Who Is the Connected Vehicle Actually Connected To?

Global cellular IoT data traffic is projected to reach 218.6 exabytes annually by 2035, according to Omdia's April 2026 forecast. The number itself is significant, but the composition is what demands corporate attention. Of that total, 135.4EB (62%) will originate from a single sector: Automotive. That is a fourfold increase from the 30.7EB generated by connected vehicles in 2025, growing at roughly 16% compound annually for a decade.

What drives that growth? In-vehicle infotainment, primarily. Video and audio streaming inside the cabin, combined with firmware over-the-air updates for software-defined vehicles, generate data volumes that dwarf every other IoT category. A connected car is no longer just about a sensor reporting GPS coordinates in kilobytes. It is a broadband endpoint consuming and generating gigabytes per month, every month, across its entire service life. Manufacturers from Volkswagen to BYD are now shipping vehicles whose primary value proposition is software, and software requires persistent cellular connectivity to stay current, safe, and commercially relevant.

The regulatory picture has kept pace with the technical one, if not faster. Since March 2025, a US Department of Commerce rule on Vehicle Connectivity System supply chains has prohibited, on a staggered schedule through model year 2030, the integration of hardware and software produced by entities under PRC or Russian jurisdiction. Software restrictions apply from model year 2027. For manufacturers sourcing components internationally, compliance is a design parameter. The EU Data Act, with design obligations taking effect in September 2026, adds a parallel obligation that requires connected products to enable user access to vehicle-generated data on fair terms. Across Asia-Pacific, Cullen International's 2026 analysis confirms that roaming-based IoT connectivity models face intensifying national scrutiny: China and India prohibit permanent roaming outright, while Australia, Japan, and Aotearoa permit it conditionally.

These regulatory pressures converge on a market where Asia and Oceania already generates the majority of global cellular IoT traffic. China alone had 4.486 million 5G base stations deployed by mid-2025, accounting for 35.3% of its total mobile infrastructure, with 5G-Advanced already live in over 300 cities. The Mordor Intelligence estimate of the Asia-Pacific connected cars market puts it at 35.52 billion USD in 2025, on a trajectory toward 77.08 billion USD by 2030 at a 16.76% annual growth rate. Five-generation hardware is moving quickly too: Quectel's AR588MA, launched at CES 2026, became the first 5G-Advanced automotive-grade cellular module, supporting 9.0Gbps downlink and integrating satellite communication fallback through 3GPP Release 18's non-terrestrial network specifications.

Two additional traffic categories identified by Omdia's research had no measurable presence in prior forecasting cycles. The first, remote vision, covers cellular-connected cameras deployed in mobile or hard-to-reach contexts like delivery robots, agricultural machinery, construction equipment, etc. where wired backhaul is unavailable. These endpoints generate persistent uplinks that reverse the traditional asymmetry of cellular networks designed for consumer download-heavy usage. The second, agentic AI, introduces machine-to-machine traffic generated by autonomous AI agents coordinating across IoT endpoints without human mediation. The Wevolver 2026 Edge AI Technology Report describes this architecture, here is what is says: Edge-based agents executing perception, reasoning, and action cycles locally, communicating laterally with peer systems, not routing through central cloud infrastructure.

It seems the practical consequence for corporate strategy is a separation that network planners and procurement teams have not yet fully internalized. Connected vehicles generate roughly 113GB per connection per year. All other non-automotive IoT connections average approximately 17.7GB. That is a 6.3-times differential, and it means that pricing models, capacity investments, and service architectures designed for legacy IoT sensors are fundamentally mismatched to the automotive segment. The eSIM provisioning layer is catching up: GSMA SGP.32, the standard enabling over-the-air SIM profile management for long-lived IoT devices, is transitioning from roadmap to early commercial deployment, with eSIM hardware shipments growing 32% year-on-year in 2025. For a vehicle expected to stay in service for 10 to 15 years and potentially cross multiple national borders, remote re-provisioning of cellular connectivity is not optional.

Now, what is consistently absent from the corporate IoT conversation, at the precise moment when Asia and Oceania dominates global traffic generation, is the linguistic layer. The Pacific Islands region alone hosts approximately 800 languages, which represent roughly one-eighth of global linguistic diversity, concentrated among just 12 million consumers. IoT interfaces (voice alerts, infotainment dashboards, firmware update flows, telematics reporting tools) are built for mainstream languages like English, Mandarin, Spanish, and French. But Sāmoa's ICT Policy 2025–2030, for example, explicitly frames local-language digital services as a national priority. In fact, the ITU Smart Villages and Smart Islands initiative is actively deploying IoT infrastructure across Fiji, Sāmoa, Tonga, the Solomon Islands, and Vanuatu. Yet the language technology to make those deployments usable by the populations they target is rarely within scope.

That said, the technical gap is narrowing on the research side. Meta's old NLLB-200 model (superceded by the broader Omnilingual Machine Translation project since March 2026) includes Samoan, Tongan, Fijian, Māori, and Hawaiian in its 200-language machine translation architecture, even if the output is far from reliable (both NLLB-200 and OMT). Microsoft Research's Paza project, released in February 2026, shows very promisingly that fine-tuned ASR models for low-resource languages can be built with minimal data, tested with real users on everyday hardware, and iterated based on community feedback. Edge hardware capable of running localized speech recognition offline is available at price points suited to solar-powered island deployments. The precedent for sovereign language data governance is established, through Intellectual Property tools like Te Hiku Media's Kaitiakitanga License, which ensures community ownership of language data and governs its use in AI development.

For localization departments and public-benefit corporations working in or through the Asia-Pacific region, the question is no longer whether linguistic localization of IoT systems matters. For, regulatory compliance in the vehicle connectivity space now requires supply chain due diligence, declarations of conformity, and market-specific design decisions in which language and interface are embedded from the start. A connected vehicle dashboard built for the Filipino or Kiwi market cannot be retrofitted with a language layer after the product cycle closes. Neither can a disaster early-warning IoT system, nor an agricultural sensor network, nor any platform where the end user's ability to receive and act on information is the entire point.

At Huri Translations, we think the firms that will hold durable positions in Asia-Pacific IoT markets are those that consider Pacific language consultation as a design input, not a post-launch concession as we have seen so many times. The difference between those two postures is, in practice, the difference between a product that works and one that sits unused.