IOT News
San Jose IoT News Roundup: AI, Chips, Robotics and Connectivity
Published by IOT San Jose Research & Editorial Team

The pattern across this year's IoT announcements is easy to miss in any single headline. A chip maker adds a neural accelerator to a sub-dollar microcontroller. A robotics company leases a factory-sized campus. A Wi-Fi router for sensors goes on sale for 129 dollars. Each story is modest on its own. Together they describe an industry where intelligence is moving into every layer, from the sensor to the facility, and where South Bay companies sit on both the supply and the demand side. This roundup collects the developments that matter most for San Jose and Silicon Valley readers and says what each one means in practice.
Chips: Intelligence Gets Cheap and Small
The semiconductor story of 2026 is that neural processing units are now routine in low-cost parts. Texas Instruments announced on March 10 its TinyEngine NPU in microcontrollers, with one part priced under a dollar in volume and claims of up to 90 times lower latency and more than 120 times lower energy per inference than similar parts without an accelerator. NXP, STMicroelectronics, Ambiq, and others have comparable parts. Qualcomm launched its Dragonwing Q-2390 and IQ-2390 processors on September 1 with an NPU rated at up to 1.1 TOPS, aimed at appliances, kiosks, and industrial devices. Qualcomm has also set a target of more than 14 billion dollars in IoT revenue by fiscal 2029, up from 6.6 billion in fiscal 2025.
What it means: on-device AI is no longer a premium feature. Product teams should assume a small neural accelerator will be available in most new designs and plan software accordingly. We covered the details in our article on the AI chip race in IoT devices and our guide to smaller, faster, more efficient edge AI chips. The open instruction set RISC-V is rising alongside, a trend explored in our piece on RISC-V in IoT hardware.
Robotics: Machines Become Connected Systems
Robotics and drone companies have grown their Bay Area footprint from under 500,000 square feet in 2020 to more than 7.6 million in 2026, according to JLL figures reported by Bisnow. Figure AI opened a campus of roughly 98,700 square feet on North First Street in San Jose, and Tesla leased 375,000 square feet in Fremont for its Optimus program. Nvidia's Jetson AGX Thor, which became available in August 2025, offers up to 2,070 FP4 teraflops in a 130-watt module for on-machine AI. A constraint appears in the real estate data: pilot production often needs around 4,000 amps, and JLL estimates less than 10 percent of Bay Area industrial supply can provide it.
What it means: robots are becoming fleet-managed, network-dependent systems, and the buildings around them matter as much as the machines. Power, wireless coverage, and facility sensors are now part of the robotics business case. See our article on robotics companies and physical AI systems and our analysis of NVIDIA's physical AI push.
Connectivity: More Choices, Better Matched
Several connectivity developments arrived in 2026. Morse Micro announced general availability of its HaLowLink 2 Wi-Fi HaLow router at CES in January, priced at 129 dollars in the US. KORE and Kigen announced in April that they would roll out SGP.32 eSIM connectivity with multi-network resilience, giving device makers a way to change carriers without touching hardware. Industry reviews report that about 75 percent of US private 5G networks run on the shared CBRS band. Nordic Semiconductor said its nRF92 cellular chip with an on-chip NPU will reach general release in early 2027.
What it means: no single network wins. Enterprises will mix Wi-Fi, private cellular, low-power radios, and long-range options, and the discipline is matching each application to the cheapest network that meets its needs. We compared the options in our private 5G piece, explained the long-range option in our Wi-Fi HaLow article, and covered remote provisioning in our eSIM and iSIM guide.
Software and Platforms: Agents Arrive
AWS published a reference design for a device management agent in October 2025, using natural language and layered guardrails. SecuriThings launched an Agentic Device Orchestrator in March 2026 that lets administrators state a goal and have agents build and run the workflow across mixed device fleets. Qualcomm has been buying embedded AI software companies, including San Jose-based Edge Impulse, to close the gap between chips and deployed models.
What it means: the next layer of the IoT platform may be an agent that operates the fleet, and trust, permissions, and audit trails will decide who adopts it first. See our article on AI agents in IoT platforms and our look at gateways becoming edge AI computers.
Infrastructure and Security
AI racks are pushing density from roughly 20 to 40 kilowatts toward 120 to 140 kilowatts, and data center vacancy in primary markets fell to under 2 percent in 2025, according to CBRE. That is driving demand for liquid cooling instrumentation and AI-driven monitoring. On the security side, STMicroelectronics released post-quantum libraries for STM32 chips and Infineon announced a security controller with a certified library including the NIST ML-KEM and ML-DSA algorithms. See our data center monitoring article and our post-quantum security guide.
| Theme | Headline development | Practical takeaway for South Bay teams |
|---|---|---|
| AI chips | Sub-dollar MCUs with NPUs, 1.1 TOPS industrial processors | Plan for on-device AI in every new design |
| Robotics | 7.6 million square feet of Bay Area space, 4,000-amp power needs | Check power and wireless coverage before piloting robots |
| Connectivity | HaLow routers at 129 dollars, SGP.32 eSIM rollouts, CBRS dominance | Match each application to the cheapest network that works |
| Platforms | Agentic device management, Edge Impulse acquisition | Pilot agents in advise mode with strong audit trails |
| Infrastructure | 120 kW racks, liquid cooling, under 2 percent vacancy | Invest in telemetry before density outruns visibility |
| Security | Post-quantum libraries and certified controllers | Protect roots of trust in long-lived devices now |
Three Stories Behind the Headlines
Some developments matter more than their size suggests. The first is the quiet spread of the NPU. When a sub-dollar part includes one, the question facing hardware teams changes from whether they can afford AI to what they will do with it. The second is the move from selling robots to operating fleets. Companies that hold the software, data, and simulation layers will capture value long after the hardware is sold, which is why power and connectivity are becoming part of the vendor conversation. The third is the shift of security from a software patch to a hardware design decision, visible in post-quantum chips and in European rules that expect secure update paths.
A fourth, less visible trend is consolidation. Acquisitions such as Qualcomm's purchases of embedded AI software firms and NXP's purchase of Kinara show larger companies absorbing startups that solved narrow but critical problems. For buyers, that means checking whether a small supplier's product line is likely to remain independent for the life of your design. For founders, it suggests that a clear technical edge in a specific layer is more valuable than a broad but shallow platform.
Finally, note what is absent. Few announcements this year concern sensors that are radically new, and few concern entirely new networks. Most of the energy is going into making existing parts smarter, cheaper, and easier to manage. That is a sign of a maturing industry, and it favors teams that execute well on integration over teams that chase novelty.
How to Read the News Without Being Misled
A roundup is easy to write and easy to misread, so a few cautions are worth stating. Many figures come from vendors describing their own products, such as speedup multipliers measured against a baseline the vendor chose. Market forecasts vary widely, as seen in estimates of the RISC-V market that differ by nearly a factor of two. Funding totals depend on how a category is defined. And announcements are not shipments: a product announced in 2026 may reach volume production in 2027, as with Nordic's nRF92.
A useful habit is to ask three questions of each story. What was measured, and against what? Who is reporting it, and what do they sell? And what would have to be true for this to matter to my product? Applying those questions turns a stream of headlines into a short list of decisions.
What to Watch Next
- Dragonwing design wins. Whether Qualcomm's new processors appear in shipping industrial and consumer products.
- Private network policy. The outcome of debate over CBRS band arrangements, which affects planning for private cellular.
- Matrix extensions for RISC-V. Their ratification would let open-architecture cores handle more AI work on their own.
- Power availability. JLL warns of scarcity for robotics and data center sites in 2027 and 2028 unless more capacity comes online.
- Regulatory deadlines. European cybersecurity obligations phase in through December 2027, and post-quantum roadmaps keep tightening.
For ongoing updates, browse our IoT news section and the company directory, which lists San Jose firms such as Supermicro, Cisco Systems, and Broadcom.
The Bottom Line
The year's news points in one direction: capability is moving to the edge, networks are diversifying, machines are becoming connected systems, and security timelines are getting shorter. For San Jose and the wider South Bay, that is a favorable setting, because the region builds the chips, writes the tools, and buys the systems. The teams that benefit most will treat each announcement as a question about their own roadmap, not as entertainment.
