Qualcomm's New Industrial Chip Bets on -30°C to 115°C Toughness
Qualcomm's IQ-2390 chip trades raw AI horsepower for a -30°C to 115°C rating and deterministic Ethernet, aimed at PLCs, gateways and machine vision.
Qualcomm used the run-up to IFA 2026 in Berlin to introduce the Dragonwing IQ-2390, the first processor in a new “IQ2 Series” of industrial edge-AI chips built around one central trade-off: a -30°C to +115°C operating range and built-in deterministic networking, in exchange for AI compute that is deliberately modest. The chip, announced September 1 alongside a commercial/consumer sibling called the Q-2390, is not a bid for AI horsepower. It is a bid to make edge AI cheap and rugged enough to show up in PLCs, gateways and machine-vision boxes that currently run without any on-device intelligence at all.
Both chips share a common silicon base: a quad-core Qualcomm Kryo CPU built from one Arm Cortex-A78 core paired with three Cortex-A55 cores, clocked up to 1.5 or 1.9 GHz depending on SKU, a Qualcomm Adreno 704 GPU, dual-channel 16-bit LPDDR4x memory up to 1.8 GHz, and eMMC 5.1/SD 3.0 storage — specifications independently documented by CNX Software’s technical write-up, which is where the clock speeds, part numbers and NPU figure below come from; Qualcomm’s own release does not publish that level of detail. Where the two chips diverge is the industrial line’s packaging and I/O: the IQ-2390 pairs that compute with dual Gigabit Ethernet carrying Time-Sensitive Networking (TSN), the IEEE extension that guarantees deterministic, low-jitter timing instead of Ethernet’s normal best-effort delivery — the feature that actually lets a PLC coordinate a conveyor, a robot cell and a bank of sensors on a shared, predictable clock.
On raw AI compute, the honest number is small. CNX Software’s spec sheet lists the onboard Hexagon NPU at up to 1.1 TOPS, and the only IQ-2390 part number that exists — IQ2390-AB — is the variant that includes that NPU, an optional SiFive E61 RISC-V real-time core (up to 600 MHz, 768 KB of memory) and a Hexagon V66 audio DSP. Qualcomm’s own release simplifies this into “both processors feature… a real-time RISC-V MCU,” which glosses over a distinction CNX’s spec table makes explicit: the commercial Q-2390 line also ships as a cheaper “AA” variant that drops the NPU, RISC-V core and audio DSP entirely, while the IQ-2390 is only sold as the fuller “AB” configuration. The -30°C to +115°C rating follows the same pattern — it is not unique to the IQ-2390 silicon, but is shared with the Q-2390’s own AB variant; the Q-2390’s cheaper AA variant is rated to a narrower -30°C to +95°C. For context on what 1.1 TOPS actually means in this market: NVIDIA’s Jetson Orin Nano 2, an entry-level edge-AI module NVIDIA itself pitches at cost-conscious robotics developers, claims 78 TOPS — a company-stated figure, like Qualcomm’s, but roughly 70 times higher. Further up NVIDIA’s own compute ladder, Delta Electronics built its dual-arm robot platform around a 275-TOPS Jetson AGX Orin module. The IQ-2390 is not positioned to compete with either; it is aimed at fixed infrastructure that has never run a neural network locally before, not at systems that need to run one well.
Jeff Arnold, Qualcomm’s VP and GM for Auto Telematics, Industrial and Embedded IoT, framed the launch around accessibility rather than performance: “AI is creating a fundamental shift in how connected devices are designed and deployed, but many product categories still face barriers around cost, complexity and integration,” he said in the announcement carried by WebWire, a wire-distribution copy of Qualcomm’s own release. “With the Dragonwing Q-2390 and IQ-2390 processors, we’re enabling intelligence to scale across a much broader range of devices, from retail systems, smart appliances and consumer robots to industrial controllers, machine vision systems and critical infrastructure.” Qualcomm names industrial automation, oil and gas, utilities and energy systems, HMIs, PLCs, gateways, industrial machine-vision systems — the kind of fixed inspection hardware covered on this site’s own inspection applications page — and building/energy management as the IQ-2390’s target categories, distinct from the Q-2390’s consumer-facing list of retail point-of-sale, kiosks, access control, smart appliances, smart agriculture, home robots, fitness equipment and enterprise terminals.
Two named hardware partners give the industrial claim some early substance. SECO’s Chief Product and Marketing Officer, Lorenzo Veltroni, said the company is building a “Compact Vision 5” machine-vision board and a separate single-board computer directly on the IQ-2390, paired with SECO’s Clea software framework “for edge AI, and long-term lifecycle management.” Engicam CTO Milco Pratesi said his company is developing a system-on-module powered by the same chip to help customers “modernize their industrial products.” Both statements name the IQ-2390 specifically, not the commercial Q-2390 — a useful signal that at least two board vendors are treating this as a real design-in, not just a press-release mention. No pricing has been disclosed for either chip, and no robot manufacturer appears anywhere in Qualcomm’s materials as a named adopter.
The near-term calendar is thin by design: Qualcomm is showing IQ-2390 and Q-2390 system-on-modules on-site at IFA 2026, running an early-access program with board partners now, and targeting evaluation kits for the first quarter of 2027 — over a year before this chip could plausibly appear in a shipping industrial product. That gap, plus the absence of any named robot maker or systems integrator, is the main reason to treat this as an early-stage components story rather than a deployment story.
Sources
- Qualcomm Introduces Dragonwing Q-2390 and IQ-2390 Processors, Expanding Access to Intelligent Connected Devices — WebWire (Qualcomm Technologies press release), Sep 1, 2026
- Qualcomm introduces Dragonwing Q-2390 and IQ-2390 for consumer and industrial AIoT applications — CNX Software, Sep 1, 2026
- Qualcomm introduces Dragonwing Q-2390 and IQ-2390 processors — Robotics 24/7, Sep 1, 2026
Frequently asked questions
Is this a chip for humanoid robots or industrial robot arms? +
No. Qualcomm's own materials name zero robot makers as adopters or partners. The IQ-2390 is aimed at fixed industrial infrastructure — PLCs, HMIs, gateways and machine-vision systems — not at robot controllers or actuators. The commercial Q-2390 sibling chip does list 'home robots' among its target consumer applications, but that is a different SKU aimed at a different market.
How does the IQ-2390 compare to NVIDIA's Jetson Orin Nano 2? +
It is far less powerful on paper. Qualcomm states up to 1.1 TOPS of NPU compute for the IQ-2390, versus the 78 TOPS NVIDIA claims for the Jetson Orin Nano 2 announced in August 2026 — both are the vendors' own figures, not independently benchmarked. The IQ-2390 instead adds a -30°C to +115°C operating range and built-in Time-Sensitive Networking that NVIDIA's module announcement does not emphasize. These are different market tiers, not competing point solutions.
When can integrators actually buy it? +
Not yet, and no price has been disclosed. Qualcomm says evaluation kits are expected in the first quarter of 2027, with on-site demonstrations at IFA 2026 in Berlin and an early-access engagement program already underway with hardware partners including SECO and Engicam.
What does Time-Sensitive Networking actually buy an industrial buyer? +
TSN is a set of IEEE Ethernet extensions that guarantee deterministic, low-jitter packet delivery timing, instead of the best-effort timing standard Ethernet provides. That matters on a factory floor because a PLC coordinating a conveyor, a robot arm and a set of sensors needs those devices to receive commands and readings in a predictable window — not just eventually — to keep motion synchronized.
More news
