Harry Dewhirst: What If the Physical World Had an API? — 375ai, DePIN, and Real-Time Data for AI

This episode explores 375AI's ambitious venture to create a decentralized sensor network, transforming real-world physical data into monetizable assets for AI and beyond, offering a new frontier for DePIN investors.

Harry Dewhirst's Journey: From Web2 Data Monetization to Web3 DePIN

• John from Delphi Ventures introduces Harry Dewhirst, co-founder and CEO of 375AI.
• Harry recounts his entrepreneurial journey, starting with a company that helped mobile operators monetize user information, raising funds from prominent VCs like Sequoia and Excel, and eventually selling to Singtel.
• He then became president at Bliss, a data business using mobile phone location data to understand real-world movement, valuable for retail and hospitality, which was later sold to T-Mobile.
• His last role before 375AI was CEO of Linksys, the Wi-Fi router company. It was here, about three to four years ago, that he and his co-founder Rob became fascinated by projects like Helium, a decentralized wireless network.
  • Helium: A decentralized wireless network that uses blockchain and token incentives to build and maintain wireless infrastructure.
• Harry mentions his early retail investments in crypto since 2015-2016 but had no professional crypto exposure before 375AI. His interest in DePIN (Decentralized Physical Infrastructure Networks)—networks that use token incentives to build and operate physical infrastructure—grew, leading him to attend conferences and connect with industry peers.
  • "It's fascinated me and and now my co-founder Rob so much that um that we left links to launch launch 375 and that's uh yeah just coming up to to almost 3 years ago to the day," Harry shares, highlighting the conviction that drove the new venture.

The Genesis of 375AI: A Pivot to Proprietary Network Building

• John probes the "light bulb moment" that led to 375AI building its own network.
• Initially, 375AI focused on being an enterprise-grade deployer of other DePIN projects, becoming a large miner for Helium, XNET, and others.
• This experience provided insights into tokenomics, concepts, demand-side dynamics, and business models of existing projects.
• A key asset was an exclusive real estate contract with Outfront, a publicly traded company, granting access to 50,000 prime billboard locations. Harry explains they felt this asset was underleveraged.
• The team decided to build their own project, leveraging their data background and lessons learned, aiming to "attack real world data in a in a new and novel way."

375AI Overview: Capturing Real-World Data at the Edge

• Harry provides a foundational overview of 375AI's current operations.
• 375AI is an edge data intelligence network deploying advanced sensors in strategic US locations.
• Devices feature an Nvidia GPU core running a proprietary LLM (Large Language Model)—an AI model trained on vast amounts of text and code to understand and generate human-like text—and computer vision software.
  • Computer Vision: A field of AI that enables computers to interpret and understand visual information from the world, like images and videos.
• Sensors include six specialized cameras, microphones, and atmospheric/environmental sensors.
• The initial focus is on vehicular data from busy highways, identifying:
  • Make, model, color, number of occupants, license plates.
  • Road incidents (collisions, erratic driving, emergency vehicles).
  • Commercial vehicle characteristics (DOT numbers, trailer IDs, carrier names like Hapag-Lloyd, Evergreen, UPS, FedEx, and contact details from vehicles).
• This detailed data on commercial vehicles is particularly valuable for understanding economic activity.

The Value Proposition: Turning Noisy Video into Actionable Data

• John questions how 375AI makes inherently noisy real-world data useful and monetizable.
• Harry emphasizes that while millions of cameras exist, they capture video, not structured data. 375AI's system converts high-definition video streams into anonymized, privacy-compliant data in real-time at the edge.
• "The video itself never leaves the device. So there is no privacy um kind of concern or or or regulation to be abided by because the video is neither stored nor used," Harry clarifies, highlighting a key differentiator.
• This process of normalizing unstructured data into a usable format, combinable with other data, is a significant advancement. It provides deterministic data about specific real-world locations for the first time, enabled by technological progress.
• Strategic Implication: For AI researchers, this offers a novel, structured dataset of real-world events. For investors, the privacy-first approach mitigates regulatory risks often associated with data collection.

Concrete Use Cases: Beyond Satellite Imagery

• John asks for concrete examples of how 375AI's data can be used, contrasting it with alternatives like satellite imagery for hedge funds.
• Harry uses the hedge fund example: satellite imagery of retail parking lots (e.g., Best Buy, Target, Walmart) helps predict financial results. However, satellite data is expensive, infrequent (snapshot in time), and struggles with calculating unique visitors.
• 375AI's devices, deployed on thoroughfares near such locations, capture every vehicle, including license plates (a unique identifier). This allows accurate measurement of both reach and frequency.
• Harry mentions a device in Redwood City, California, on Highway 101, which has detected the same Ford Transit van thousands of times, illustrating the importance of unique identification versus simple counts.
• The data also allows for demographic assumptions based on vehicle types (e.g., Kia Sportage vs. Range Rover, F-150s vs. Teslas) and how these change by time of day.
• The data is stored in perpetuity, increasing its richness over time for historical analysis and trend prediction.
• Actionable Insight: Investors should note the higher fidelity and real-time nature of 375AI's data, offering a potential edge over traditional alternative data sources for market intelligence and economic forecasting.

Current Stage and Path to Monetization: Avoiding the "Build It and They Will Come" Trap

• John inquires about 375AI's current operational stage and whether it needs critical mass for its data to be useful, referencing Helium's scaling approach.
• Harry states they aimed for immediate demand and utility, avoiding the "build it and they will come" pitfall.
• 375AI has already signed paying customers even before its TGE (Token Generation Event)—the moment a project's cryptocurrency token is first issued and made available to the public—meaning on-chain revenue and token burn will occur almost immediately.
• This was achieved by focusing on high-quality install locations. The powerful edge devices cost $50,000 each. The first 50 units were pre-sold in Q3 last year, manufactured in Q4, and deployment began in Q1.
• A geographic focus was adopted:
  • First ~20 units in LA County, particularly around Long Beach (busiest US port), targeting commercial/freight traffic.
  • Next, New York/New Jersey.
  • Then, South Florida.
• This market-by-market densification approach, rather than disparate national deployment, proved effective. In LA, they see millions of cars daily, representing a double-digit percentage of unique cars in LA County.
• "There is a level of critical mass and scale even with a small number of devices placed in the right places that can be attained," Harry asserts.
• Strategic Implication: 375AI's early monetization and focused deployment strategy de-risks the typical DePIN challenge of balancing supply-side growth with demand generation.

Scaling and Emergence of New Buyers

• John asks if reaching a larger scale, like national coverage, would attract new types of data buyers.
• Harry confirms that current data from major markets like NY, LA, and Miami is already valuable, especially for transport and logistics businesses, as their devices cover key arteries like those around Long Beach port. "There isn't a container that comes in or out of Long Beach that we don't at some point see."
• National coverage will indeed open up more use cases. The passage of time, creating richer historical datasets, will also unlock further applications, enabling robust trend analysis.
• To fill gaps between the "big rock" 375 Edge devices, they've developed a smaller form factor device (375 Street) to create a more comprehensive picture.
• Harry anticipates both contemplated and surprising new use cases as the data becomes more accessible via a platform for businesses to integrate into their BI models.

Go-to-Market Strategy: Leveraging Existing Data Marketplaces

• John inquires about how 375AI gets its data in front of buyers.
• Harry explains their initial strategy involves integrating with numerous existing data platforms and marketplaces, some owned by large companies like Google and Oracle, others independent.
• Large businesses (e.g., McDonald's, Coca-Cola, Walmart) typically have preferred data platforms. By publishing data to these platforms, 375AI allows Fortune 500 companies to consume it easily within their existing workflows.
• "It's within their muscle memory and they they don't have to there's no real change of behavior," Harry notes, emphasizing the ease of adoption for enterprise clients.
• These platforms serve diverse clients; one platform might have customers ranging from McDonald's and FedEx to Citadel.

Tech Stack Deep Dive: Edge AI vs. Cloud AI

• John transitions to 375AI's tech stack, specifically the decision to run AI at the edge.
• Harry states the decision for edge AI was intentional, despite being harder than centralized cloud processing.
• Two main reasons:
  • Efficiency: Streaming and storing vast quantities of high-fidelity video data at scale would be costly.
  • Privacy: The evolving privacy landscape is a key concern. Processing data locally on the device, without transmitting or storing raw video, ensures privacy compliance and security. "If we don't send the data anywhere, there's no ability for it to be intercepted."
• Raw footage is processed by AI chips on the device, structured text data is outputted, and the video is deleted.
• The model can be trained to stop capturing specific data if regulations change or for deployment in new geographies with different rules.
• While the heaviest AI models can't run at the edge due to cost, 375AI uses sample data sets sent to cloud-based multimodal systems for training, which then feeds improvements back to the edge devices. This offers a "best of both worlds" approach, focusing the edge models on relevant parameters.

Sensor Design and Modularity

• John asks about the trade-offs in sensor selection and device cost.
• Harry explains the system is modular. If new data types become valuable, new sensors can be relatively easily added. Cameras are the most expensive sensor components.
• The current six cameras on the 375 Edge device are specialized:
  • They have different apertures and fields of vision, looking in two directions (e.g., northbound and southbound traffic).
  • Some are honed in for details like license plates, logos, and text.
  • Others have a wider field for identifying vehicle make/model and occupants.
• "It's not six of the same cameras. It's different cameras doing, you know, programmed and and tuned to do different things simultaneously and then marry up together um that information," Harry details. This ensures comprehensive data capture for a single vehicle without multiple counts.

Expanding the Network: 375 Edge and 375 Street Devices

• John discusses the different device form factors and ensuring data quality with self-deployed units.
• 375 Edge: The large, powerful devices installed in prime billboard locations, representing a curated network scaling approach.
• 375 Street: A miniaturized version of the Edge device, roughly shoebox-sized, with an Nvidia GPU at the edge (smaller chip) and two cameras.
  • Designed for self-deployment by individuals or small businesses with interesting vantage points (balconies, rooftops, storefronts).
  • Can be co-located with other DePIN devices like Helium hotspots, allowing deployers to monetize locations in multiple ways.
  • Sales for 375 Street are planned for later this year.
• Harry notes that while a few thousand Edge devices will provide significant coverage with diminishing returns beyond that, tens or hundreds of thousands of Street devices could create an incredibly rich national dataset. The company also has international expansion aspirations.
• Actionable Insight: The dual-device strategy allows for both high-value, strategically placed data capture (Edge) and broader, community-driven coverage (Street), appealing to different types of network participants and investors.

Comparison with Hivemapper: Stationary vs. Mobile Data Capture

• John draws a comparison with Hivemapper, which uses vehicle-mounted dashcams.
• Hivemapper: A DePIN project that incentivizes users to collect street-level imagery using dashcams, primarily for mapping and autonomous vehicle applications.
• Harry, a fan and early user of Hivemapper, notes that while there's data overlap, use cases differ. Hivemapper's data is from within the car, limiting its view of overall road conditions.
• "It doesn't give you a a a really a a view into what's going on on a particular road because for it to do so, it would require... every other car to have a hive mapper in it," Harry explains. It's a sample-based system not ideal for general traffic flow determination.
• 375AI's stationary cameras, conversely, capture nearly all vehicles passing their fixed points, regardless of in-vehicle tech.

375 Go: The Mobile App Layer

• Harry introduces 375 Go, expanding the network to mobile phones.
• Launched in testnet around November/December last year, 375 Go allows anyone to contribute data using their phone, envisioned as the "smallest sensor" on the network.
• Phones have cameras, microphones, LiDAR (Light Detection and Ranging)—a remote sensing method that uses light in the form of a pulsed laser to measure ranges (variable distances) to the Earth—and other sensors.
• Currently, around 200,000 testnet users contribute anonymized location data and RF (Radio Frequency) signatures (cellular, Wi-Fi, Bluetooth networks detected, and connection speeds).
  • This data is useful for ISPs, MNOs, etc., for understanding network congestion and quality of service.
• Future plans include more active participation tasks, e.g., users in Paris collecting real-time gas prices for a customer, with users rewarded for their contributions, driving network utility and token burn.
• "Like the you had the rocks and and sand analogy. So this is kind of like the water I imagine," John remarks, to which Harry agrees it's like "the dust of the water."

Synergies with AI: Physical World Data for LLMs (Comparison with Grass)

• John compares 375AI with Grass, another Delphi Ventures portfolio company focusing on web data for LLMs.
• Grass: A DePIN project that allows users to sell their unused internet bandwidth, which can be used to scrape public web data for training AI models.
• Harry sees a strong analogy: Grass provides real-time web data for LLM inference; 375AI can be the equivalent for real-world physical data.
• "This information being um incredibly valuable to the LLMs uh or or to you know customers of the LLMs," Harry states. Businesses like FedEx, for whom vehicle movement is paramount, would want this data in their proprietary LLMs.
• 375AI aims to develop relationships with OpenAI, Anthropic, XAI, etc., to become a leading proprietary data source for physical world events.
• John points out a key difference: Grass focuses on inference as pre-training web data is abundant. For the physical world, high-quality data is scarce, so 375AI could also serve training use cases.
• Harry agrees: "LLMs don't have eyes and ears in the real world... they have a huge sensory blind spot to uh the real world." This data can unlock a "fourth dimension" for LLMs, especially in robotics, automation, and autonomous vehicles, complementing their onboard sensors.

The Endgame: Proprietary AI Models?

• John asks if 375AI might eventually build its own AI models beyond data provision.
• Harry mentions a recent conversation with their Chief AI Officer, Chad, who noted their increasing expertise in vehicle-related identification (vehicles, incidents, erratic driving).
• "He believes that given the quantity and scale of of of data that we're training upon... we quite quickly, if we're not already becoming like the very best at identifying vehicles," Harry shares.
• These specialized AI capabilities could become a commercial offering, as their unique, large-scale dataset provides an edge over generic open-source object recognition models.
• Strategic Implication: For AI researchers and investors, 375AI is not just a data provider but potentially a developer of highly specialized AI models for physical world understanding, creating additional value streams.

Biggest Risks and Challenges: The Hurdles of Hardware

• John inquires about the biggest risks keeping Harry up at night.
• Harry identifies manufacturing physical hardware as a significant and constant challenge.
  • Experiences with chip shortages, GPU high demand, tariffs, and logistics complexities.
  • "There are so many elements out of your control," he admits, though his and Rob's experience at Linksys (making millions of routers) is invaluable.
• He empathizes with DePIN enthusiasts waiting for pre-ordered hardware, acknowledging the difficulties faced by manufacturers. "It could be one little chip... and they've got 99 other chips, but if they don't have that one, it doesn't work."

The Story Behind the Name: 375AI

• John asks about the origin of the name "375AI."
• Harry reveals it's inspired by Highway 375, the "Extraterrestrial Highway," which runs through Area 51 in Nevada.
• "We we kind of thought there was some analogy between I guess the extraterrestrial world and what's now capable and possible with the advent of AI to to create otherworldly type experiences that that have never been done before," he explains. It signifies the art of the unknown and aligns with their focus on vehicles and transport.

Conclusion: A New Data Layer for the Physical World

This episode reveals 375AI's strategy to build a critical data infrastructure for the physical world, directly feeding AI and enterprise. Investors and researchers should monitor its unique edge-processing approach and multi-layered network growth as a bellwether for real-world data monetization and AI integration.