Field AI

1.1 Identity and business model

Field AI was founded in 2023 and is consistently described by third-party profiles as a private robotics-intelligence company headquartered in Irvine, California. Its business is the intelligence layer rather than a proprietary robot body: the company licenses autonomy software and related integration for robots made by other vendors. Public descriptions call the core product a hardware-agnostic, physics-first Field Foundation Model intended to operate across quadrupeds, wheeled platforms, drones, and other embodiments in dynamic environments without relying on GPS or pre-built maps. That positioning reduces direct hardware manufacturing exposure but makes commercial scale dependent on integrations, partner hardware, and independently unverified model performance. The snapshot below separates closed financing from a pending term sheet and labels combined revenue-and-contract figures as exactly that, rather than converting them into revenue, ARR, cash, or backlog.[CO001, CO002, CO003, CO007, CO008, CO009, CO010, CO011, CO012, CO013]

1.2 Leadership, founder-market fit, and governance

The publicly identified founder group combines autonomy research and operating leadership. Ali Agha is identified as founder and CEO; Shayegan Omidshafiei as co-founder, president, and chief scientific officer; David Fan as co-founder and chief technology officer; and Duncan McIntyre as chief financial officer. Agha's NASA Jet Propulsion Laboratory and DARPA autonomy work gives the company direct founder-market fit for robots operating in uncertain, GPS-denied environments. Public profiles also identify Justin Saeheng as a co-founder and former COO, which is the only leadership-role change visible in the assigned evidence. However, the evidence does not establish a complete current executive roster, board composition, voting control, observer rights, founder vesting, or succession plans. This creates material key-person and governance diligence needs despite the technical pedigree.[CO004, CO005, CO006]

1.3 Financing, operating scale, and milestone interpretation

The defensible closed-financing baseline is $405 million raised across rounds, including a $314 million August 2025 round associated with a $2 billion valuation. Later reporting says Field AI signed a term sheet to raise roughly $700 million at about a $10 billion valuation, but the financing had not formally closed and no lead investor was named; it therefore is neither closed funding nor cash available. Public operating indicators are directional rather than audited: sources report more than $135 million of combined revenue and customer contracts, more than 30 customers as of June 2026, and deployments across hundreds of sites or environments. Headcount is directly conflicting, with one profile showing 200–225 employees and another showing 346 in October 2026. The timeline treats publication-backed milestones as evidence events, not proof of audited performance.[CO007, CO008, CO009, CO010, CO011, CO012, CO013, CO014, CO016]

1.4 Stakeholder map and diligence priorities

The disclosed financing syndicate includes financial investors, strategic technology investors, and industrial investors. Bezos Expeditions, Prysm Capital, and Temasek were reported as co-leads of the August 2025 financing, while Khosla Ventures, Canaan Partners, Gates Frontier, Intel Capital, NVentures, BHP Ventures, Samsung, and others appear in public investor lists. Their participation validates access to capital and relevant ecosystems, but it does not disclose ownership, liquidation preferences, protective provisions, board rights, or commercial dependencies. Customer and deployment claims indicate early commercial reach, yet an industry benchmark cited by Construction Dive found active construction robotics use fell from 65% to 46% even as positive evaluations increased. The central diligence task is therefore to reconcile headline traction with signed, recognized, and renewable economics at customer and contract level.[CO008, CO009, CO010, CO011, CO014, CO015, CO016]

1.5 Exhibits

2.1 Market boundary and status quo

Field AI's relevant market is software and integrated autonomy that lets existing or purpose-built robots perceive, navigate, inspect, map, and move through unstructured industrial environments. The company describes applications including site mapping, facility and equipment inspection, anomaly detection, large-scale data capture, unmanned material transport, security, telepresence, and search; that breadth is a product surface, not proof that all adjacent spending is immediately addressable. The boundary should therefore include edge autonomy software, sensor-compute payloads, integration, and recurring analytics tied to industrial workflows. It should exclude robot hardware, general factory automation that depends on fixed repeatable motion, broad enterprise AI services, and the full revenue of construction, mining, energy, or manufacturing customers. The status quo is a mix of human inspection and documentation, fixed automation, specialist robot vendors, system integrators, and manual or semi-automated workflows. This boundary preserves the strategic upside of a hardware-agnostic layer without treating every robotics or AI dollar as Field AI revenue.[CM001, CM002, CM004]

2.2 Sizing lenses and what they do not prove

Public estimates support a growing demand environment but not a single defensible TAM, SAM, or SOM for Field AI. The Business Research Company defines industrial AI broadly across hardware, software, platforms, and solutions and forecasts $13.69 billion in 2026 and $73.54 billion in 2030; that includes sectors and offerings well outside autonomous robots. GMI estimates the AI-powered industrial robot market at $9.9 billion in 2026 and $27.9 billion in 2035, while Mordor estimates the oil and gas automation market at $46.16 billion in 2026 and $63.19 billion in 2031. Fortune's narrower inspection-robotics-in-oil-and-gas estimate is $0.93 billion in 2026 and $1.51 billion in 2034. These figures are complementary lenses, not additive pools: they overlap in geography, software, hardware, and end markets, and some use proprietary estimation frameworks. The most decision-useful reading is that Field AI can enter large industrial budgets through narrower, safety- and uptime-linked workflows; public evidence does not isolate its serviceable share or support a precise SOM.[CM005, CM006, CM007, CM008, CM009]

2.3 Buyers, users, and adoption path

The initial buyer is likely an industrial operator or asset owner with a costly inspection, mapping, safety, or material-movement problem; the user is an operations, reliability, construction, mining, energy, or safety team; and the economic payer may be a site, business unit, or capital-project owner. Field AI's published applications span facilities and equipment inspection, mapping, anomaly detection, transport, and security, while market sources identify hazardous access, workforce safety, remote monitoring, and data quality as adoption drivers. The adoption path is consequently workflow-first rather than robot-first: prove a defined mission, integrate sensors and site processes, demonstrate reliable operation and usable data, then expand across fleets, sites, or adjacent missions. Construction Dive reports deployments across hundreds of industrial environments and cites use in the U.S., Japan, and Europe, but it also reports active robotics use among surveyed firms falling from 65% to 46% even as positive evaluations rose above 95%. That divergence makes paid deployment, repeat usage, and outcome evidence more important than enthusiasm or funding alone.[CM010, CM011, CM012, CM013]

2.4 Drivers, constraints, and valuation relevance

Demand drivers include labor scarcity, hazardous or inaccessible work, the need for remote inspection, asset uptime, and the ability to turn field activity into structured operational data. GMI describes a shift from fixed repeatable motion toward perception- and data-led execution, while Fortune identifies remote inspection and workforce safety as drivers. Constraints are equally material: Fortune cites extensive capital and maintenance requirements and a shortage of skilled personnel; Mordor highlights cybersecurity and safety-system economics in operational technology; and Construction Dive's usage data suggests conversion from positive evaluations to sustained deployment remains unresolved. Field AI's hardware-agnostic, risk-aware positioning could reduce dependence on a single robot form factor and widen the number of workflows it can address, but it does not remove site integration, certification, liability, change-management, or ROI hurdles. For valuation, the upside case requires evidence that deployments convert into recurring software or repeatable multi-site contracts; the downside case is that revenue remains project- or integration-led inside a capital-intensive adoption cycle. Public sources do not disclose enough gross margin, retention, cohort, or contract detail to resolve that mix.[CM003, CM014, CM015, CM016]

2.5 Exhibits

3.1 Competitive landscape and buyer alternatives

Field AI does not compete in a single product category. A buyer can choose a robot maker's native autonomy, a specialist mapping or inspection system, a general robot-foundation-model provider, a multi-vendor industrial platform, an internal development program, or continued manual and fixed automation. Independent landscape sources name Physical Intelligence and Skild AI as model peers; Google Intrinsic, Fanuc, and Boston Dynamics as incumbent or platform alternatives; and Brain Corp, Exyn, and Gecko Robotics as narrower operating competitors. These lists are illustrative rather than exhaustive. The relevant boundary is therefore the full industrial workflow budget, not merely the market for one robot body or one autonomy model.[CP001, CP002]

3.2 Capability, trust, and deployment comparison

Field AI's strongest positioning is the combination of cross-robot portability, local operation, physics-based probabilistic reasoning, and experience in unstructured industrial environments. Business Insider reported that its software spans humanoids, robot dogs, drones, and industrial rovers, while the company said it had more than 30 customers. That is meaningful commercial evidence, but it is not a task-matched benchmark. Physical Intelligence is a well-funded direct model peer with a hardware-abstraction approach, yet one refreshed source said it still lacked a commercial product or disclosed revenue. Field AI's own portability claim also remains conditional: adapters, tuning, testing, safety validation, and site-specific integration can prevent a common model from producing common deployment economics.[CP003, CP004, CP006, CP008, CP009]

3.3 Pricing, packaging, and distribution power

Public evidence does not provide comparable list or realized pricing for Field AI and its principal alternatives. Field AI appears to combine pilots, integration, software licensing, hardware or sensor payloads, and ongoing support, so its reported revenue-plus-contracts milestone cannot be treated as ARR or divided into a representative unit price. Competing capital is abundant: Physical Intelligence raised a reported $600 million Series B, while Skild AI was reported at $30 million of revenue and a $14 billion valuation. Distribution may matter more than financing. Fanuc's reported 1.1 million installed robots and Google's Intrinsic platform create an incumbent route to upgrade existing factory fleets without introducing a standalone autonomy vendor.[CP005, CP006, CP008, CP010, CP011, CP012]

3.4 Switching costs, displacement risk, and verdict

Field AI's moat will be durable only if deployments create reusable assets: validated robot adapters, mission configurations, customer integrations, operating data, safety cases, and trusted support procedures. Those assets can raise switching costs, but they can also reveal a services-heavy model if each new site requires extensive engineering. Incumbent platforms can compress the opportunity through installed hardware and established procurement channels. Specialists can displace a general platform by delivering a complete workflow, as illustrated by Gecko Robotics' reported Navy inspection contract. Adoption is not automatic: Construction Dive cited industry research showing active construction-robotics use declining even as favorable evaluations increased. Field AI's humanoid-manipulation hiring signals a broader roadmap, but public, independent production benchmarks remain the decisive gap.[CP007, CP013, CP014, CP015, CP016]

3.5 Exhibits

4.1 Monetization and public traction

Public descriptions point to a B2B monetization model combining software licensing with hardware-integration services. Sacra describes sensor-compute payloads that retrofit existing robots and software designed to run on edge devices. This supports a mixed software and integration model rather than a pure subscription assumption, but it does not establish realized pricing or revenue recognition. A Silicon Valley Investclub profile reports more than 30 customers and combined revenue plus customer contracts above $100 million in June 2026, rising above $135 million by October; the same profile says those totals derive from CEO comments and that GAAP revenue and profitability are undisclosed. Field AI's own site emphasizes thousands of missions across hundreds of sites and three continents, which is useful evidence of activity but not evidence of recognized revenue, gross margin, or customer retention.[CI001, CI002, CI003, CI004, CI005, CI015]

4.2 Pricing, sales motion, and efficiency proxies

No public list price, contract rate card, realized price, CAC, payback period, sales-cycle distribution, or renewal data was found in the assigned evidence. The available operating descriptions imply enterprise sales into construction, industrial, energy, security, and related safety-critical settings, with deployments requiring a compatible robot, sensor-compute payload, site onboarding, and operator workflow. That configuration can support higher contract values than commodity software, but it also creates implementation and support costs that could pressure gross margin. Construction Dive reports deployments across hundreds of industrial environments and says the company planned to use its 2025 financing for global growth, product development, and hiring; this is a growth investment signal, not a sales-efficiency measure. Underwriting should treat any implied revenue-per-customer or recurring-revenue conversion as unavailable until contracts and invoice-level revenue are provided.[CI005, CI006, CI007]

4.3 Cost structure and capital adequacy

The cost base is likely shaped by robotics research, edge-compute integration, field deployment, cloud analytics, safety validation, and a growing technical team, but public sources do not disclose expense categories or gross margin. Construction Dive reports that Field AI intended to accelerate global growth, continue locomotion and manipulation development, and double headcount by year-end after its 2025 financing. SetterVC lists 200–225 employees, while Silicon Valley Investclub lists 346 as of October 2026; the discrepancy reinforces that headcount is a signal rather than a settled input. Capital availability appears strong relative to the disclosed $405 million 2025 financing, but current cash, burn, debt, runway, use-of-funds detail, and financing covenants remain undisclosed. A reported October 2026 term sheet for approximately $700 million at a $10 billion valuation is not a closed financing and cannot be treated as cash on hand.[CI008, CI009, CI010, CI011]

4.4 Underwriting verdict and diligence blockers

The financial profile is promising on financing momentum and deployment breadth, but weakly underwritten on revenue quality. A $2 billion 2025 valuation was reported alongside the $405 million raise, while later coverage describes a proposed or reported $10 billion valuation that had not formally closed; without recognized revenue, the implied revenue multiple cannot be calculated responsibly. The most material blocker is the absence of a public bridge from customer contracts and missions to recognized revenue, contribution margin, renewal behavior, and cash generation. A second blocker is operating leverage: the payload-and-services model may produce durable software economics, but it may also require labor-intensive integration and field support. The practical verdict is track rather than commit on headline financing signals: obtain the general ledger revenue split, backlog definition, gross-margin bridge, deployment cost per site, customer concentration, receivables aging, cash balance, burn, debt terms, and financing-close evidence before underwriting valuation or runway.[CI012, CI013, CI014, CI015, CI016]

5.1 Product and operating model

Field AI is positioned as a software company that supplies the “brain” for third-party robots rather than manufacturing a robot platform. Its Field Foundation Models (FFMs) are described as physics-first, risk-aware models that combine sensor inputs and physical reasoning for robots working without prior maps, GPS, or predefined paths. Intel Capital describes deployment on any robot embodiment or vehicle and any sensor, while Sacra describes a sensor-compute payload attached to compatible robots and edge execution. This makes the product definition a control and autonomy layer, with integration work and software licensing as the likely delivery mechanism rather than a vertically integrated hardware sale. The public record does not disclose realized pricing, service-level commitments, or a standard implementation package.[CE001, CE002, CE003, CE004]

5.2 Architecture and customer workflow

The disclosed architecture has three functional layers: a physics- and uncertainty-oriented model layer, a multi-sensor and robot-adaptation layer, and an edge runtime that makes decisions on the vehicle. Sacra identifies Dynamics and Multi-Agent Foundation Models plus a Safety and Risk Awareness layer; the same source says camera, LiDAR, radar, and IMU data are combined and that processing occurs on-device with latency under 100 milliseconds. Intel Capital independently describes no prior map training, no GPS, and edge operation. In a typical workflow, an operator mounts the payload, positions the robot, starts the system through a tablet, and receives inspection or mapping outputs. These are source-described mechanisms, not proof that every supported embodiment has equivalent reliability or latency.[CE005, CE006, CE007, CE008]

5.3 Maturity, deployment, and differentiation

Maturity is strongest where Field AI has disclosed field deployment rather than laboratory architecture. Construction Dive reports deployments in the United States, Japan, and Europe across hundreds of industrial environments, while the company-issued release says robots operate daily across multiple industries and embodiments. The public construction proof is still concentrated: Field AI's reported DPR case study lists 45,000-plus photos, more than 100 miles walked, and 500,000 square feet of interiors scanned. TechCrunch reports contracts across construction, energy, and urban delivery but says the company declined to name customers. The resulting differentiation is a credible hardware-agnostic and risk-aware thesis plus deployment data, not independently verified benchmark superiority against competing autonomy stacks.[CE009, CE010, CE011, CE012]

5.4 Trust, safety, and diligence limits

Risk awareness is central to the product narrative: Field AI says the models reason about uncertainty and physical constraints instead of retrofitting language or vision models after the fact. That approach is relevant to hazardous, remote, or connectivity-constrained sites, but the assigned evidence does not establish a formal safety case, certification, incident history, cybersecurity controls, privacy policy for collected site data, or customer-validated failure rates. The company’s GitHub organization states that it has no public repositories, limiting direct inspection of code, releases, issue history, and developer adoption. Product diligence should therefore prioritize embodiment-specific reliability distributions, rollback and human-override procedures, security attestations, data retention and model-training terms, and evidence that field performance generalizes beyond selected construction deployments. The company's August 2025 roadmap statement directed new capital toward locomotion and manipulation, but it did not provide delivery dates or acceptance criteria.[CE013, CE014, CE015, CE016]

5.5 Exhibits

6.1 Buyer, user, and segment structure

Field AI's observable customer motion is enterprise and operational rather than self-serve. Construction is the clearest documented wedge: DPR's superintendent describes the system as supporting photography and routine documentation, while the solution is positioned for safety, progress tracking, scanning, and defect workflows. The buyer and daily user are not necessarily the same person: contractors or project owners procure the capability, superintendents and safety teams review outputs, and robot or platform partners can provide deployment leverage. Company material also names energy, manufacturing, urban delivery, inspection, mining, logistics, and federal applications, but those references do not establish paid production use for each vertical. The resulting segmentation supports a land-and-expand thesis, while leaving procurement owner, pricing, and revenue contribution by vertical open.[CU001, CU002, CU003, CU004]

6.2 Adoption trajectory and deployment proof

The best public adoption evidence is specific to work performed, not logo count. A DPR deployment is described as having run for approximately a year and a half, with 45,000-plus photos, more than 100 miles walked, four floors mapped, 125,000 square feet of roofing documented, and 500,000 square feet of interiors scanned. Boston Dynamics separately says Spot equipped with Field Foundation Models has been deployed across numerous construction sites and reports inspection and documentation time reductions of more than 90 percent versus manual processes. These are meaningful workflow and utilization signals, but they are not a cohort, active-account count, renewal rate, or independently audited outcome. Field AI's 2025 announcement reports deployments across hundreds of complex industrial environments and multiple countries; because that statement is company-issued, it should be treated as directional rather than a verified customer denominator.[CU005, CU006, CU007, CU008]

6.3 Named proof, expansion, and durability limits

Named evidence spans DPR Construction, Big-D Construction, Ryan Companies, Boston Dynamics, Certis, and a McLaren Construction trial, but the evidence quality differs. DPR has the most concrete operating detail and a named superintendent quote; the Boston Dynamics release provides a partner-confirmed construction deployment and enterprise-scale expansion language; Robots in Construction lists additional organizations and a UK trial but is a secondary product directory. Certis is evidence of a strategic partnership and intended multi-site orchestration, not proof of contracted recurring revenue. Public sources do not disclose churn, GRR, NRR, renewal terms, contract length, customer concentration, or expansion revenue. Construction-market evidence also cautions that positive sentiment can coexist with selective adoption: active robotics use fell from 65% to 46% in BuiltWorlds data while repeated use became more selective. This makes customer durability the central unresolved underwriting issue.[CU009, CU010, CU011, CU012]

6.4 Commercial implications and diligence gaps

Field AI appears to be selling a workflow product whose value is realized through integration into live site operations, human review, and repeat capture rather than through a one-time robot sale alone. That model can support expansion from one project to multiple projects, fleets, or adjacent inspection and security use cases, but it also creates implementation and procurement friction. Customers must validate safety, integrate outputs with BIM or command systems, train teams, and decide who owns remediation. The available sources show partner ecosystem activity and concrete site work, yet do not establish gross retention, net expansion, average contract value, deployment payback, or the share of deployments that converted from pilot to production. Investors should require a customer cohort schedule, contract and renewal evidence, and project-level unit economics before treating deployment breadth as durable recurring demand.[CU013, CU014, CU015]

6.5 Exhibits

7.1 Ranked risk profile

Field AI's risk profile is dominated by the gap between a safety-oriented product narrative and the public evidence needed to underwrite safe operation at scale. The company says its models are designed for dynamic, unstructured environments and are already deployed across construction, energy, manufacturing, delivery, and inspection sites, but those descriptions do not disclose incident rates, near misses, false negatives, rollback performance, insurance coverage, or the denominator of successful missions. The International AI Safety Report states that general-purpose AI failures can create physical harm, financial loss, reputational damage, and legal liability, and that no combination of current methods eliminates failures entirely. For Field AI, the investment-relevant risk is therefore operational proof: whether bounded operating domains, human override, safe shutdown, testing, and monitoring are sufficient for each customer environment. The heatmap below treats the residual exposure as high where public evidence is thin, not as proof that a harmful incident has occurred.[CR001, CR002, CR003]

7.2 Legal and regulatory exposure

The legal surface is expanding in parallel with the company's physical deployments. K&L Gates describes a shift toward treating AI functionality as a product, with plaintiffs targeting architecture, guardrails, warnings, escalation pathways, and foreseeable misuse rather than only individual outputs. It also identifies potential liability up and down the supply chain, which matters to a software company that integrates with robot manufacturers and customer workflows. The firm's March 2026 analysis says the EU Product Liability Directive treats software, including AI systems, as products and covers substantial modifications, while member states must transpose it by December 2026. Baker Botts separately reports 2026 state laws that impose risk-framework, incident-reporting, impact-assessment, disclosure, or discrimination obligations in relevant AI contexts. These sources are not evidence of a Field AI violation. They establish a changing compliance and litigation perimeter, making product definition, version history, testing records, safety claims, data governance, export controls, and customer allocation of responsibility material diligence items.[CR006, CR007, CR008, CR009]

7.3 Operational, partner, and financial dependencies

Deployment execution creates several linked risks. Field AI's commercial model combines software with hardware integration and deployment services, so delivery quality depends on sensor-compute payloads, compatible robot platforms, site data, customer operators, and reliable support. A job posting describes on-device autonomy, customer project-tool integration, and commercial deployment services, while public descriptions emphasize operation without maps, GPS, or predefined routes. That breadth can accelerate adoption but also expands integration, calibration, cybersecurity, and support obligations across different robots and jurisdictions. Construction Dive reports that positive evaluations of construction robotics increased while the share of firms reporting active use fell from 65% to 46%, a warning that enthusiasm does not equal durable utilization. Funding and valuation signals also do not establish a settled current mark: PrivateMarketView carries forward the $2 billion last priced event, while SetterVC labels a possible $700 million raise at a $10 billion valuation as an unclosed, unverified signal. Without audited financials, customer-level concentration, renewal data, or unit economics, capital intensity, procurement friction, and the ability to convert pilots into repeatable revenue remain material open risks.[CR011, CR012, CR013, CR015]

7.4 Mitigations, monitoring, and thesis breaks

The most credible mitigation path is evidence-backed operational discipline rather than a broader marketing claim. Field AI's public materials describe risk-aware architectures, edge processing, and uncertainty-aware control, but the public record does not quantify how those controls perform by robot type, operating design domain, customer site, or software version. The SRES safety analysis argues that separation and very slow operation can reduce harm but limit utility, while safe shutdown and safety-of-intended-function testing are needed for robots operating around people. A practical diligence program should require a site-by-site hazard analysis, defined operating envelope, test and incident logs, override and shutdown demonstrations, cybersecurity and privacy controls, insurance and indemnity terms, and customer renewal or expansion evidence. Thesis-break indicators are a recordable safety event, a regulator or customer suspension, inability to reproduce safe behavior after model updates, a material outage or breach, repeated pilot non-conversion, or financing terms that expose a large gap between the latest priced round and current marks. Until these are measured, risk should remain high and the recommendation should not rely on an assumption of incident-free operation.[CR017, CR019, CR020, CR021]

7.5 Exhibits

8.1 Investment case and anti-thesis

Field AI has a credible software-first investment case. CNBC reports $405 million raised across two rounds and a $2 billion latest-round valuation, while Construction Dive describes deployments across hundreds of industrial environments and a physics-first autonomy stack intended for unstructured sites. The platform thesis is attractive because hardware-agnostic software could scale across robot form factors without requiring Field AI to own factories. The anti-thesis is equally material: current public evidence does not establish audited recurring revenue, gross margin, profitability, customer concentration, or conversion of deployments into durable contracts. The company therefore has strategic momentum but limited public underwriting visibility.[CV001, CV002, CV003, CV004, CV005]

8.2 Financing context and entry discipline

The observable financing anchor is the August 2025 $2 billion valuation reported by CNBC, not a completed 2026 transaction. SetterVC and the Silicon Valley Investclub tear sheet describe an approximately $700 million round at a $10 billion valuation as reported, in talks, or not confirmed closed. Those marks should be treated as indications rather than a settled price. The implied step-up from $2 billion to $10 billion is fivefold, but no public revenue denominator supports a defensible valuation multiple. Entry discipline should therefore require confirmed closing terms, primary financials, preference rights, and an independently verified revenue bridge.[CV006, CV007, CV009, CV010]

8.3 Scenarios, comparables, and downside

A scenario framework is more decision-useful than a point estimate. The bull case requires Field AI to convert multi-site deployments into repeatable, high-margin software contracts and to preserve differentiation as NVIDIA, Physical Intelligence, Skild AI, and other robotics platforms compete for the same layer. The base case assumes meaningful technical and customer progress but continued opacity, making a track decision more defensible than a buy. The bear case combines a financing reset with slower construction adoption, weak customer ROI, or a market multiple contraction. Public comparables in the tear sheet show wide historical price-to-sales ranges for autonomy and AI names, but those IPO-era yardsticks are not directly comparable without Field AI revenue and margin disclosure.[CV011, CV012, CV013, CV014]

8.4 Recommendation and thesis-break triggers

The evidence supports track rather than buy. A buy would require a verified price, a GAAP-to-contract-revenue reconciliation, customer-level retention and concentration data, gross-margin evidence, and cap-table terms showing that preference and dilution risk are acceptable. The thesis should be upgraded if a closed financing or secondary transaction confirms a price near the reported mark while audited operating metrics demonstrate durable software economics. It should be downgraded if the reported round fails to close, if revenue is primarily nonrecurring services or hardware integration, if deployments do not convert into paid expansion, or if safety and commercialization costs prevent attractive margins. The immediate diligence path is to obtain the data room and independently verify the round, contracts, unit economics, and governance rights.[CV015, CV016, CV017]

8.5 Exhibits