A turbine only ever talks about itself. A truck brings back the world.

Last edition, our signal engine collapsed thousands of themes into a single dominant answer: predictive-maintenance time-series, the timestamped pulse of machines that nobody keeps. 43% of every data holder we'd mapped. 100% time-series. A $1.4-trillion-a-year problem underneath it.

That edition was about a use case cutting across every sector. This one is about the opposite: a sector cutting across every use case. And it broke the pattern.

Because when we sorted our map by sector rather than by use case, one name came back that refuses to behave like the rest of the physical economy — mobility. Not because it's the loudest. Because it's the only one whose data carries an entire world inside it.

Here's what six weeks of signals taught us.

1. What the machine heard: 532 dated signals, and mobility punching above its weight

A reminder of how our engine works, because it matters for what follows: we call it signal-first. We never start from an opinion about where the data is. We start from facts — dated, sourced, external. A funding round. An OEM opening its telematics network. A regulator approving an autonomous system. A fleet electrification mandate. Each signal is a real article from the trade press, with a URL and a date, or it doesn't exist.

Between 1 June and 15 July 2026, mobility holders in our map were sourced by 532 dated signals — every single one press-sourced, none synthetic, none inferred.

That number is the first thing that made us look twice. Mobility accounts for 30.8% of the holders we've published — but 34.6% of all the dated signals we collect. The real economy talks about mobility more than mobility's share of the map would predict.

And it's talking about mobility specifically, not as a by-product of "industry" in general. Of the 271 unique signals behind those holders, 196 — 72% — appear on mobility holders and nowhere else. This is a distinct conversation, not the spillover of the industrial one.

The press doing the talking is exactly the press you'd want: FreightWaves (133 signals), Journal de l'Automobile (103), Supply Chain Dive (87), Supply Chain Magazine (49), Manufacturing Dive (23) — twelve trade outlets, no general-news filler. 88% of our mobility holders are sourced by at least one such signal; 91 of 134 are corroborated by two or more independent outlets.

A sample of what the engine actually picked up, verbatim, with dates:

• "Webfleet intègre les marques du groupe Volkswagen à son réseau de télématique OEM" — Journal de l'Automobile, 17 June 2026. An OEM's fleet telemetry becoming reachable through a third party. That is the whole thesis in one headline.
• "Volvo Autonomous Solutions to remove safety drivers in Q1 2027" — FreightWaves, 12 June 2026.
• "Gatik to bring autonomous freight to PepsiCo's North American supply chain" — The Robot Report, 12 June 2026.
• "La Belgique approuve à son tour le système de conduite autonome de Tesla" — Journal de l'Automobile, 12 June 2026.
• "Samsara Ride Along pushes fleet safety AI beyond incident flagging" — FreightWaves, 26 June 2026.
• "A Driver's Paper Logs Said He Was in One Place. A Roadside Camera Network Said Otherwise" — FreightWaves, 4 June 2026.

Read them together and they stop being news. They're a countdown. Physical AI is arriving in mobility first — and every one of these stories is, underneath, a story about who holds the data.

2. Mobility is the largest sector on our map

We've now published 435 data holders. Sorted by sector, the distribution is not flat:

• industrial — 213 (49%), but that's an umbrella: plants, grids, pumps, process lines, a dozen unrelated industries in one bucket.
• mobility — 134 (30.8%). One coherent industry. Vehicles, fleets, telematics, EV infrastructure, logistics assets.
• Then a long drop: other (37), healthcare (30), retail (12), finance (7).

Mobility is the single largest coherent concentration we've mapped. And it is not a niche of enthusiasts — it's where the operators are: UK (26), France (24), Canada (22), US (21), Germany (20), Netherlands (7). Contactable operators running real fleets, not institutions.

3. The twist: mobility is the only niche that isn't just a signal

This is the finding that changed how we think about it, and it only appears when you put editions #2 and #3 side by side.

Predictive maintenance is 100% time-series. All 191 holders. Every one. It's a beautifully pure dataset — and that purity is exactly its limit. A turbine's vibration trace tells you about the turbine. Nothing else. It's a signal to forecast.

Mobility doesn't do that. Sorted by modality, our 134 mobility holders break down:

• Time Series — 80 (59.7%) — telemetry, CAN traces, fleet telematics
• Text — 28 (20.9%) — regulatory records, maintenance notes, inspection write-ups
• Tabular — 18 (13.4%) — trips, claims, transactions
• Document — 5, Multimodal — 3

Mobility is the only large niche on our map that is genuinely multimodal. And that isn't noise in our classifier — it's the physics of the thing.

A turbine sits still. A vehicle moves through the world. So its data doesn't just encode the machine — it encodes everything the machine drove through: road surface, gradient, weather, traffic, load, the behaviour of the human at the wheel, the regulation it operated under, the incident it was involved in and the paperwork that followed. A truck is a sensor rig dragging an instrumented probe across the physical world, ten hours a day.

That is why this dataset is different in kind. Predictive-maintenance telemetry lets you predict a machine. Mobility data lets you predict a world — and "predicting the world, one step ahead" is the literal definition of a world model.

You can see the split inside our own map. The purest slice of mobility is what we classify as Mobility Telemetry Datasets — 48 holders, and all 48 are simultaneously 100% time-series and 100% predictive maintenance. That's the part of mobility that behaves like Edition #2: a clean forecasting substrate. Meanwhile mobility is also 31.9% of every predictive-maintenance holder we've mapped (61 of 191) — the largest single sector inside last edition's winning niche.

So mobility does two things at once. It is the biggest contributor to the forecasting niche we already identified — and it carries the multimodal context that pure industrial telemetry will never have. It's the one dataset that is both.

4. Why none of this can be scraped — and why Europe just put a price on it

The reason this data is valuable is the same reason it's hard to get: it never touches the open web. A vehicle's signals live on a proprietary CAN bus, behind an OEM's telematics backend, inside an ELD or a tachograph. There is no public corpus to crawl. The only way in is a licence.

Two things happened in the last year that turned that constraint into a market.

First, the EU wrote the licence into law. Regulation (EU) 2023/2854 — the Data Act — has applied since 12 September 2025. It does something subtle and, for us, decisive: it makes in-vehicle data free to the vehicle's user (Articles 4-5), but when that user directs it to a third party, the data holder is owed compensation that need only be "non-discriminatory and reasonable" and — verbatim — "may include a margin" (Art. 9(1)). The law didn't open the data for free. It opened a priced B2B channel, and said out loud that the price is allowed to carry a margin.

And it drew the boundary in exactly the place that matters for AI. The Commission's own guidance (Notice C(2025) 6119, published the same day) puts raw and pre-processed signals — sensor readings, speed, fuel and battery state, odometer, component status — inside the access right, and puts "inferred or derived" outputs — ADAS object detection, trajectory predictions, risk scores, eco-scores — outside it. Read that twice: the mandate opens the training substrate and leaves the OEM's model outputs proprietary. It is almost as if it were written for the data-licensing question specifically.

Second, privacy law made the front door consent-only. Under the ePrivacy rules, the EDPB treats the connected vehicle as "terminal equipment" — like a phone or a laptop. That triggers a prior-consent requirement to extract data from it, whether or not the data is personal, and it takes precedence over the GDPR's "legitimate interest" basis. You cannot quietly pull data off a vehicle and argue your interest justified it. Someone has to say yes.

Put those together and you get the shape of the whole market: the raw data is legally accessible, explicitly priceable, and gated by consent. That is not a scraping problem. It's a matching problem — connecting the holder who can say yes with the buyer who'll pay, on terms the law has already sketched.

One date makes this urgent rather than theoretical. The Data Act's access-by-design obligation (Art. 3(1)) bites for vehicles placed on the market after 12 September 2026 — six weeks after this newsletter goes out. The channel doesn't just exist on paper; it's about to be built into the metal.

(Two honest caveats, because this is new law. The Commission still hasn't published its method for calculating that "reasonable compensation" — the Art. 9(5) guidelines were in draft as of February 2026 and remain unadopted, so the price is real but unsettled. And a pending "Digital Omnibus" proposes a trade-secret refusal right where data would flow to third countries with weaker protection — a live risk for anyone hoping to license EU vehicle data straight to a US lab.)

5. Who actually pays for it — and it isn't who the headlines say

Here we have to be honest, because the easy version of this story is wrong, and our own research killed it.

The obvious pitch writes itself: self-driving is here, the AV labs are starving for driving data, sell them your fleet's telemetry. We went to check. It doesn't hold.

The frontier labs are data-rich and closed, not hungry and buying. Waymo's June 2025 scaling study runs on an internal dataset of roughly 447,000–500,000 hours of driving — the nearest published comparison tops out around 8,192 hours, about sixty times smaller. That's not a company shopping for data; it's a company sitting on a moat. And Wayve, asked the same question, answers it the other way entirely: its response to the long tail of rare events is synthetic generation, not buying more real miles — it publicly points to data abundance, not scarcity. Pitch "the AV labs need your telemetry" and you'd be inverting what the labs actually say. So we won't.

The real buyers are quieter, and they're already paying. Two of them, on the record:

Insurers. This is the clearest live market for real vehicle telemetry that exists today. Hyundai and Genesis exclusively license their connected-vehicle data to Verisk's Data Exchange, which turns it into filed insurance-pricing models used by major US auto insurers. The reason is boring and durable: the actuarial record is unambiguous. As the US insurance regulators' own research body put it, using actual vehicle operation "has been proven to be significantly more predictive of expected loss costs than proxy variables commonly employed for auto insurance ratemaking." Telematics isn't a nice-to-have for pricing risk; it's simply better — and insurers pay for better.

Everyone who has to grade a battery, a part, or a rule. This is where mobility's multimodality (Section 3) pays off. Real fleet telemetry doesn't just flag a failing part — it estimates the state of health of an EV battery at scale. Geotab's January 2026 study read the health of 22,700+ EVs across 21 models and found batteries degrading ~2.3% per year on average — roughly doubling under heavy DC fast-charging. Peer-reviewed work goes further: a Nature Communications framework estimates battery health across 300 real-world EVs to within a 2.83% average error. A battery you can grade is a battery you can price, warranty, resell, or retire on time — worth money to lessors, insurers, second-life buyers and OEMs alike. And it exists only if someone can reach the real charging-and-driving data underneath it.

That maps precisely onto who our own signals say wants this. Of our 134 mobility holders, the buyers our engine identifies are 45.5% industrial-AI & maintenance vendors, 15.7% RegTech and compliance-AI, 11.9% industrial-AI integrators. Not one is an autonomous-driving lab. The demand is real; it just isn't where the hype points.

Why we're a matcher, not a marketplace. There's a graveyard behind this business, and we'd rather name it than pretend it isn't there. Wejo raised its way to a $657M public listing and ingested trillions of data points from 12 million+ vehicles — then went into administration in May 2023, revalued at under $10M. Otonomo listed at $1.4B and was absorbed for $270M. The pure data-marketplace model — hoard the pipes, resell the exhaust — has been tried at scale, and it broke. What it lacked wasn't data. It was trust and fit: qualification, provenance, a reason for a specific buyer to want a specific holder's data. That missing layer is the entire thing we build.

The bottom line

Edition #2 found the biggest use case on our map. Edition #3 found the one sector that refuses to sit inside a single use case — because a vehicle's data carries a whole world, not one machine's pulse.

Mobility is our largest coherent sector (134 holders, 30.8%), our loudest in signals (34.6% of all we collect), and our only genuinely multimodal one. Underneath it sits a substrate that can't be scraped, that Europe has just made legally licensable with a margin, and that already grades things worth paying for — insurance risk, battery health, compliance. The buyers aren't the ones in the headlines; they're the maintenance, insurance and RegTech vendors our own signals keep naming. And the one model that tried to stand between holder and buyer by owning the data — the marketplace — is a cautionary tale, not a template.

On one side: fleet operators sitting on a licensable asset most don't know is worth anything, six weeks before the law starts building the access channel into new vehicles. On the other: insurers, maintenance-AI vendors and compliance platforms who already pay for exactly this signal. What's missing between them is the layer that qualifies, matches and makes the exchange trustworthy.

That's what we're building.

If you run a fleet, a telematics estate, or EV infrastructure — you are almost certainly sitting on data that is now, by law, yours to license and someone else's to pay for. If you're building AI that needs real-world mobility signal — insurance, energy, maintenance, compliance — this is where it lives. Either way, now's the time to talk.

— Salim Labriki, d-nvest

Methodology note: legal figures cite primary EU texts and Commission/EDPB guidance; where a primary site blocked automated access we cross-checked against independent mirrors and string-matched every quote. We do not publish market-sizing numbers we could not trace to a primary or clearly attributed source — several widely-cited connected-vehicle "market size" estimates did not survive that test and are deliberately absent here. Battery figures are state-of-health estimation (current state), not remaining-useful-life forecasting. A "signal" is a dated external fact (press/registry); a niche is counted as qualified only when ≥3 independent sources converge on it.