agent errors, and hallucination driven decisions are real enterprise liability categories with no viable insurance products or underwriting frameworks. This is where cyber insurance was in the late 1990s- the risk existed before the pricing models did.

04. AI-Designed Molecules Unlock Venture-Scale Industrial Biology

AI has fundamentally changed the economics of molecular discovery. What used to take years of directed evolution and tens of millions of dollars can now be accomplished in weeks using generative protein models, autonomous enzyme engineering platforms, and cell-free screening systems. Researchers have demonstrated 90-fold improvements in enzyme specificity in a single design cycle. New computational methods can design enzymes for industrial chemical reactions from scratch in a one-shot process, producing biocatalysts stable enough for harsh manufacturing environments.

This compression creates a new category of venture backable company. A pre seed check can now fund the critical gap between an AI designed molecule with a provisional patent and validated industrial process economics- unit economics, a scalable production pathway, and a patent strategy that unlocks institutional capital. The $6 trillion chemical industry still runs on petroleum-derived processes that are energy intensive and increasingly regulated. Enzyme-based alternatives can deliver 50-70% energy savings and 40-60% water reduction. But cost parity with petrochemicals remains the central challenge, and the companies that win will close the gap between "works in the lab" and "works at industrial scale and price."

We'd love to meet with founders who are thinking about:

• Process economics validation: Taking an AI designed enzyme or molecule from provisional patent to demonstrated industrial viability- proving yield, titer, rate, and cost of goods at the scale that makes the next round of funding obvious. The 12-18 month window between provisional filing and full process disclosure is our target area.
• Enzyme-as-a-Process for specific industrial replacements: Selling the validated enzyme and the production process together into high-value verticals where cost parity with petroleum-derived already exists or is within reach- specialty chemicals, pharmaceutical intermediates, food ingredients, textile processing, and bio-based polymers. 
• AI-guided retrobiosynthesis optimized for economics: Platforms that design biosynthetic pathways not just for chemical feasibility but for industrial cost structure from the start, incorporating feedstock cost, energy consumption, yield targets, and capital expenditure into the design loop.
• Patent strategy infrastructure for AI discovered molecules: Tools and services that help founders build defensible IP estates around AI generated molecular innovations, managing the complexity of human inventorship documentation, continuation-in-part strategies as process data accumulates, and freedom-to-operate analysis across overlapping AI generated chemical spaces.
• Scale up and fermentation infrastructure: Cell free biomanufacturing platforms, continuous fermentation systems, and contract development organizations purpose built for bridging the gap between computational enzyme design and pilot scale production. The bottleneck has shifted from discovery to scale-up.
• Enzymatic recycling and circular economy applications: Enzymes that depolymerize plastics, textiles, and industrial waste back into virgin quality feedstocks. Regulatory mandates around circular economy and extended producer responsibility are creating real market pull, and enzymatic approaches are proving commercially viable at scale for the first time.

05. The Skilled Labor Gap 

Pulse of the Market

The US is facing a structural labor crisis that no amount of recruiting can solve. The country is losing 400K+ skilled tradespeople per year to retirement with no replacement pipeline, while manufacturing carries 500,000+ unfilled positions. At the same time, AI is displacing white-collar workers who need new career paths. The conventional response—“train more workers”—misses the point. The real opportunity is using AI to fundamentally redefine what it means to do these jobs, decouple decades of institutional knowledge from the physical act of doing the work, and build new pathways for people entering the trades from non-traditional backgrounds.

We’d love to meet with founders who are thinking about:

• AI augmented skilled trades platforms: AR and spatial computing tools that overlay expert guidance onto physical work environments in real time, turning a two-year apprentice into someone who can execute complex jobs with AI assistance. Platforms that decouple diagnostic and technical knowledge from physical delivery so less experienced technicians can handle work that previously required a 20-year veteran.
• Institutional Knowledge Capture: Platforms that record and structure skilled trade workflows from observation, video, sensor data, tool usage patterns, into reusable procedural knowledge. This is fundamentally different from white-collar knowledge capture because the knowledge is embodied, spatial, and contextual. The most defensible version builds a live skills graph: every tap, voice query, and completed task becomes a timestamped record that trains future models.
• The manufacturing workforce platform: AI can automate roughly 80% of a manufacturing operator's cognitive work: diagnostics, sequencing, quality checks, documentation. What remains is physical and situational: loading parts, handling exceptions, managing setup. If you cleanly decouple those two layers, the job description changes entirely. It no longer requires years of accumulated knowledge, just someone who can execute physical tasks while AI handles the decision-making overhead. That dramatically expands the eligible labor pool and compresses ramp time from months to weeks.
• AI native skilled trades services firms: Full stack service businesses that use AI to compress the expertise required for HVAC, plumbing, electrical, and other trades. AI handles the diagnosis, generates the repair plan, walks the technician through execution, and manages quality assurance, enabling less experienced technicians to deliver expert level service.
• The white-collar to trades transition pipeline: Vocational retraining platforms that combine AI accelerated learning with structured physical practice to compress traditional 2-4 year training into 6-18 months, with credentialing and placement built in, purpose-built for career changers rather than 18 year olds entering the workforce for the first time.

06. Insurance in the AI Era

Pulse of the Market

Insurance is a $6T+ global industry built on actuarial tables derived from decades of human-operated physical systems. AI is rewriting the risk profile of virtually every insurable physical asset, but the data to price this new risk doesn’t exist.  Liability frameworks for human/AI shared decision making haven’t been written, and incumbent carriers can’t make the transition because their organizational structure, regulatory framework, and capital model are built around the old paradigm.  An autonomous warehouse has a completely different fire risk profile than a human-operated one. When an AI system miscalculates a structural load or misjudges a drone flight path, existing insurance products, regulatory frameworks, and legal precedents have no answer for who is responsible. 

We’d love to meet with founders who are thinking about:

• Liability allocation for human/AI shared outcomes: Engines that determine responsibility when human operators and AI systems jointly control physical outcomes. As AI embeds into construction, logistics, manufacturing, and healthcare, the interaction between human judgment and AI recommendation creates a new category of risk that existing frameworks can’t resolve.
• Underwriting infrastructure for autonomous systems: As autonomous vehicles, drones, and robotic systems scale commercially, insurers need to underwrite the behavior of the AI operator, not just the asset. That requires entirely new rating variables and the underwriting models for this don't exist and neither does the data pipeline to feed them.
• Black box recording and audit infrastructure: When an AI-assisted physical operation goes wrong and ends up in litigation, someone needs to reconstruct what the AI recommended, what the human decided, and why. Recording infrastructure for AI assisted operations that creates defensible records for courts and regulators.
• Risk data platforms for AI operated physical assets: Platforms that instrument AI-operated facilities, factories, fleets, and buildings and generate the datasets insurers need to price new categories of coverage create the foundation the entire market depends on.
• The AI native insurance carrier: A vertically integrated carrier built from scratch around continuous physical world data collection, dynamic risk pricing, automated loss prevention, and instant sensor-based claims settlement. Applied to commercial property, industrial equipment, fleet, and infrastructure (categories worth trillions in annual premiums) the outcome is fundamentally different from consumer focused predecessors.

07. Software Abstraction Layer for Edge Hardware

When chip vendors rushed to bring intelligence on device starting around 2015, each took a different architectural approach with no shared standard to build toward. Every NPU vendor built proprietary toolchains, and none are compatible. A developer who wants an AI feature to run across Qualcomm, Apple, Intel, and MediaTek hardware isn't writing one codebase. They're maintaining several, one per hardware target, each requiring its own optimization every time something changes. On top of that, no neutral runtime exists to manage what happens when multiple AI workloads compete for the same chip on the same device.  Vendor lock-in is a massive headwind on edge AI advancement, and every platform shift tells us the company that removes it captures outsized value.

We’d love to meet with founders who are thinking about:

• Edge runtime and compiler abstraction: A neutral runtime and compiler layer that lets developers write once and execute optimally across any edge silicon. Today's fragmentation means separate codebases, separate optimization passes, and separate maintenance burdens for every hardware target. The company that builds the layer sitting between AI applications and the diverse NPU hardware underneath them becomes  the infrastructure every edge AI application runs through.
• Intelligent inference routing: As the alternative silicon ecosystem matures, enterprises are running inference across a mix of NVIDIA, AMD, Intel, Qualcomm, and custom ASICs. A system that profiles each workload's actual requirements and matches it to the most cost effective silicon automatically, accumulating performance data across every workload and chip combination, makes the right hardware decision at the workload level rather than the fleet level.
• Unified operational planes: Enterprises managing mixed hardware fleets have no single view into what's running where, what it costs, and where workloads are overpaying for silicon they don't need. A management layer that treats edge and on-prem hardware as a single resource pool, with cost attribution that surfaces exactly where the performance premium isn't worth the price premium, is the operational backbone for a multi-vendor inference world.

08. The New Maintenance Economy

Pulse of the Market

AI is transforming every dimension of how physical assets are maintained,  from what breaks and why, to who fixes it and how the work gets paid for. As AI becomes the operator of physical systems, failure modes change entirely: software/hardware interaction failures, sensor drift, model degradation in physical environments. Predictive maintenance has underdeliveried for a decade because “your machine will probably fail in 30 days” isn’t actionable. The real shift is prescriptive, going from predicting failure to orchestrating the intervention. Meanwhile, the US has a $2.6T infrastructure maintenance backlog, military depots can’t hire enough technicians, and $500B+ in hazardous government work isn’t getting done.

We’d love to meet with founders who are thinking about:

• Maintenance for AI-Controlled Physical Systems: Increasingly the failure mode will be the AI itself, whether that's model drift causing mechanical stress, sensor calibration decay, or software updates interacting with physical systems in ways no one anticipated. The diagnostics require a fundamentally different skillset than maintaining human-operated equipment, and the service contracts look nothing like traditional maintenance agreements.
• Prescriptive maintenance that closes the loop: Systems that go beyond prediction to generate the specific intervention, order the parts, schedule the technician, and verify the repair. 
• Uptime-as-a-service business models: Platforms where the AI takes financial responsibility for equipment availability, absorbing the cost of downtime and earning a premium for guaranteed performance. This shifts maintenance from selling software to owning the outcome, the physical world equivalent of outcome first service firms applied to industrial assets.
• Autonomous infrastructure maintenance: AI controlled robotic systems for the most repetitive, dangerous, and labor constrained maintenance tasks: pipe inspection and repair, road surface maintenance, bridge inspection, power line monitoring. Not one off robots, but a networked fleet that municipalities subscribe to like a utility.
• Autonomous and AI-augmented military depot maintenance: Military maintenance often exceeds the original acquisition cost of a platform over its lifetime: an F-35 costs $80M to buy and $40K+ per flight hour to maintain. Robotics for repetitive inspection and repair, AI for diagnostics and planning, and AR for amplifying remaining human technicians. 
• Embodied AI for contaminated and hazardous environments: Nuclear decommissioning, chemical weapon disposal, and deep infrastructure inspection in environments too dangerous for humans. The government has identified over $500B in deferred nuclear cleanup alone. Autonomous systems that navigate unstructured spaces, manipulate objects, and make real-time decisions about hazardous materials unlock work that is currently impossible.

09. Cybersecurity for AI-Controlled Physical Systems

Pulse of the Market

As previously air-gapped physical systems get connected to enable AI control, the attack surface is expanding rapidly. The threat model is categorically different from traditional industrial cybersecurity. You're not just protecting the network. You're protecting against adversarial manipulation of the AI itself: feeding it bad sensor data, corrupting its decision model, exploiting the gap between what the AI thinks is happening and physical reality. When an attacker compromises a traditional IT system, you lose data. When an attacker compromises an AI controlling a factory, a power grid, or a fleet, the consequences are physical. The global cybersecurity market is already $200B+ and growing. The physical world AI security layer could be larger because the consequences of failure are existential and governments will mandate it

We’d love to meet with founders who are thinking about:

• Adversarial robustness testing for physical infrastructure AI: Testing platforms that systematically probe AI systems controlling critical infrastructure for vulnerabilities.
• Physical/ digital integrity verification: Systems that cross check AI decisions against independent physical sensors to detect when the AI’s model of reality has been compromised.  An independent verification layer that catches the moment an AI controlled system starts acting on corrupted data before consequences become physical.
• Incident response and forensics for AI physical breaches: When a breach occurs in an AI controlled physical system, responders need to reconstruct what the AI “believed” versus what was actually happening physically. Purpose built forensics tools that untangle the interaction between cyber compromise and physical consequence.
• Physics-Aware Cyber Defense: Protecting AI-controlled physical systems requires understanding the physics of the system, not just the network. Security infrastructure that combines cyber defense with physical system modeling, detecting anomalies that only make sense when you understand how a power grid, factory, or autonomous fleet actually operates in the physical world.

10. The Continued Evolution of Real World Models

The physical world is drowning in data it can't access. Governments sit on massive amounts of physical world data locked in analog formats: paper permits, hand drawn utility maps, physical land records. Millions of miles of underground pipes, cables, and conduits were never digitally mapped. Roughly $500T in physical assets on Earth have no real time digital representation of their condition, utilization, or value. And the environments that matter most for the next wave of AI models have never been sensed at scale, not because they're exotic, but because conventional optical sensing simply doesn't work there. Underground infrastructure, subsea systems, and interior industrial processes require acoustic, pressure, chemical, and thermal sensing modalities that are only now becoming viable. As physical space becomes multi-dimensional, with drone corridors above the surface, utility corridors below it, and autonomous vehicle routing zones across it, the absence of coherent physical world data infrastructure becomes the binding constraint on everything from infrastructure investment to insurance pricing to national security.

• The Analog-to-Digital Conversion of Government Infrastructure: Permits, deeds, environmental assessments, inspection histories, hand drawn utility maps, all locked in analog formats across thousands of agencies. 
• Subsurface sensing and mapping: Platforms combining ground-penetrating radar, electromagnetic detection, and AI to build 3D maps of underground infrastructure that was never digitally recorded. 
• The multi-dimensional land registry: Physical space is being carved into new dimensions: drone corridors, solar rights, air rights, autonomous vehicle routing zones. A spatial rights platform for mapping, registering, pricing, and transacting across every dimension of physical space.
• Water data infrastructure: Distributed sensing networks providing real time water quality and composition data at the point of delivery. Leak detection combining acoustic sensors, pressure analytics, and AI. Water accounting and trading infrastructure for regions implementing water markets.
• Interior industrial processes: What happens inside a blast furnace or casting mold during operation is almost entirely unobserved. Temperatures exceed what optical sensors tolerate. New sensing modalities like acoustic emission and microwave are beginning to make interior process observation possible for the first time- a model trained on direct interior process data represents a step change over anything trained on external proxies.
• Expert in the loop labeling platforms: Labeling infrastructure designed for physical-world domains where annotation requires trade or engineering expertise, not just visual pattern recognition- welding inspectors identifying defect types, structural engineers classifying load stress patterns, agronomists interpreting soil and crop conditions.

11. Economic Infrastructure for Agents

Pulse of the Market

When agents become the customers, the entire commercial infrastructure built around human psychology, attention, and judgment becomes irrelevant or actively counterproductive. Agents don’t respond to brand, narrative, or relationship. They query, evaluate, transact, and move on. Every layer of the commercial stack gets rebuilt around what agents actually optimize for: verified performance, price, availability, and compatibility. The infrastructure required to support agent to agent commerce, such as identity, credentialing, trust verification, real time performance registries, and auction based procurement, doesn’t exist at commercial scale. This is closer to programmatic advertising infrastructure than to traditional procurement, and programmatic advertising created an entirely new industry when human ad selection was replaced by algorithmic selection.

Examples of potential shifts we might see:

• Agent Identity and cross-organizational trust infrastructure: An agent transacting on behalf of a corporation needs a verifiable identity that counterparties can authenticate, a capability scope defining what it is authorized to do, and a decision log- the equivalent of SSL certificates for agent-to-agent commerce. The harder problem is cross-organizational: your agent interacting with your customer's agent, your supplier's agent, your bank's agent, each operating under different policies and trust models with no shared coordination layer. The infrastructure that solves individual agent identity is the same infrastructure that enables cross-organizational trust, and whoever builds the standard owns the authentication layer for the entire agent economy.
• Verified performance registries: If agents select based on verified quantitative performance data, there’s an opportunity for a registry that accumulates continuously verified performance records across vendors (uptime, accuracy, latency, error rates, compliance) as a dataset of ground truth on which vendors actually perform as claimed.
• Anti collusion infrastructure for agent markets: When thousands of pricing, procurement, and trading agents operate in the same markets, they naturally converge on equilibria that may constitute illegal collusion. Market surveillance that detects emergent cooperative behavior, proves or disproves intent, and provides the compliance evidence regulators will demand. "We didn't tell it to collude" won't be a defense.
• Agent negotiation infrastructure: When agents negotiate on behalf of both parties in every transaction, markets emerge where fixed pricing previously dominated. Protocols for agent-to-agent bargaining, market-clearing mechanisms, fairness constraints, deadlock resolution, and deal verification. Every transaction where both sides have agents becomes a micro-market that needs market infrastructure.
• The data brokerage layer for agents: The entire $250B+ data industry restructured for agent consumption, delivered programmatically, in real time, with clear provenance, usage rights, and consumption-based pricing. Data providers list assets, agents discover and evaluate sources, usage rights are programmatically enforced, and pricing is based on actual agent consumption rather than seat licenses.

12. The Waste Stream Is an Unpriced Data and Resource Problem

Pulse of the Market

Industrial and commercial waste is one of the last major physical flows with almost zero data infrastructure. Companies pay to have waste hauled away and that’s where visibility ends. But waste streams contain valuable signals, for example- a change in a factory’s waste composition can indicate process drift, quality problems, or regulatory exposure before any other metric catches it. On the recovery side, the economics of recycling and materials recovery are entirely dependent on knowing exactly what’s in the stream, which today requires manual sorting or expensive lab analysis. AI-powered waste characterization turns a cost center into an intelligence layer.

We’d love to meet with founders who are thinking about:

• Real time waste stream characterization: Computer vision and spectroscopic sensors at the point of generation, turning waste bins and dumpsters into data collection points that capture composition, volume, and frequency in real time.

• Process intelligence from waste composition: Platforms that use waste composition data as a leading indicator for manufacturing quality, efficiency, and compliance. Your waste tells you what’s going wrong before your production metrics do.

• Regulatory compliance for waste generators: Platforms that automatically track, classify, and report hazardous and regulated waste, an area where penalties are severe and record keeping is still largely manual.

• Waste stream matching and material arbitrage: A platform that maps what industrial operators are actually discarding and matches it to processors who can extract value, taking a spread on material arbitrage without requiring chemistry IP or capital-intensive processing infrastructure.

• Co-location recovery modules: Semiconductor fab waste contains high purity source materials at concentrations above any mining operation. Coal ash contains germanium and gallium being disposed of as waste. The recovery infrastructure doesn't exist at the site but there is opportunity for a modular recovery system.

13. The Future of American Manufacturing 

Rebuilding US manufacturing capacity is a national priority backed by hundreds of billions in government incentives, but the infrastructure to actually make it happen barely exists. There are 600,000+ manufacturing facilities in the US, and virtually all of them run on tacit, embodied knowledge held by floor managers who are retiring. Most reshoring efforts are failing because companies are trying to replicate supply chains that took decades to develop in Asia. Individual factories operate as isolated islands, each solving the same problems independently. And the economics of production still assume high volume runs that exclude a massive long tail of products. 

We'd love to meet with founders who are thinking about:

• The manufacturing intelligence layer: Systems that continuously absorb operational intelligence, from sensors, operator behavior, and process outcomes, and make it available to every shift, every facility, every new hire. Machine 7 runs hot on Thursdays, Supplier B's aluminum needs different feed rates, humidity above 60% doubles paint line rejections. This tacit knowledge has never been captured digitally, and the people who carry it are retiring across every factory in the country.
• The factory capital markets platform: Connecting real time factory performance data to capital providers. A machine on the factory floor generates continuous data about its own productivity, uptime, and output quality, none of which flows to the lender underwriting a $5M equipment loan based on tax returns. 
• Manufacturing knowledge pooling: A network where manufacturers share operational learnings without sharing proprietary data- the same way hospitals contribute to medical research without exposing patient records. A shop in Ohio cuts scrap rate on a specific alloy by 30%, and that knowledge automatically benefits 500 other shops running the same material. Particularly powerful in fragmented sectors like machining, metal fabrication, plastics, and food processing where tens of thousands of small shops can't afford their own data teams.
• On demand and micro manufacturing infrastructure: Platforms that combine AI driven quoting, process planning, and quality assurance to make production runs of 1-100 units economically viable. Distributed manufacturing networks matching micro-production jobs with nearby shops that have available capacity. Digital inventory platforms that replace warehousing 50,000 SKUs of spare parts with manufacturing-on-demand from stored digital designs. 

14. When Every Physical Asset Gets a Financial Identity

Historically, only large physical assets had enough value to justify the overhead of individual financial tracking: a building, an aircraft, a ship. AI and IoT are driving the cost of monitoring individual physical assets toward zero, which means you can now create a financial identity for a single machine tool, a shipping container, a construction crane, or even a pallet of goods. When every physical asset can report its own location, condition, utilization, and performance in real time, you unlock entirely new financial products: asset level financing, usage based insurance, real time collateral valuation, and liquid secondary markets for physical assets that were previously illiquid.

We'd love to meet with founders who are thinking about:

• Persistent asset identity platforms: Giving individual machines, vehicles, or equipment a digital financial identity tied to real time physical data: location, condition, utilization history, maintenance records, and performance metrics that follow the asset throughout its entire lifecycle
• Dynamic collateral valuation: Systems that use IoT data to update the value of physical assets in real time for lenders. A well maintained machine in active use is worth more than an idle one in a warehouse, but today's lending infrastructure has no way to reflect that. Real time collateral valuation transforms how physical assets are financed.
• Usage based financing platforms: Financing where payments on physical equipment are tied to actual utilization data rather than fixed schedules, aligning cost with value for the operator. When the machine is producing, you pay more. When it's idle, you pay less. This unlocks equipment access for companies that can't justify fixed payments on assets with variable usage.
• Liquid secondary markets for physical assets: Transparent condition and performance data reduces the information asymmetry that makes used equipment markets inefficient. Secondary markets where buyers can see verified utilization history, maintenance records, and real time condition data transform illiquid physical asset categories into something approaching a functioning market.

‍

We’re investing in 80+ companies this year through our AI Venture Studio, accelerator, and pre-seed fund. If you are spending time in these spaces, we’d be excited to meet and explore whether there is an opportunity to work together.

Forum Ventures® is the leading early-stage fund, accelerator and venture studio for B2B startups. Founded in 2014 and based in New York, San Francisco, and Toronto, we’re on a mission to make the B2B startup journey easier, more accessible, and successful for early-stage founders. We invest in founders at the earliest stages and work together to launch, build and scale their businesses. To date, we have made 550+ pre-seed and seed investments globally across industries like fintech, healthcare, applied AI, AI infrastructure, vertical AI, supply chain, and more.