Satyajit Sinha

In short

• Embedded systems, a $585 billion market in 2025, is undergoing several important structural changes, according to IoT Analytics’ 87-page Embedded Systems Landscape & Trends Report 2026 (published August 2026).
• AI is a major driver of these changes across hardware architecture, software workflows, and security requirements, as intelligence is being distributed across various embedded compute tiers.
• The IoT Analytics team shares 3 opportunities and 3 challenges for developers, along with 5 strategic priorities for embedded systems vendors going forward.

Why it matters

• Workflows are becoming more connected and automated, but not without system and regulatory constraints. Development teams must remain apprised of changes and opportunities to stay competitive.

In this article

• Embedded systems market: The overlooked third pillar of electronics
• 3 key opportunities for embedded systems developers
  • Opportunity 1: Developers can now build edge AI through one connected toolchain
  • Opportunity 2: Developers can now put a general-purpose coding agent to work on embedded code
  • Opportunity 3: Developers can now validate software before the hardware exists
• 3 key challenges for embedded systems developers
  • Challenge 1: Developers can no longer size a design on compute alone
  • Challenge 2: Developers can lose time fitting new silicon into an existing workflow
  • Challenge 3: Developers must now design for security across the product lifecycle
• Analyst opinions: What developers should take forward, and 5 strategic priorities for embedded system vendors
• FAQ about embedded systems
• Further analysis
• Embedded systems market by sector (Insights+ Exclusive)
• Embedded systems competitive landscape (Insights+ Exclusive)
• 18 key embedded systems technology trends (Insights+ Exclusive)
• Embedded systems market segment and competitive landscape deep dives (Insights+ Exclusive)

Embedded systems market: The overlooked third pillar of electronics

The global electronic systems market is divided into 3 major groups.

1. Personal and mobile computing – Draws attention because it sits in everyone’s pocket.
2. Data center systems – Draws attention because AI currently drives it
3. Embedded systems – Least publicly visible of the 3, even though it runs much of the physical world around us: cars, machines, medical devices, or smart meters.

According to IoT Analytics’ 87-page Embedded Systems Landscape & Trends Report 2026 (published in August 2026), spending on embedded systems reached an estimated $585 billion in 2025, roughly a third of the $1.8 trillion global electronic systems market. Personal and mobile computing systems accounted for $850 billion and data center systems for $369 billion.

Currently, embedded attracts the least attention and is also the most fragmented of the 3 groups. IoT Analytics tracks 40 product categories across 5 technology layers within the embedded systems market, from microcontrollers and AI chipsets at the silicon level to industrial PCs, gateways, and embedded operating systems higher up the stack.

Embedded systems market to surpass $800 billion on mobility and IoT growth. IoT Analytics forecasts that the embedded systems market will grow at 5.6% CAGR, reaching roughly $857 billion by 2032. Automotive and mobility, together with industrial and robotics, have accounted for and will continue to account for more than half of embedded systems demand. Embedded systems are also behind each of the world’s 21.1 billion installed connected IoT devices, a base that is set to surpass 50 billion by 2034.

Edge AI deepening electronics content across devices. Aside from rising volume, electronics content for each of these devices and products is also increasing, adding more connectivity options, additional safety, local processing, and local software. Edge AI is one structural piece of that shift. About 1% of global IoT connections carry a dedicated edge AI component in the report model today, a share projected to reach about 25% by 2035.

Growth is not the story here. This is already a large, mature market.

What matters more is the structural change underway inside it.

AI is making embedded systems more distributed and complex. Embedded systems are undergoing several important structural changes across hardware architecture, software workflows, and security requirements. AI is an increasingly important part of the story: More intelligence is being distributed across compute tiers, from constrained sensing nodes and embedded control loops to local edge systems and cloud infrastructure, with each handling a different part of the same workload. The result is a more distributed and more heterogeneous system than embedded designs have typically assumed.

For developers, this cuts both ways:

• Some workflows are becoming more connected and more automated, which lets teams build and validate earlier.
• At the same time, system constraints and lifecycle responsibilities are getting harder to design around.

The report identifies 18 technology trends across embedded hardware, software, and applications. Below, we dive into the 3 opportunities and 3 challenges that matter most to embedded systems developers.

Insights from this article are derived from

Embedded Systems Landscape & Trends Report 2026

An 87-page report examining the embedded systems market, 18 technology trends, and vendor approaches, with a special focus on edge AI.

Request sample

Already a subscriber? View your reports and trackers here ?

3 key opportunities for embedded systems developers

Opportunity 1: Developers can now build edge AI through one connected toolchain

Unified toolchains are simplifying edge AI deployment workflows. Edge AI models can now be built and moved from raw data to a running device through a largely single toolchain, instead of hand-carrying them across separate data-science, firmware, and hardware steps (and tools from different vendors). The payoff is fewer manual handoffs between teams, and in some workflows, data from deployed devices can even feed retraining and optimization before an updated model is pushed back out.

Broader toolchains are closing gaps between models and hardware. Edge AI tools are now extending beyond model inference alone. They can prepare data, optimize and adapt models for specific hardware targets and generate deployment-ready code. Bringing these steps into a single workflow reduces the manual effort required to translate a model into something that can infer within the processor, memory, and software constraints of an embedded device. Developers can enter these workflows at different points, although many toolchains remain tied to specific vendor hardware and deployment targets.

How vendors enable the opportunity. Example: Texas Instruments

US-based semiconductor and embedded processor company Texas Instruments (TI) frames Edge AI Studio as a single environment for the full path to an MCU, covering:

• Data capture and annotation
• Model training and inference preview
• Device targeting
• Generation of source code or a linkable library

A trained model can be imported from ONNX, compiled for the specific device, and dropped into an IDE-based application, such as TI’s CCStudio, through generated C files. For a developer, the effect is a shorter route from trained model to embedded code, with fewer disconnected steps.

At Embedded World 2026, TI framed deployment as a built-in stage of Edge AI Studio, linking model output to MCU inference and a live app demo (source: IoT Analytics at Embedded World 2026)

Opportunity 2: Developers can now put a general-purpose coding agent to work on embedded code

There is an opportunity to offload more tedious, low-level coding tasks to a general-purpose coding agent. Rather than acting as chatbots limited to code suggestions, agents can read actual source code, use build tools, simulate virtual hardware, and interpret runtime output. In practice, that means developers can use natural language to ask an agent to inspect a fault or error, edit the source, rebuild, and test the result on virtual hardware.

These agents become useful when embedded IDEs and virtual hardware give them access to the same source code, build tools, and runtime context that developers use. Interfaces such as MCP servers provide that connection. The Embedded Systems Landscape & Trends Report 2026 also notes a bring-your-own-model approach in some implementations, where the development tool provides access to the environment while the developer supplies the preferred AI model or API key. The capability is still early, but current demonstrations show agents moving beyond autocomplete into debugging and integration work.

How vendors enable the opportunity. Example: Cadence

US-based EDA company Cadence Design Systems’ VLAB virtual development machine, exposed through a VS Code integration, lets an external coding agent access the source code, build tools, and runtime diagnostics of a simulated system. Cadence has demonstrated this with Anthropic‘s Claude debugging a virtual EV charging station: the agent read the runtime output, identified a hardware-register conflict, edited the code, rebuilt the project, and brought the emulated system online. The enabler is the virtual environment and its open interface, not the model itself, which the developer brings.

Cadence’s demonstration at Embedded World 2026 connected a cloud-based AI coding agent in VS Code to a VLAB virtual development machine, Zephyr source code, build tools, and runtime diagnostics to identify and fix an EV charging-station software issue (source: IoT Analytics)

Opportunity 3: Developers can now validate software before the hardware exists

Virtual hardware is removing physical boards from the critical development path. Waiting on physical boards has long gated embedded software work. Virtual ECUs, functional twins, virtual boards, and cloud validation environments reduce that dependency. This allows a team to write and test software while the electronics are still unfinished or scarce. The practical wins are earlier starts, repeatable verification, and far less time queued behind a limited number of hardware benches.

Virtualization is moving beyond the MCU. Developers can now test ECU behavior, PCB-level functions, and system integration before real boards arrive, then repeat those tests in cloud-based continuous integration and testing (CI/CT) environments. This allows more software validation to happen earlier and reduces the amount of integration work left for the physical hardware stage. The strongest evidence is in software-defined vehicles, although the same approach can be applied more broadly to embedded engineering.

How vendors enable the opportunity. Example: Aumovio
Germany-based automotive electronics company Aumovio makes vSDV Composer, a virtual ECU development tool that lets teams build and test ECU software before physical hardware is available. It combines virtual ECUs with plant models, functional twins, and virtual PCB elements, and runs either locally on a developer PC or as a headless build in a cloud-based CI/CT pipeline. Aumovio positions it beyond virtual MCU simulation alone, using PCB- and ECU-level virtualization to move validation earlier and support continuous testing across desktop and cloud environments. The value is the earlier, more repeatable validation, not the digital twin label.

Aumovio’s vSDV Composer visual at Embedded World 2026 showed virtual ECU development spanning desktop, cloud, and CI/CT-based testing environments, reducing dependence on a single physical bench (source: IoT Analytics at Embedded World 2026)

3 key challenges for embedded systems developers

Challenge 1: Developers can no longer size a design on compute alone

Memory constraints can outweigh peak compute in embedded designs. Compute performance alone does not decide whether an embedded workload ships. Constraints such as memory, thermals, bill of materials (BOM), and software overhead can each become the limiting factor, and the current memory squeeze makes that system-level reality hard to ignore.

Memory has become a primary bottleneck for both device performance and BOM cost in some designs. Cloud AI demand for high-bandwidth memory is reshaping supply economics. Additionally, vendors report intense pressure on the availability and pricing of LPDDR memory, with some noting input costs up to 8x over the past year. A processor with high peak compute can still underperform if memory bandwidth or capacity falls short.

Embedded design requires balancing system performance against memory economics. Evaluation therefore is not about peak TOPS alone. A design that works in the lab can become uneconomical if memory pushes up the BOM or the chosen memory type becomes hard to source. Developers can offset part of this through model optimization, such as quantization and memory-aware compilation. Still, those techniques ease the burden rather than remove the underlying limits on memory capacity, bandwidth, availability, and cost.

How vendors see it. Example: Micron

US-based memory manufacturer Micron supplies LPDDR memory used in power-constrained edge designs. It has highlighted that the growing demand for high-bandwidth memory in data centers is putting pressure on LPDDR supply because HBM requires substantially more wafer capacity. Micron frames the core issue as the memory wall: an edge AI compute engine paired with memory that lacks sufficient speed or bandwidth cannot use its full processing capability and can spike power as the system forces data through the channel. The takeaway for developers is that raw compute means little without matching memory, so memory choice cannot be separated from model and accelerator choice.

Micron’s memory portfolio on display at Embedded World 2026 (source: IoT Analytics)

Challenge 2: Developers can lose time fitting new silicon into an existing workflow

Toolchain compatibility increasingly determines silicon adoption speed. New silicon can be technically strong and still cost a developer time. The hold-up is rarely the chip itself. Rather, it is getting it into a workflow the team already knows. When a part does not fit the compilers, debuggers, model formats, and libraries that developers already use, adoption slows as they wrestle with setup and model conversion instead of building. Proprietary onboarding adds friction even when the hardware is strong.

Framework compatibility lowers friction from evaluation to deployment. AI frameworks are where this shows up most clearly today. The report points to PyTorch and exported formats such as ONNX as the common entry points across many edge-AI stacks, so silicon that accepts those inputs spares developers from rebuilding a model workflow for every target. Framework support is only part of the picture, though. Profiling, debugging, deployment, and documentation all shape whether a part makes it past first evaluation.

How vendors are responding to the challenge. Example: AMD

US-based semiconductor company AMD demonstrates the developer experience that alternative AI silicon must match. Vitis AI, its edge AI development platform, lets developers bring models from familiar frameworks, such as PyTorch, and optimize, compile, and deploy them onto AMD embedded hardware. Because the NPU in a standard AMD PC shares the underlying technology of its embedded edge parts, a developer can quantize and compile PyTorch models on their own machine before any target board exists, shortening the verification cycle. That native fit with a standard PyTorch install is the point: it removes onboarding friction, while parts that need separately downloaded environment packages carry a setup penalty that can stall adoption.

The Vitis AI software stack, as provided by AMD. The AMD Quark Quantizer and Vitis AI Compiler sit above the ONNX and VART runtimes, which drive the NPU via a flexible runtime on the AMD Versal platform (source)

Challenge 3: Developers must now design for security across the product lifecycle

Embedded security is becoming a lifecycle responsibility. Shipping a secure device used to generally end at secure boot and a device identity. That is no longer the case. Development teams now have to support security long after launch, including:

• Pushing updates
• Tracking and patching vulnerabilities
• Producing software BOMs
• Monitoring devices in the field
• Protecting models
• Keeping cryptography current
• Holding onto evidence for all the above

For products that stay in service for a decade or more, that is a standing commitment, not a pre-release checklist.

Regulations like the EU Cyber Resilience Act (CRA) turn vulnerability management, security updates, and evidence of compliance into requirements rather than nice-to-haves. Meanwhile, algorithms and security standards can change over a product’s lifetime, so industrial and automotive designs need room to update or replace cryptographic mechanisms later. AI adds one more asset to guard, as models can call for encryption, integrity checks, controlled loading, and secure updates of their own. The practical effect is that security has to be designed into the architecture up front, and the build, release, and fleet pipelines have to preserve compliance evidence for years.

How vendors are responding. Example: Kontron

Germany-based embedded computing company Kontron makes AI Shield, a CRA-oriented runtime security layer that runs independently of Kontron’s own hardware and OS while shipping inside the delivered platform. It watches network flows, panel-level logs, and container behavior, learning an application’s normal behavior over roughly three months before flagging anomalies. Rather than a static signature scan or a bolt-on appliance, Kontron positions it as ongoing runtime monitoring tied to CRA readiness, an example of security moving off the boot path and into the device’s full operating life.

Kontron’s AI Shield dashboard showing CRA readiness moving into runtime monitoring, with network events, active threats, alerts, and blocked items managed from a central control layer (source: IoT Analytics at Embedded World 2026)

Analyst opinions: What developers should take forward, and 5 strategic priorities for embedded system vendors

IoT Analytics Senior Principal Analyst Satya Sinha and IoT Analytics CEO Knud Lasse Lueth prepared the Embedded Systems Landscape & Trends Report 2026. They based their research and insights on conversations with over 50 field experts, screening of over 50 announcements/releases, and observations/conversations while attending conferences such as Embedded World 2026.

To them, the 3 opportunities and 3 challenges are not isolated developments. They are all effects that are related to embedded systems becoming more interconnected and intelligent. Below are broader analyst opinions drawn from the full report, including 5 strategic priorities for embedded systems vendors.

Analyst opinion for embedded systems developers and product teams

Satyajit Sinha is a senior principal analyst in our Hamburg, Germany office. He leads the hardware and connectivity research team, focusing on IoT components, chips, modules, and other hardware, along with IoT connectivity and security.

? View publications
? Follow on LinkedIn

For developers and product teams, the 4 points below are the broader lessons worth carrying forward:

• Developers need to engineer the embedded AI stack as one system. As AI workloads spread across MCU-class nodes, application processors, accelerators, edge systems, and cloud infrastructure, selecting individual components is no longer enough. Hardware, memory, runtimes, development tools, deployment, and model-improvement workflows need to work together.
• For embedded AI, model quality is not enough if the model does not fit the device. Memory, power, latency, thermals, and cost can determine whether an AI workload is deployable. Quantization, model compression, memory planning, and hardware-specific optimization are therefore becoming core engineering tasks, particularly as larger models reach the edge.
• RISC-V should be viewed as an additional architecture option, not simply an Arm replacement story. Its value is strongest where chipmakers need workload-specific customization within heterogeneous processors. However, production adoption will depend as much on software and toolchain maturity as on architecture itself.
• Security can no longer be treated as a hardware feature or a pre-launch checklist. Connected and updateable embedded products require security across hardware, firmware, software, deployment, and long-term maintenance. AI adds further assets to protect, while long device lifetimes also require vulnerability remediation and the ability to update cryptography after deployment.

Analyst opinion on embedded systems vendors

Since founding IoT Analytics in 2014, Knud Lasse Lueth’s focus has been to build a team that produces high-quality research in areas such as IoT, AI, Cloud, and smart manufacturing.

? View publications
? Follow on LinkedIn

5 strategic priorities for embedded systems vendors:

1. Design for heterogeneous architectures. Support deployment across different MCUs, application processors, accelerators, local edge systems, and cloud environments rather than assuming one compute tier.
2. Optimize the whole system. The embedded system you sell requires the right balance of memory usage, power, latency, thermals, connectivity, and application performance. Optimizing for one (say, only peak compute) is not enough.
3. Own more of the development workflow. Reduce handoffs between data science, embedded software, firmware, hardware, testing, and deployment.
4. Treat security as a lifecycle capability. Build updates, SBOMs, vulnerability handling, observability, model protection, and compliance evidence into the product architecture.
5. Build around vertical use cases and partners. Translate horizontal technology into validated architectures for automotive, industrial, robotics, medical, energy, and other regulated or domain-intensive markets.

FAQ about embedded systems

What are embedded systems?

Embedded systems are computing systems built into physical products rather than sold as general-purpose computers, such as running cars, industrial machines, medical devices, and smart meters.

How big is the embedded systems market?

The embedded systems market generated an estimated $585 billion in 2025, about 32% of the $1.8 trillion global electronic systems market, according to IoT Analytics’ Embedded Systems Landscape & Trends Report 2026. That makes it the 2nd-largest pillar of electronics, behind personal and mobile computing ($850 billion) and ahead of data center systems ($369 billion).

How fast is the embedded systems market growing?

The embedded systems market is forecast to grow at a compound annual growth rate of 5.6%, from $585 billion in 2025 to roughly $857 billion by 2032, according to IoT Analytics. Growth is led by automotive and mobility together with industrial and robotics, which account for more than half of demand.

What role does edge AI play in embedded systems?

Edge AI is one structural growth driver for embedded systems, not the whole market. About 1% of global IoT connections carry a dedicated edge AI component today. IoT Analytics projects that share will reach about 25% by 2035. Its main effect is distributing more AI workloads across MCU-class nodes, application processors, local edge systems, and cloud infrastructure.

Can general-purpose AI coding agents be used in embedded engineering?

Yes, general-purpose LLM-based coding agents are beginning to work inside embedded development environments, with access to source code, build tools, virtual hardware, and runtime diagnostics. The capability is early-stage rather than mature or universal.

How is the EU Cyber Resilience Act (CRA) changing embedded security?

The CRA is turning vulnerability management, security updates, and compliance evidence into product requirements rather than optional features, extending embedded security beyond secure boot into full lifecycle protection. The CRA is increasing the need for secure updates, SBOMs, vulnerability management, and evidence of compliance. Long product lifetimes and edge AI add further requirements, such as runtime monitoring, model protection, and cryptographic agility.

Further analysis

In our Insights+ section below, we share several insights from our IoT Analytics’ Embedded Systems Landscape & Trends Report 2026, including a breakdown of the embedded systems market, select market leaders across 40+ product categories, 18 key embedded systems technology trends, and deep dives into the IoT MCUs, cellular chipsets, industrial PCs, and gateways markets.

Click here to continue reading with Insights+

Access key market data for $99/month per user

The Insights+ Subscription unlocks exclusive facts & figures. You will gain access to:

• Additional analyses derived directly from our reports, databases, and trackers
• An extended version of each research article not available to the public

Full report access not included. For enterprise offerings, please contact sales: sales@iot-analytics.com

Share this with others:

[social_warfare]

Satyajit Sinha

Satyajit is a senior principal analyst in our Hamburg, Germany office. He leads the hardware and connectivity research team, focusing on IoT components, chips, modules and other hardware, along with IoT connectivity and security.