Cyberphysical Data: The Most Defensible Asset Class Investors Aren’t Pricing

Cyberphysical data is reshaping AI investment strategies by offering an edge that software-only ventures can’t match. While venture capital heavily funds AI infrastructure and foundation models, these areas are rapidly commoditizing. The real value lies in data generated through physical systems – like factory sensors, vehicle telemetry, and medical devices – which competitors cannot easily duplicate.

Key insights:

• AI commoditization: Costs for cutting-edge AI tools dropped 90% from 2024 to 2026, eroding software-only advantages.
• Cyberphysical data growth: The market is projected to grow from $15.2 billion in 2023 to $111.9 billion by 2033 (22.1% CAGR).
• Investment gap: Vertical AI startups, focused on physical data, received only 30% of AI capital in 2025 despite accounting for 53% of deal volume.

This overlooked asset class creates a self-reinforcing advantage through exclusive, hard-to-replicate data tied to physical infrastructure. Investors who prioritize cyberphysical applications over commoditizing infrastructure stand to gain significantly.

The Mispricing: AI Capital Flows to Commoditizing Infrastructure

Capital Concentration in Mega Deals

Recent investment patterns show a heavy tilt toward easily replicable digital infrastructures, leaving the unique value of cyberphysical data largely overlooked. By 2025, 73% of total AI investment value came from mega deals of $100 million or more (OECD, 2026). Instead of funding application-layer companies with strong defenses, much of this capital is flowing into foundation models and infrastructure that are quickly losing their edge. A prime example: OpenAI’s $40 billion Series F round in March 2025, led by SoftBank, set a record for the largest private raise ever. Just months later, Anthropic closed a $15 billion Series G, which alone accounted for nearly half of all AI funding that month. Together, these two deals outstripped the entire vertical AI sector’s funding for 2025.

The imbalance is striking. EY reported that VC-backed companies raised $80.1 billion in Q1 2025, but removing OpenAI’s $40 billion deal reveals a 36% quarter-over-quarter drop. The top 10 AI mega-rounds in 2025 funneled $84 billion into horizontal infrastructure. Meanwhile, vertical AI startups – those creating specialized solutions for specific industries – might have captured 53% of deal volume, but they only received 30% of total capital deployed. This shows a clear preference for investing in layers that are commoditizing rapidly, while underfunding the application layer where long-term defensibility actually exists.

Pure-Software Moats Are Collapsing

The dominance of mega deals also highlights a troubling trend: software-only ventures are losing their defensibility at an alarming rate. Frontier model capabilities now double every seven months, and open-source options like Llama 3, Mistral, and DeepSeek-V3 are performing within 5-10% of proprietary models. This rapid narrowing of performance gaps has slashed pricing power. For instance, the cost of cutting-edge AI coding assistants fell 90% between January 2024 and Q1 2026 – from $0.03 per 1,000 tokens to under $0.003. Even Microsoft has felt the pinch, reporting a 400 basis point year-over-year decline in Azure AI gross margins by early 2027 due to intense pricing competition.

Bowmark Capital’s 2025 analysis highlights how AI has automated processes like data collection, cleansing, and aggregation – tasks that once created valuable proprietary data moats. Today, large language models make mass data scraping 100 times cheaper and faster, allowing new competitors to replicate years of work for just 1% of the original cost. Founders Factory’s research echoes this, pointing out that startups relying solely on digital data face severe replication risks. They argue the only way to build a defensible position is through domain-specific machine learning applied to physical, industry-specific challenges. This approach leverages deep expertise and vertical specialization to create barriers that competitors can’t easily overcome.

"The market got the scarcity wrong. Intelligence is scaling rapidly and becoming commoditized… What is genuinely scarce is the physical infrastructure required to run that intelligence at scale." – Luciano Colos, Author, PitchGrade

These dynamics underline why chasing commoditized layers in AI investment is creating a structural mismatch.

The Paradox: Chasing Commoditized Layers

Investors are channeling resources into the least defensible areas of AI. Between 2023 and 2024, $560 billion flowed into AI infrastructure, which generated only $35 billion in revenue – a glaring mispricing issue. Despite this, 93% of venture capital dollars in Silicon Valley are now directed toward AI, even though 95% of AI pilots fail to deliver measurable P&L impact. Most of this capital is funneled into horizontal mega-rounds that back foundation models, leaving cyberphysical applications – where data is inherently more defensible – chronically underfunded.

The numbers tell the story. In 2025, horizontal AI companies averaged $63.1 million per transaction, while vertical AI startups averaged just $24.0 million. As Alberto Onetti, Chairman of Mind the Bridge, explains: "AI is absorbing a disproportionate share of venture capital: 93 cents of every VC dollar in the Valley now goes into AI." This focus on replicable intelligence is overshadowing the value of real-world, context-driven data generated through physical interactions. Such data, which requires specialized systems and physical presence, represents a massive untapped opportunity for competitive advantage.

Curious about which AI investments truly stand out? Subscribe to our AI Acceleration Newsletter for weekly insights on spotting defensible cyberphysical data opportunities over commoditizing infrastructure bets.

sbb-itb-32a2de3

What Makes Cyberphysical Data Defensible

Defining Cyberphysical Data

Cyberphysical data emerges at the intersection of physical hardware and digital systems. Think of sensors on factory floors, robotics in warehouses, biotech instruments analyzing clinical samples, or IoT devices monitoring environmental changes. This data reflects the "ground truth" – real-world conditions captured through hardware working in unpredictable environments. Unlike simulations, it accounts for variables like weather, lighting, and mechanical breakdowns. Nick Bunick from NewView Capital sums it up well:

"Physical AI creates proprietary context through devices operating in real environments that are unpredictable and constantly changing. Every interaction becomes a learning opportunity."

In this setup, hardware isn’t just a cost – it’s the engine driving data generation. Each interaction and deployment enhances the software, creating a unique, inaccessible context for competitors. For instance, the robotics and AI robots market is projected to leap from $15.2 billion in 2023 to $111.9 billion by 2033, with an annual growth rate of 22.1%. This growth underscores the value of physical systems in generating defensible data. Let’s explore why replicating this data is no small feat.

Why Replication Requires Physical Presence

Cyberphysical data can’t be conjured from a computer screen. Take Halter, a virtual fencing platform using solar-powered collars to track cattle behavior and grazing patterns. These collars act as live data nodes, constantly feeding insights into pasture management systems. For competitors to match this, they’d need to deploy similar hardware on actual farms and gather data over entire seasonal cycles.

Another example is Netradyne, which uses in-vehicle cameras to monitor driving conditions across fleet vehicles. These cameras capture rare, real-world scenarios – like specific weather patterns or unique driver behaviors – that can’t be adequately simulated in a lab. With every new deployment, Netradyne collects exclusive data points that refine its models in ways competitors can’t match without scaling up their own hardware.

As Buildloop AI puts it:

"The moat isn’t the model – it’s the continuous learning loop tied to a specific workflow, distribution channel, and service envelope."

Additionally, industries like healthcare, defense, and infrastructure demand strict compliance with security and data residency regulations. They also require deep integration into complex workflows – hurdles that software-only competitors struggle to surmount. These factors make physical deployment essential for creating data assets that are difficult to commoditize.

The Self-Reinforcing Data Flywheel

The challenges of replicating cyberphysical data create a self-reinforcing flywheel that solidifies competitive advantages. Here’s how it works:

• Physical nodes collect raw data from dynamic environments.
• AI transforms this data into proprietary intellectual property.
• Improved system performance drives customer value and ROI.
• Revenue from this success funds further deployment of physical nodes.

Each cycle strengthens the moat, as new deployments uncover unique edge cases that competitors without physical presence can’t replicate.

Take Verkada, for example. Its integrated physical security platform combines video, access control, and sensors. As its systems expand to more buildings, they adapt to real-world variables – different lighting, layouts, and threats – enhancing detection accuracy while reducing false alarms. Nick Bunick describes this as the "edge case advantage":

"Anywhere deployment meets variance, context compounds."

Once hardware becomes part of a customer’s operations, it creates high switching costs and locks in data, making it difficult for competitors to gain a foothold. The recurring software revenue built on top of this physical infrastructure is far more resilient than traditional SaaS models. As NewView Capital explains:

"Atoms do not compile. But once hardware is embedded in a customer’s operations, it becomes the foundation for recurring software revenue that is harder to displace than a SaaS login."

A prime example of this dynamic is Project Prometheus, launched in October 2025 with $6.2 billion in backing from Jeff Bezos. Focused on industries like chip packaging and aerospace, the company generates proprietary experimental data through AI-driven physical experiments on factory floors. This approach creates a flywheel of hardware deployment and data refinement that software-only competitors simply can’t match. It’s a powerful illustration of how cyberphysical systems maintain their edge in a competitive landscape.

Where Cyberphysical Data Moats Exist: Sector Analysis

Sports Performance and Biometrics

Athletes produce data that simply can’t be duplicated by sitting at a desk. Wearable devices used during training and competitions gather insights like biomechanical movements, heart rate variability, and recovery patterns. Collecting this kind of data requires direct access to athletes, their training environments, and the high-pressure settings of competition. These barriers make it nearly impossible for competitors to scrape or simulate equivalent datasets. As a result, AI models built on this data become exclusive, leveraging performance metrics from real-world, high-stakes scenarios. This concept of unique, hard-to-replicate datasets extends well beyond sports and into other industries.

Healthcare and Biotech

Take Tempus AI, for example. This company has built its edge using genomic sequencing data tied to physical samples and clinical outcomes. To replicate such a setup, you’d need partnerships with hospitals, adherence to strict HIPAA regulations, and access to specialized facilities for processing samples. Another standout, Abridge, raised $316 million in Series E funding in June 2025, achieving a valuation of $5.3 billion with a team of 488 employees. Abridge’s platform transcribes and summarizes clinical conversations, capturing diagnostic insights protected by regulatory restrictions. These processes create datasets that competitors can’t easily replicate remotely, giving these companies a solid advantage.

Industrial IoT and Manufacturing

Factory floors are treasure troves of sensor data that reveal the complexities of real-world operations. In 2025, investor interest in Manufacturing and Industrial AI surged, with deal counts jumping 41% from Q1 to Q4. Why? Sensor data capturing things like thermal signatures, acoustic patterns, and mechanical failures can be used to build predictive maintenance models. Collecting this data often requires boots on the ground, as it reflects the unique characteristics of specific equipment, workflows, and production environments. Each deployment generates proprietary edge cases, creating a significant barrier for software-only competitors.

Cleantech and Environmental Monitoring

Environmental sensors deployed across grids, farms, and industrial sites generate data that’s not only valuable but also operates within strict regulatory frameworks – creating natural barriers. For instance, Halter uses solar-powered collars on cattle to enable virtual fencing while monitoring animal behavior and land use. These collars act as live data nodes, feeding into models that improve grazing efficiency. Software-only competitors can’t replicate this feedback loop between livestock and land. Additionally, in cleantech, factors like energy access limitations and data residency requirements further strengthen these barriers. Cyberphysical data plays a crucial role here, supporting ESG verification, grid optimization, and carbon accounting, while also transforming supply chain management.

Supply Chain and Logistics

In January 2025, Fleetio raised $454 million in Series D funding to support a network managing over 8 million vehicles and processing 13 million repair orders annually across 110,000 repair shops. Similarly, Netradyne uses in-vehicle vision systems to continuously collect driving data from thousands of vehicles, turning raw streams into actionable insights. These datasets, gathered from vehicles, warehouses, and logistics systems, fuel innovations like route optimization and fraud detection. The physical infrastructure needed to gather such data creates significant barriers for competitors, who would have to independently collect similar real-world edge cases.

Across these industries, cyberphysical data moats offer a level of exclusivity that digital-only data simply can’t match. Physical infrastructure is the key to generating datasets that are defensible and impossible to replicate, giving companies a competitive edge that’s hard to break.

The Investment Arbitrage: Underfunded Cyberphysical Data Moats

Measuring the Mispricing

The numbers tell an interesting story. In 2025, vertical AI startups – those focusing on cyberphysical applications – accounted for 53% of all deal volume but secured just 30% of the capital deployed. To put this into perspective, the average deal size for these startups was $24.0 million, compared to $63.1 million for horizontal infrastructure companies. This disparity becomes even more striking when you consider that 93% of venture capital in Silicon Valley is funneled into AI, yet most of it goes to foundation models and GPU infrastructure, leaving application-layer businesses underfunded.

Data from the OECD reveals that mega deals over $100 million are overwhelmingly directed toward commoditized infrastructure. Meanwhile, EY’s Q1 2025 report highlights a concentration risk, with a single $40 billion transaction dominating the landscape. Between 2023 and 2024, a staggering $560 billion was poured into AI infrastructure, yet it generated a mere $35 billion in revenue – a lopsided 16:1 investment-to-revenue ratio. This misallocation of resources creates a prime opportunity for cyberphysical data companies to outshine traditional software-only models.

Why Cyberphysical Data Companies Will Outperform

With such uneven capital distribution, companies rooted in physical infrastructure are positioned for stronger performance. By embedding hardware and sensors into a customer’s operations, these businesses establish recurring revenue streams that are far more stable than those of SaaS models. The physical integration creates data advantages that digital-only competitors struggle to replicate. According to a 2025 MIT study, 95% of AI pilots fail to achieve measurable profit and loss impact, but companies with physical infrastructure see much better success rates because their value is tied directly to real-world operations instead of theoretical productivity gains.

"As the cost of intelligence goes down, the premium is shifting to what informs decisions: context… some of the hardest signals to replicate won’t be scraped from the internet. They’ll be found in the physical world." – Nick Bunick, NewView Capital

The market for robotics and physical AI is projected to grow from $15.2 billion in 2023 to $111.9 billion by 2033, with a 22.1% CAGR. Additionally, vertical AI companies in industries like Aerospace & Defense and Manufacturing are seeing valuation step-ups of 2.15x and 2.08x, respectively, compared to 1.80x for Financial Services AI. These valuation multiples reflect a growing investor belief that physical data moats are becoming more valuable as software-only advantages erode.

The Window for Early-Stage Entry

Given the long-term benefits of physical infrastructure, the early-stage market remains a goldmine for cyberphysical data ventures. This potential hasn’t yet been fully reflected in Series A and Series B valuations. The most promising activity for vertical AI is happening at the $1–5 million seed and pre-seed levels, with vertical deals accounting for 60% of activity by late 2025. In November 2025 alone, $7.7 billion was invested in physical applications like manufacturing, robotics, and defense – indicating a shift in capital allocation, though early-stage valuations still lag behind the strategic value these businesses will eventually command.

One standout example is Project Prometheus, which raised a record-breaking $6.2 billion in its first financing round in October 2025. While this mega-round validates the potential of vertical AI at the growth stage, the real opportunity lies in seed and Series A rounds, where the foundational physical infrastructure is still being built – offering a chance to invest in assets that will become irreplaceable over time.

How to Evaluate Cyberphysical Data Companies

Data Non-Replicability

One of the first things to consider is whether competitors could replicate the data using only remote digital tools. If physical presence is required to gather the data, it’s a strong indicator of defensibility. Pay close attention to regulatory and relationship barriers that ensure the data is collected through on-the-ground deployments, capturing scenarios that competitors simply can’t duplicate.

For instance, take Abridge, which reached a $5.3 billion valuation in February 2025. Abridge captures clinical conversations between doctors and patients – data that demands HIPAA compliance, partnerships with hospitals, and seamless integration into electronic health record systems. This isn’t the kind of data you can scrape from the web or generate artificially.

The key question is whether the data is rivalrous (limited to one user at a time) and non-substitutable (irreplaceable by other sources). In industries with heavy infrastructure requirements, physical constraints create lengthy timelines that competitors can’t shortcut. For example, when Constellation Energy signed a 20-year power purchase agreement with Microsoft in September 2024 to restart the Three Mile Island Unit 1 nuclear reactor, its stock price tripled by March 2026. Why? Because no competitor could replicate that kind of baseload energy capacity within a reasonable timeframe.

This unique non-replicability lays the groundwork for assessing a company’s commercial traction, where real-world results further validate its moat.

Commercial Traction and Customer ROI

With 95% of AI pilots failing to deliver measurable profit or loss improvements, commercial traction becomes the ultimate test of defensibility. The companies that stand out are those deeply integrated into workflows where removing their solution would disrupt essential operations. A great example is CompanyCam, which raised $415 million at a $1.99 billion valuation in 2025. Their photo documentation tool for contractors in the Architecture, Engineering, and Construction industries achieved a 3.21× valuation increase from the previous round by embedding tightly into critical workflows.

It’s important to verify that pilot projects evolve into broader rollouts, proving their ability to improve unit economics. Look for contracted revenue visibility – long-term agreements or multi-year service contracts that highlight the solution’s indispensability. Companies with "execution authority", or the ability to drive decisions in high-stakes workflows, hold a rare advantage that generic AI systems struggle to match.

"The moat isn’t the model – it’s the continuous learning loop tied to a specific workflow, distribution channel, and service envelope." – Buildloop AI

Strong market performance is essential, but it’s often the team’s expertise in a specific domain that truly sets these companies apart.

Domain Expertise and Vertical Focus

True defensibility often comes from data tied to physical interactions and deep sector knowledge, making the team’s expertise a critical factor. Founders with proven credibility in their industry are better equipped to access and leverage cyberphysical data. For example, Project Prometheus raised a record-breaking $6.2 billion in late 2025 by assembling a team of over 100 experts from DeepMind and Tesla. Their goal? Solve factory-floor challenges in chip packaging, automotive assembly, and aerospace – highly specialized problems that require intimate knowledge of these industries.

It’s also important to confirm that the company’s product is embedded in workflows that generate sensitive, hard-to-replicate data. Abridge’s founder, cardiologist Shivdev Rao, built the company to capture diagnostic reasoning during clinical conversations. This required real-time integration with patient care, enabling the company to scale to 488 employees and solidify its position in a space where general-purpose AI tools couldn’t compete.

In 2025, vertical startups accounted for 53% of venture capital deal volume, with companies raising $30 million or more making up 58% of those transactions. This trend highlights how industry-specific moats are highly attractive to investors. The ultimate question is whether the team has developed the expertise and processes needed to handle complex, nuanced datasets. As Nick Bunick of NewView Capital put it:

"As the cost of intelligence goes down, the premium is shifting to what informs decisions: context."

Conclusion: Cyberphysical Data as the Defensible Asset Class

Core Takeaways

The market often undervalues asset classes. While 93% of venture capital investments in Silicon Valley are funneled into AI, much of this funding goes toward infrastructure – things like foundation models, computing power, and horizontal tools. However, the real edge lies in the data that fuels these models, especially when that data is generated through physical means that competitors can’t easily duplicate.

Cyberphysical data comes from real-world deployments, making it uniquely difficult to replicate. Examples include sensors on manufacturing floors, lab-processed clinical specimens, telemetry from autonomous vehicles, and environmental monitors across energy grids. The cyber-physical systems market is projected to hit $255.3 billion by 2029, growing 15.5% annually. Despite this growth, much of the capital still leans toward infrastructure investments with shrinking margins. Companies that create feedback loops through physical data collection often remain underfunded, even though their data creates strong competitive barriers.

This reinforces the idea that the most resilient AI systems are powered by exclusive cyberphysical data. These insights shape our investment strategy and guide the next steps.

Next Steps

With these takeaways in mind, our focus is on scalable AI architectures that leverage this overlooked asset class. M Studio prioritizes companies that generate irreplaceable cyberphysical data. We design AI frameworks to organize that data into valuable intellectual property and connect it with capital markets. What AI strategies are you using to assess cyberphysical data moats? Subscribe to our AI Acceleration Newsletter for regular updates on AI systems and strategies for defensible data.

Discover more about M Studio’s approach to building AI-driven systems with strong data moats: https://maccelerator.la/en/#eluid1e3e2401

FAQs

How do you verify a cyberphysical data moat is truly non-replicable?

To confirm that a cyberphysical data moat cannot be duplicated, check whether its creation relies on physical infrastructure, exclusive access, or regulatory hurdles. For instance, data collected from sensors in industrial equipment or genomic sequencing in healthcare often requires partnerships with hospitals and adherence to strict regulations. These factors involve substantial investments of time, money, and specialized knowledge, making duplication nearly impossible and ensuring long-term protection.

What metrics show the data flywheel is driving customer ROI?

Metrics that showcase ROI from a cyberphysical data flywheel include enhanced security (such as quicker threat detection and response), increased operational efficiency (like minimized downtime and reduced expenses), and improved customer retention (evident in higher renewal rates). These metrics underscore the importance of proprietary physical data, which is difficult for competitors to mimic. This advantage translates into lasting customer benefits, including cost reductions, improved results, and a more resilient position against competition.

What are the biggest risks in cyberphysical data businesses (hardware, regulation, deployment)?

Cyberphysical data businesses face several challenges, with hardware dependencies, regulatory hurdles, and deployment challenges standing out as major risks.

• Hardware Dependencies: Physical systems like sensors or biotech instruments can be expensive to acquire and maintain. They are also vulnerable to wear and tear, technological obsolescence, and disruptions in the supply chain.
• Regulatory Hurdles: Industries like healthcare are heavily regulated, which can lead to delays in operations and increased compliance costs. Navigating these regulations often requires time, resources, and expertise.
• Deployment Challenges: Implementing these systems in real-world environments is no small feat. Physical settings are often unpredictable, and managing logistics or deploying specialized equipment demands skilled personnel and constant adjustments.

These risks highlight the complexities involved in operating cyberphysical data businesses, where both digital and physical components must align seamlessly.

Related Blog Posts

• Five Questions Every Investor Should Ask Before Backing an AI Company
• Why VCs Are Pricing Companies on Data Defensibility (And What That Means for Your $1M ARR Business)
• The Data Defensibility Thesis: Why Smart Capital Is Repricing Software
• Due Diligence for Data Moats: An Investor’s Evaluation Framework