Vannevar Labs reached a $1.5B valuation in October 2024.

The company raised a $75M Series B in January 2023 after a $12M Series A in 2021 and $4.5M seed round in 2019. Total disclosed funding is approximately $91.5M.

Vannevar Labs builds AI software that converts fragmented, hard-to-access foreign information into searchable intelligence and operational recommendations for military analysts, planners, investigators, and commanders. The system collects data from roughly 3,000 unique sources across 23 countries and regions in nine languages. Its holdings comprise over one billion structured data points and approximately one petabyte, growing by about one terabyte per week.

The collection layer uses managed attribution and obfuscated proxy infrastructure to gather information in adversarial digital environments without activity visibly originating from a U.S. government network. Self-hosted machine-learning models process raw material through translation, OCR, entity extraction, relationship identification, geospatial enrichment, and semantic indexing. Four AI models run in production, handling roughly six million enrichments and 24,000 retrieval-augmented-generation calls per day.

Analysts use mission applications rather than a generic chatbot. They can select an entity to map its ownership network, draw a polygon on a map to explain unusual activity, or highlight a chart spike to identify its cause. The software displays intermediate steps, links conclusions to source documents or map objects, and provides Accept, Edit, and Undo controls, leaving consequential decisions to the operator.

The portfolio has expanded from Decrypt, an intelligence search application that began as a two-person prototype, to five products and dozens of mission-specific agents. These include planning agents that simulate adversary reactions and compare courses of action, force-protection agents that detect surveillance campaigns against friendly forces, information-operations agents that monitor propaganda and recommend counter-messaging, and geospatial agents that identify cross-border anomalies. Serra, a geospatial product, and a shared RAG platform moved from forward-deployed engineering prototypes to reusable offerings.

The stack runs in government cloud, classified networks, air-gapped systems, and low-connectivity tactical environments. An internal system called Millrace batches pipeline workloads that can tolerate minutes of delay, moving suitable inference to less expensive CPUs and simplifying deployment in constrained environments. A separate citation service segments agent responses and uses hybrid similarity scoring to match assertions with source material, providing more deterministic traceability than citations generated voluntarily by an LLM.

Vannevar Labs sells to the U.S. government through contracts and task orders rather than published per-seat pricing. Contract economics combine annual or multiyear software access with data collection and enrichment, integration engineering, mission support, training, and infrastructure. The mix varies by deployment.

The go-to-market motion follows a land-prove-institutionalize sequence. Vannevar enters through rapid-acquisition mechanisms such as SBIR awards, Defense Innovation Unit prototype agreements, and Other Transaction Authority contracts, then embeds forward-deployed engineers with users to produce measurable operational outcomes. Its earliest Decrypt pilot grew from a roughly $25K three-month engagement to a $1.3M contract within four months after analysts produced intelligence that received senior-level attention.

The next commercial step is converting operational adoption into durable program funding, a transition the company describes as the second valley of death. Its $99M DIU production vehicle allows additional organizations to order production capabilities without starting a new acquisition process for each deployment. Through its SMX partnership, Vannevar can use existing prime-contract positions rather than serve as the prime contractor and full-service integrator for every opportunity.

The cost structure sits between high-margin SaaS and labor-heavy government services. Shared platform engineering, search infrastructure, and ML inference can be reused across deployments, while data collection, forward-deployed engineers, security accreditation, and environment duplication for classified and air-gapped networks add recurring costs. Vannevar reported a 45% reduction in ML inference costs, while Millrace lowers enrichment costs by batching workloads onto CPUs. Three consecutive profitable years through 2024 indicate that the software-and-data model generates higher margins than hardware-dependent defense companies.

Expansion comes from new organizational adoption and increased usage within existing accounts. A single command might start with intelligence search, then add targeting support, planning agents, and information-operations workflows. The 125 deployments across every service branch and combatant command, served by an estimated 40 to 55 paying organizational customers, imply average ACV per organization of roughly $2.5M to $3.5M, with deployment density rising faster than the number of top-level buyers.

Vannevar Labs operates in a defense AI market consolidating around a small number of enterprise platforms, pressuring specialist vendors to become indispensable components of those platforms or risk displacement.

Palantir is the most consequential competitor. Its Gotham, AIP, and Maven products span data integration, intelligence analysis, targeting, planning, and deployment across classified and tactical environments. As the software prime for the Army's TITAN intelligence ground station, Palantir controls the data and application layer through which competing products must integrate. Palantir's reported $10B enterprise contract creates a procurement pathway that could consolidate analytics and decision-support spending under one incumbent.

Vannevar differentiates through hard-to-access adversarial data and mission-specific intelligence workflows rather than a universal operating system. It could serve as a specialized intelligence layer for Palantir-enabled workflows, though this approach risks relegating Vannevar to a data supplier.

Anduril competes from an adjacent position, using Lattice to connect sensors, command-and-control software, and autonomous platforms. Lattice is moving upstream into planning and intelligence synthesis while controlling the downstream execution environment where data becomes tracks, tasks, and effects. Vannevar operates primarily before and around the kill chain, but the boundary is narrowing.

Primer is likely the closest product-level competitor in foreign-language intelligence, OSINT, and information-operations workflows. Its Delta platform structures unstructured intelligence, Primer Command monitors narratives, and its APIs embed extraction and verification into other mission systems. Primer also supports IL5+, JWICS, SIPR, and air-gapped environments, reducing a deployment advantage historically available to specialized defense startups.

Scale AI's Donovan platform allows defense customers to connect mission data, select models, and deploy agents for intelligence analysis, planning, and report generation. As a horizontal agent builder, its quality depends on the underlying data and configuration, but it could commoditize capabilities that once required a specialized vendor. Babel Street overlaps in global OSINT collection, entity resolution, and identity intelligence, particularly in homeland-security investigations where Vannevar is expanding.

Booz Allen Hamilton, Leidos, and SAIC can bundle AI into existing mission-support contracts alongside analysts, integration, accreditation, and sustainment. Their main weakness is development speed, but procurement consolidation around fewer accountable prime contractors would make their scale and contract access more valuable.

OpenAI, Anthropic, Google, and other foundation-model providers are entering national-security workflows through DoD contracts, reducing the cost of basic summarization, translation, and OSINT assembly. They may remain model suppliers rather than complete intelligence-workflow vendors, but integrators can increasingly deploy these models in secure environments and connect government data through retrieval. This raises the performance threshold a specialist such as Vannevar must meet to justify standalone procurement.

Vannevar Labs is expanding from intelligence search into a broader mission-agent platform while entering customer segments and geographies that increase its addressable market.

The largest expansion vector is moving beyond analyst productivity software into operational planning, targeting support, force protection, and information operations. These workflows draw from larger command-and-control and decision-support budgets. The FY2026 defense request allocated $13.4B to autonomy and autonomous systems, including approximately $1.2B for enabling software. Vannevar does not need to manufacture autonomous platforms to benefit. It can supply the intelligence, planning, and orchestration layer connecting sensors, commanders, and autonomous systems.

The company already operates dozens of agents across multiple environments. Continued investment in edge deployment, model monitoring, and deterministic citation infrastructure could extend the platform into higher-consequence workflows where operators must inspect evidence rather than receive only a generated answer.

Homeland security expands the addressable market beyond traditional defense and intelligence organizations. Vannevar's FedRAMP High authorization supports digital forensics, identity intelligence, foreign-media analysis, and network mapping across counter-narcotics, money laundering, human trafficking, sanctions enforcement, and border security missions.

Counter-fentanyl operations provide an entry point, with Vannevar mapping precursor chemicals, online marketplaces, logistics providers, and financial facilitators across fragmented networks. Because much of the underlying data is unclassified, outputs can be shared across agencies and with international partners without the declassification burden attached to traditional intelligence reporting.

International expansion adds another growth vector. Vannevar frames its Indo-Pacific product around coalition operations and allied campaign credibility, while its unclassified common operating layer allows Five Eyes partners, Japan, South Korea, NATO members, and other states to collaborate without exposing sensitive U.S. collection methods.

Allied customers can add regional expertise, language coverage, and local data to the product's sensing grid. Vannevar could enter these markets through U.S.-sponsored coalition deployments, foreign military sales, or local integrator partnerships, competing with European companies such as Helsing that benefit from sovereignty preferences in their home markets.

Procurement concentration: Nearly all of Vannevar Labs' demand comes from the U.S. government, so changes in appropriations, administration policy, continuing resolutions, or program priorities can reduce or eliminate revenue, while the 18-to-24-month gap between operational adoption and durable program funding requires the company to support products before recurring budgets are secured.

Platform displacement: Palantir, Anduril, Scale AI, major cloud providers, and incumbent integrators can bundle analytics, agents, sensors, and integration under larger enterprise contracts, potentially reducing Vannevar to a data supplier or niche application within another vendor's architecture rather than an independent platform.

Data-source fragility: Vannevar's differentiation depends on sustained, discreet access to adversarial digital environments, but foreign governments, platforms, and criminal networks can block collection, seed false information, change technical interfaces, or identify and disrupt proxy infrastructure, degrading the proprietary data layer underlying the product.