Discover the Top 10 Semiconductor Trends in 2026

Executive Summary: What are the Top 10 Semiconductor Technology Trends in 2026 & Beyond?

1. AI Compute & Custom Silicon: Cloud and enterprise spend keeps tilting to accelerators and in-house chips as power and supply become the limiting factors. NVIDIA’s data-center revenue hit USD 39.1B in fiscal Q1’26, +73% YoY.
2. Chiplets, 3D IC, & Advanced Packaging: CoWoS/SoIC capacity is expanding aggressively – industry trackers expect ~75k wafers per month of CoWoS in 2025 and further ramps tied to Blackwell-class GPUs.
3. High Bandwidth Memory (HBM) & Memory-Centric Architectures: HBM is the scarce currency of AI. TrendForce says HBM already represents 20% of DRAM revenue and will exceed 10% of bits in 2025.
4. Automotive Chips: S&P Global expects USD 2000+ silicon content per vehicle by mid-decade. Regulators are mandating safety stacks (EU GSR II in force for new models, US AEB standard phased by Sept 2029).
5. 2 nm Race & Angstrom-Class Roadmaps: TSMC reiterates N2 HVM in late-2025 and A16 (1.6 nm) in H2’26 with backside power (SPR) to lift voltage droop and routing. Intel’s 18A (RibbonFET + PowerVia) targets 2025 with 25% performance or 36% power gains from the previous generation.
6. Supply Chain Geopolitics & Re-Shoring: The US has mobilized at unprecedented scale – about USD 30.6B awarded across 19 firms (TSMC, Intel, Samsung each >USD 6B). Also, semiconductor announcements (while only 5% of reshoring events) drove USD 102.6B in capex, two-thirds of all FDI
7. Photonic & Quantum Integration: NVIDIA’s Spectrum-X Photonics targets 100-400 Tb/s switch throughput with silicon photonics, and industry trackers see CPO high-volume from 2027.
8. Edge AI & Domain-Specific Processors: Microsoft’s Copilot+ baseline is pushing AI PCs and workstations. AI-capable PCs to be ~57% of shipments in 2026, while Counterpoint forecasts >400M GenAI smartphones in 2025.
9. Wide-Bandgap Power: SiC & GaN have crossed from niche to core power architectures across EVs, fast charging, renewables, and data centers. Momentum is visible on both demand and supply: Navitas reported USD 450M in GaN design wins with 50% YoY revenue growth in 2024.
10. Sustainability: The sector’s footprint is already vast, and a single large fab can demand ~1.59M ft³/day of water, generate >5000 tons of waste, and consume >100 000 MWh of energy. Water is the tightest choke point: fabs may need up to 5M gallons/day of ultrapure water (UPW) by 2035; TSMC used 101M m³ in 2023.

Frequently Asked Questions

1. What is the new technology for semiconductors?

Researchers at imec and Ghent University have achieved a breakthrough by successfully stacking 120 alternating layers of silicon (Si) and silicon-germanium (SiGe) on a 300 mm wafer using precise epitaxial deposition techniques. This 3D-stacked DRAM architecture with higher memory density will enable high-performance memory solutions, Gate-All-Around (GAAFET) transistors, and AI and quantum-optimized stacked logic.

2. Who is leading in semiconductor technology?

TSMC remains the world’s foremost chip manufacturer, especially dominant in cutting-edge process nodes like 5 nm and below. In parallel, Nvidia has surged ahead as the most valuable player in the sector – its market capitalization recently surpassed USD 4 trillion amid the growing demand for AI-optimized GPUs.

ASML stands alone as the exclusive global supplier of extreme ultraviolet (EUV) lithography systems, while SK Hynix and Samsung Electronics remain dominant in the memory domain.

Methodology: How We Created the 2026 Semiconductor Trend Report

For our trend reports, we leverage our proprietary StartUs Insights Discovery Platform, covering 7M+ global startups, 20K technologies & trends, plus 150M+ patents, news articles, and market reports.

Creating a report involves approximately 40 hours of analysis. We evaluate our own startup data and complement these insights with external research, including industry reports, news articles, and market analyses. This process enables us to identify the most impactful and innovative trends in the semiconductor industry.

For each trend, we select two exemplary startups that meet the following criteria:

• Relevance: Their product, technology, or solution aligns with the trend.
• Founding Year: Established between 2020 and 2025.
• Company Size: A maximum of 200 employees.
• Location: Specific geographic considerations.

This approach ensures our reports provide reliable, actionable insights into the semiconductor innovation ecosystem while highlighting startups driving technological advancements in the industry.

Innovation Map outlines the Top 10 Semiconductor Industry Trends & 20 Promising Startups

For this in-depth research on the Top Semiconductor Trends & Startups, we analyzed a sample of 2016 global startups & scaleups. The Semiconductor Innovation Map created from this data-driven research helps you improve strategic decision-making by giving you a comprehensive overview of the semiconductor industry trends & startups that impact your company.

Tree Map reveals the Impact of the Top 10 Semiconductor Trends in 2026

The Semiconductors Tree Map highlights the Top 10 Semiconductors Trends in 2026. AI accelerators and custom silicon pull capex toward HBM-rich, chiplet/3D-packaged platforms, while 2-nm/angstrom nodes add PPA gains.

Demand diversifies with automotive (ADAS/SDV), edge AI/NPUs, and early photonics, yet profit pools concentrate in accelerators, memory, and advanced packaging. Meanwhile, SiC/GaN became the power backbone for EVs, charging, renewables, and AI-class data centers.

Geopolitics and re-shoring are fragmenting supply chains into multi-region footprints, and sustainability constraints turn into cost variables.

Global Startup Heat Map covers 2016 Semiconductor Startups

The Global Startup Heat Map showcases the distribution of 2000+ exemplary startups and scaleups analyzed using the StartUs Insights Discovery Platform. It highlights high startup activity in the United States and India, followed by the UK. From these, 20 promising startups are featured below, selected based on factors like founding year, location, and funding.

Want to Explore Semiconductor Innovations & Trends?

Top 10 Semiconductor Trends & Innovations in 2026

1. AI Compute & Custom Silicon Explosion

Capital expenditures for hyperscalers increase dramatically as AI changes the economics of data centers. Amazon, Microsoft, Alphabet, and Meta raised capital expenses by roughly 54% in 2024, adding almost USD 80 billion annually. This demonstrates AI’s dominance in investment priorities.

AI spending in 2025 is expected to range from USD 300 billion, according to Morgan Stanley. HyperFrame Research has revised its estimate by 16% to USD 335 billion.

According to The Guardian, the total AI spending in AI has already surpassed USD 155 billion by the middle of the year. It is anticipated to surpass USD 400 billion in the upcoming fiscal year.

At the heart of this AI computing surge is NVIDIA. Its data center revenue jumped to USD 39.1 billion in Q1 FY26 (ending May 28, 2025), up 73% year-over-year (YoY). Its GB200 NVL72 architecture offers up to 30 times the LLM inference performance compared to H100. Inference-token creation has increased tenfold.

The cooling infrastructure and server demand need to catch up. As funds move from general-purpose systems to GPUs and bespoke ASICs, it is anticipated that shipments of AI servers will increase by more than 20% in 2025.

With racks reaching 130-140 kW and Google saying modern AI facilities are moving to 1 MW per rack designs, usage of liquid cooling is predicted to more than quadruple from 14% in 2024 to 33% in 2025.

At the same time, the footprint of power is growing. According to the IEA, AI will be the main factor driving the increase in data center power consumption worldwide.

Additionally, networking-focused vendors benefit. In the first quarter of 2025, Broadcom reported AI semiconductor revenue of USD 4.1 billion (77% YoY) and over USD 4.4 billion in Q2 2025 (46% YoY). This demonstrates the hyperscaler adoption of bespoke ASICs in conjunction with NVIDIA platforms.

Arm anticipates that almost 50% of hyperscaler compute will be Arm-based by 2025, and IDC predicts that Arm server shipments will increase by 70%.

In 2025, TSMC intends to increase its chip-on-wafer-on-substrate (CoWoS) capacity once again to satisfy the inherent demand for AI accelerators.

Rebellions provides Domain-specific AI Processors

Rebellions is a South Korean startup that offers domain-specific AI processors. The startup builds AI accelerators by bridging silicon architectures and deep learning algorithms.

It uses silicon kernels to change the processor architecture so they are suitable for complex deep-learning algorithms. This way, this specialized AI hardware accelerates machine learning computations, improves performance, and reduces deployment costs.

SEMIFIVE provides a Custom Silicon Platform

SEMIFIVE is a South Korean startup that offers a purpose-built custom silicon platform. It integrates with silicon-proven IPs and optimized design methodologies to improve system-on-a-chip solutions. The platform utilizes continuous integration and deployment for the entire configuration process.

Further, its framework merges customer requirements into the design, simulates the environment, and analyzes potential issues. This enables small businesses to mitigate outsourcing while lowering costs, reducing risks, and improving turnaround time.

2. Chiplets, 3D IC, and Advanced Packaging

Demand for AI and HPC is expected to drive advanced packaging revenue, which surpassed USD 12 billion in Q2 2025 and is expected to surge toward USD 83 billion by 2030.

With a 10.7% compound annual growth rate (CAGR), the larger 3D IC and 2.5D IC packaging market is expected to reach USD 167.57 billion by 2034.

According to Coherent Market Insights (CMI), the global advanced chip packaging market size is expected to reach USD 50.38 billion in 2025 and USD 79.85 billion by 2032.

Capacity increase is led by TSMC. With a roughly 50% CAGR from 2022 to 2026, TSMC’s CoWoS throughput is expected to nearly quadruple to around 75 000 wafers per month in 2025.

Form-factors are growing beyond the constraints of reticles. However, super-carrier interposers that approach the 6X reticle (~5148 mm2) are starting production in 2025. TSMC’s CoWoS-S interposers currently span ~3.3X reticle (~2700 mm2).

This change is supported by NVIDIA’s packaging approach. Blackwell chips are switching to CoWoS-L, according to CEO Jensen Huang, in response to higher-bandwidth, multi-HBM requirements. Packaging is still a bottleneck even after fourfold capacity increases in less than two years.

Amkor intends to spend around USD 850 million on high-density fan-out, SiP, and test in 2025. ASE is increasing its 2025 capital expenditure to approximately USD 5.5 billion as AI strains advanced packaging lines. The company is also expanding its Penang facility from approximately 1 million to 3.4 million square feet to accommodate the demand.

3D stacking and hybrid bonding technologies are also advancing. Toolmakers predict growing demand for hybrid bonding into 2H, while Intel promotes Foveros Direct (Cu-Cu bonding).

TSD Semiconductor builds Advanced Packaging Machines

Chinese startup TSD Semiconductor manufactures advanced packaging machines. The startup’s range of products includes wafer grinding, chemical and mechanical cleaning, lapping, and polishing machines.

They find applications in flip chip assembly, wafer bumping, and production of SIPs. TSD Semiconductor’s surface processing equipment enables chip makers to thin the die for better semiconductor properties and improve electrical execution.

JetCool provides Fluid-to-Package Cooling Technology

JetCool is a US-based startup that offers a fluid-to-package cooling solution. It is a micro-convective liquid cooling technology that uses arrays of fluid jets to cool chip surfaces. In comparison to other techniques, this improves heat transfer for high-power microelectronics.

The direct-to-chip cooling also eliminates the need for thermal pastes and interface materials, saving space for enhanced packaging.

Also, FleX and JetCool will develop liquid cooling-ready servers for AI and high-density workloads.

3. High Bandwidth Memory & Memory-Centric Architectures

High bandwidth memory is quickly taking the lead in terms of both cost and performance in AI systems. Because of its pivotal role in building AI accelerators, HBM’s revenue is expected to double in 2025, reaching nearly USD 34 billion.

HBM’s proportion of DRAM value and capacity is also growing. HBM already makes up around 20% of DRAM sales in 2024, while the share of total DRAM bit capacity rises above 10% in 2025 (from 2% in 2023 and roughly 5% in 2024).

HBM is a key growth engine for the memory sector as a whole. According to Yole Group, the memory market is expected to generate close to USD 200 billion in revenue by 2025. HBM prices were expected to increase by 5-10% in early 2025, and TrendForce projects a total addressable market (TAM) estimate of over USD 36 billion.

SK hyniX shipped 12-layer HBM4 samples in March 2025, surpassing 2 TB/s speeds, while HBM3E 36 GB 12-high entered volume in late 2024 with >1.2 TB/s per stack. Systems are also changing to keep up.

By the end of the year, Micron hopes to have 23-24% of the HBM market, with HBM sales jumping roughly 50% from the previous quarter in Q3 of FY25.

Further, Compute Express Link (CXL) and other memory-centric fabrics are starting to be used in real-world applications. Samsung previously showed 512 GB CXL modules and has started mass producing 128 GB CXL 2.0 DRAM.

UnifabriX develops CXL Products

UnifabriX is an Israeli startup that offers a CXL solution: Memory over Fabrics. It features 32TB of DDR5 DRAM in a 2U chassis and shares memory across servers using CXL/UALINK.

The solution enables a fast, scalable, and efficient AI and high-performance computing (HPC) data center memory infrastructure. This allows businesses to save DRAM usage and costs.

XCENA provides Computational Memory

XCENA is a South Korean company that offers computational memory. It is a CXL-based, memory-centric architecture that embeds processing within memory modules. This brings computation closer to the data.

It leverages thousands of proprietary 64-bit RISC-V cores to offload computation and utilizes vertically optimized on-chip cache to mitigate latency in external memory access. The architecture also enables software-defined allocation of memory resources to minimize idle DRAM and improve system efficiency.

The company also unveiled the MX1 computational memory at the Future of Memory and Storage (FMS) event in August 2025.

4. Automotive Chips: Silicon & ECU BOM ~USD 2.3K/Vehicle

Electronics drive essential automotive operations like software-defined architectures, electrification, and safety compliance. An obvious sign of structural silicon uplift in each unit is the average semiconductor/ECU content per car.

According to S&P Global Mobility estimates, automotive ECUs cost USD 1982 globally on average in 2025 and USD 2256 in North America (+14%).

With a 7.29% CAGR through 2030, the automotive semiconductor industry is expected to reach USD 100.5 billion in 2025 and reach USD 142.87 billion by 2030.

As software-defined/ADAS demand escalates and inventory cycles stabilize, the S&P anticipates even greater growth for automotive semiconductors – 16.5% YoY for 2025-2026.

Global light-vehicle (LV) sales are also predicted to reach 89.6 million units in 2025, establishing a baseline for semiconductor content increases. Vehicle volumes continue to be a pillar of support.

Moreover, platforms for computation are growing quickly. Qualcomm’s Q3 FY25 automotive sales were USD 984 million, up 21% YoY. The company has a USD 45 billion design pipeline, which includes about USD 15 billion in ADAS.

In Q1 FY26, NVIDIA reported USD 567 million in automotive revenue (72% YoY). It was driven by the growth of L2+ platforms and centralized compute.

However, demand is locked in by regulatory requirements. ISA, AEB, lane-keeping, and other requirements are being incorporated into cameras, radar, MCUs, and networking silicon as part of the EU’s GSR (2024-2029).

The architecture is also changing from having several separate ECUs to having a central compute unit together with zonal/domain controllers. In 2025, the automotive Ethernet (for domain connection) market is expected to reach USD 3.36 billion, with double-digit growth anticipated.

As ADAS technologies become more common, model content inflation is continuing, with advanced markets setting the budget for silicon (and ECU BOM) at approximately USD 2-2.3k per vehicle.

Yuntu Semiconductor creates Automotive-Grade Chipsets

Chinese fabless semiconductor startup Yuntu Semiconductor makes automotive-grade chipsets. The startup focuses on integrated circuit design to provide clients with their own chipset solutions.

Its auto-level micro control unit (MCU) chips deliver high stability and security crucial for automotive control. They find application in electrical control units (ECUs), engines, fuel systems, infotainment systems, autopilot systems, and more.

Lidwave develops Automotive System-on-Chips

Lidwave is an Israeli startup that offers automotive SoCs. Its patented sensing architecture is useful for the 3D perception industry. It develops a lightweight system on the chip with time-based sensing.

This approach enables manufacturers to produce LiDAR solutions free of bandwidth limitations. Thus, Lidwave’s 3D sensing technology advances driver assistance systems, making them safe and more reliable.

5. 2 nm Race & Angstrom-Class Roadmaps

2 nm-class logic is entering volume production, driven by GAA transistors, backside power, and evolving EUV strategies.

Early in 2025, TSMC used its N2 (2 nm) process to finish risk production of about 5000 wafers at Baoshan. It yields already in the 60 percent range, which is considerably above mass-production requirements.

Peak production is anticipated to reach 50 000 wafers per month by the end of the year, with further growth anticipated in 2026. Its Angstrom-class A16 (1.6 nm with backside Super Power Rail) and N2P variant (2 nm “Plus”) are aimed toward H2 2026.

Samsung is also getting closer. With yields allegedly increasing from 20-30% to over 40%, Samsung’s Exynos 2600 prototype on 2 nm GAA is currently in the mass-production phase. It will go into full-scale manufacture in H2 2025.

Intel makes progress through backside power and Angstrom. Intel’s 18A node (1.8 nm class) has RibbonFET transistors and PowerVia backside power, and is scheduled for H2 2025. Compared to its predecessor, this architecture provides 25% higher frequency as well as 36% lower power and 30% density improvement.

In Q1 2025, TSMC continued to dominate the foundry market with a share of approximately 67.6%, while Samsung trails at approximately 7.7%.

In Q2 2025, ASML shipped their first High-NA EUV scanner (EXE:5200B), which increased productivity by over 60% and made high-throughput lithography possible. While TSMC continues to use low-NA through A16, Intel intends to implement High-NA for its 14A node in 2027-2028.

AlixLabs advances Atomic Layer Etch Pitch Splitting

Swedish startup AlixLabs develops APS, an atomic layer etch pitch splitting process. This process etches existing nanostructures to split them and achieve sub-10 nm features without extra lithography masks or EUV exposure.

It also integrates with existing production workflows on standard 300mm logic and DRAM wafers. The process also supports GaN and other power electronics substrates. This way, it addresses physical limits and high costs associated with traditional lithography.

Rapidus offers a 2 nm Gate-All-Around (GAA) Transistor

Rapidus is a Japanese startup that offers 2 nm logic chips using a GAA transistor. The startup’s AI-powered design support platform utilizes manufacturing data to guide design decisions. Additionally, design-manufacturing co-optimization aligns design and production in real time to reduce iteration loops.

6. Supply Chain Geopolitics & Re-Shoring

With 2025 seeing a spike in state-led reshoring, nearshoring, and policy-driven reshuffles, geopolitics now dominates semiconductor industry strategy. However, execution risk looms because of labor, utility, and permission restrictions.

In the United States, federal mobilization is unprecedented. As of late 2024, the Department of Commerce has completed USD 30.6 billion in funding across 19 businesses, with TSMC, Intel, and Samsung each earning over USD 6 billion in contracts.

Semiconductor-related announcements made up about 5% of reshoring activity between October 2024 and April 2025, but they also contributed USD 102.6 billion in capital. It was two-thirds of all foreign investment during that time and created over 17 600 new jobs.

There are enormous corporate investments as well. GlobalFoundries is investing USD 16 billion to expand its operations in Vermont and New York. Intel is increasing its operations in Arizona and Ohio, and TSMC is building a USD 40 billion factory in Arizona.

The global fab build-out is still going strong. According to McKinsey, businesses anticipate investing around USD 1 trillion in manufacturing through 2030.

Concern is raised by China’s export limitations. China halted exporting rare earth minerals in April 2025. This upset international supply systems that depended on its 99% processing share of heavy rare earths.

By August 2025, India had approved 10 semiconductor projects, including 3D packaging facilities and high-volume fabs. They were supported by incentives totaling around INR ~1.6 billion.

According to a McKinsey report, Taiwanese mature-logic factories have 35% lower operating expenses and 10% higher construction costs. In contrast, China’s subsidized capital and operating expenses provide benefits of up to 40% and 20%, respectively.

However, there is a risk of execution. There are delays in the project timeframes at Dresden (TSMC) and Magdeburg (Intel) in Europe. The expansions of GlobalFoundries highlight full-stack de-risking times by pairing front-end fabs with backend packaging/photonic facilities in New York.

SourceSentinel ensures Counterfeit Prevention

SourceSentinel is a US-based startup that offers LeadID, a barcode-based data platform that creates a single source of truth for parts throughout their lifecycle.

A barcode captures and conveys essential information accessible to all supply chain partners. They are able to access product-specific details like customer-defined information, counterfeit indicators, and product change notifications.

This way, LeanID enables electronic manufacturers and supply chain partners to ensure precise traceability and just-in-time (JIT) manufacturing in supply networks.

iVP Semi enables Local Semiconductor Production

iVP Semi is a fabless Indian startup that produces locally-made, high-performance power semiconductor modules for EVs, renewables, automation, and consumer electronics. Its product line includes MOSFETs, DC-DC converters, LDO regulators, EV chargers, vehicle control units (VCUs), and more.

The company allows electronics manufacturers to reduce reliance on imports and mitigate associated supply chain risks, costs, and delays.

7. Photonic & Quantum Integration

With the promise of breaking down bandwidth barriers and standardizing quantum technology, optical I/O and quantum-ready semiconductors are moving from lab-scale demonstrations to semiconductor roadmaps.

Co-packaged optics (CPO) and silicon photonics are rapidly entering the mainstream.

NVIDIA demonstrated its CPO-based switches, Spectrum-X (Ethernet) and Quantum-X (InfiniBand), at GTC 2025 as the cornerstone of next-generation AI clusters.

Compared to conventional modules, these switches achieve 10X stronger resilience, 63X signal integrity, and 3.5X better power efficiency. Spectrum-X offers an aggregate bandwidth of 100 Tb/s to 400 Tb/s; Quantum-X offers 115 Tb/s through 144 ports at 800 Gb/s and is liquid-cooled for long-term operation. Both will debut in 2025 or 2026.

Starting with 1.6 Tb/s optical engines, NVIDIA’s roadmap is in line with TSMC’s COUPE platform. It then moves on to 6.4 Tb/s on CoWoS packaging and aims for 12.8 Tb/s within packages. This indicates a wave of integrated photonics in advanced packaging.

With goals to exceed 200 000 hours by the end of 2025, Broadcom reported 50 000 operational hours on 32 CPO systems. LPO-MSA released its first sync specification for 100/200/400/800 G DSP-free optics in March 2025. This demonstrates that early standards are also improving.

To commercialize deterministic on-chip single-photon sources, Sparrow raised EUR 21.5 million (Series A). Likewise, QuiX Quantum raised EUR 15 million to create the first single-photon-based universal quantum computer by 2026.

2025 will be a pivotal year, particularly for the implementation of AI fabric networking. Yole Group projects that the silicon photonic integrated circuit die market will reach over USD 863 million by 2029 (at a 45% CAGR). These market signals confirm an expanding ecosystem.

According to a recent peer-reviewed study, the integration of 300 mm-line superconducting qubits (transmons) into CMOS with coherence times (T1, T2) above 100 µs was demonstrated. This indicates the early viability of large-scale quantum manufacturing.

Concurrently, Quantum Motion’s Bloomsbury chip bridges the gap between quantum and classical integration by combining qubits and cryo-control on a 22FDX node.

NcodiN offers an Optical-Network-on-Chip

NcodiN is a French company that develops NConnect, an optical-network-on-chip. It features a very small laser source and consumes less than 0.1 pJ/bet. Further, the chip supports more than 10 000 devices per square meter and delivers better bandwidth per area than copper.

This allows semiconductor companies to eliminate issues due to copper interconnects and develop chiplet-based architectures efficiently.

Deteqt manufactures Diamond Quantum Sensors

Deteqt is an Australian startup that provides diamond quantum sensors. The startup layers a millimeter-scale diamond crystal atop a custom silicon chip. This converts ambient magnetic fields into high-accuracy, vector-resolved data without cooling or bulky optics.

Moreover, the startup combines the entire sensing and processing electronics into a handheld, low-size, weight, power (SWaP) device. It finds use in navigation, resource sensing, biomedical, and environmental applications.

8. Edge AI & Domain-Specific Processors

Edge AI is now widely used for on-device inference. This change is indicated by silicon-IP revenue.

The demand for chiplets, subsystems, and multi-die assemblies is increasing in AI data centers and sophisticated edge devices. This drove electronic design automation (EDA) and silicon IP revenue to reach 12.8% in Q1 2025, totaling USD 5.098 billion compared to USD 4.522 billion in the same period last year.

AI-powered PCs are starting to become commonplace. According to an IDC forecast, an integrated neural processing unit (NPU) will be included in 57% of PCs shipped in 2026. For example, Intel’s Lunar Lake NPU delivers up to 48 tera operations per second (TOPS), while the Snapdragon X Elite from Qualcomm has 45 TOPS.

These figures establish a performance baseline for enterprise devices and are in line with Microsoft’s Copilot+ requirement of at least 40 TOPS.

Furthermore, the number of “GenAI smartphones” is rapidly growing. According to Counterpoint, there will be more than 400 million of these devices by 2025. This accounts for almost one-third of the worldwide smartphone market.

EDGED creates Neural Network Chips

EDGED is a UK-based startup that provides neural network chips for edge computing. The startup’s architectural approach designs a computation block with matrix and vector operations units. This way, EDGED replaces conventional methods that need to replicate programmable cores multiple times.

It also reduces instruction-decoding computation time and enables the tensor processor units (TPU) to pack data more efficiently. This enhances the responsiveness and performance of TPUs in limited power supply applications.

Anari delivers Personalized Chips

US-based startup Anari offers AI-driven personalized chips. The startup provides reconfigurable chips with more efficient computing than conventional graphics processing units (GPUs) to enable quick infrastructure customization.

For instance, the startup’s chip, Thor X, employs hardware acceleration and machine learning models for faster data processing. This system-on-cloud technology also uses semantic segmentation of cloud data for 3D applications.

This way, Thor X generates accurate information about the images, which finds applications in geospatial intelligence, 3D architecture, digital twins, and the metaverse.

9. Wide-Bandgap Power (SiC & GaN)

Wide-bandgap semiconductors, like SiC and GaN, are moving from specialized to fundamental architectures in various industries.

In terms of GaN growth, Navitas establishes a high water record. With USD 450 million in design wins, 2024 GaN revenue increased by more than 50% YoY. This was mostly due to applications in EVs and data centers.

In H2 2024, GaN and SiC had already started to arrive in data centers. Over USD 30 billion has been invested in SiC manufacturing worldwide by last year.

Further, Infineon creates new volumes for GaN and SiC. It intends to manufacture 300 mm wafer GaN, with the first client samples arriving in Q4 2025. The platform seeks to align with the cost structures of silicon.

Prior to this, Infineon launched its first 200 mm SiC devices in Q1 2025 and implemented a phased ramp throughout Kulim (MY) and Villach (AT).

GaN infrastructure in the US is accelerated by MACOM. With up to USD 70 million in CHIPS Act financing, the corporation announced a USD 345 million, five-year expansion. In Europe, Nexperia expands its WBG capability. To increase SiC and GaN production on 200 mm lines, it will invest USD 200 million in Hamburg.

According to Yole, SiC devices, particularly those connected to EVs, would reach a USD 10.3 billion run rate by 2030 (~20% CAGR). Rapid chargers, industrial, and AI data-center power systems will drive the growth of GaN power device sales.

In 2023, around 28% of battery EV (BEV) traction inverters had SiC installed. The use of 800 V platforms, which have a greater SiC composition, is also increasing. In Q2 2025, the percentage of >=250 kW ports in US fast-charging networks rose from 24% to 38%.

In AI data centers, high-density DC conversion using GaN and HVDC (~800 V) is becoming more popular.

Future pipelines are supported by academic and R&D arms. For example, Nexperia established a professorship in TU Hamburg to accelerate wide bandgap semiconductor research and train next-generation talent for the global energy transition.

EPINOVATECH develops GaN-based Chips

EPINOVATECH is a Swedish startup that makes GaN chips for transistor devices. The startup’s proprietary method, NovaGaN, reinforces silicon wafers at the nanoscale level. Further, it coats the wafers with a thin layer of GaN material.

This improves the thermal conductivity, breakdown resistance, and switching speeds, optimizing chip size and reducing power consumption. As a result, EPINOVATECH’s solution enables easily scalable and cost-effective microchip systems.

NXPEC Technologies builds Advanced Power Electronic Converters

NXPEC Technologies is an Indian company that provides advanced power electronics converters.

Its Si-based chargers support on and off-board use, while the GaN-based on-board chargers feature ultra-compact and lightweight designs. The company’s chargers also feature built-in health monitoring.

These products allow EV companies to mitigate in-house development and accelerate time to market.

10. Sustainability & Carbon Abatement

While the semiconductor industry is approaching a USD 1 trillion valuation by 2030, it also produced about 190 million tons of greenhouse gases in 2024. Therefore, sustainability is a big challenge for the industry.

According to a Veolia report, a large chip facility can require 1.59 million cubic feet of water per day, produce over 5000 tons of trash, and utilize over 100 000 MWh of energy. Every year, 185 million tons of CO2-equivalent emissions are caused by the production of IC alone.

Further, resources are strained by water use. By 2035, up to 5 million gallons of ultrapure water may be needed daily by fabs, while TSMC alone used 101 million cubic meters of water in 2023. 1400-1600 gallons of municipal supply are needed to generate 1000 gallons of ultrapure water.

TSMC installed 4.4 GW of contracted renewable energy by the end of 2024. This is enough to avoid ~5.23 MtCO2e annually. The company also accelerated its RE100 goal to target 60% renewable energy by 2030 and aims for 100% by 2040.

Apple also obtained more than 90% F-GHG abatement pledges from 26 semiconductor vendors.

Hard Blue Si-Carbons provides Recycled Silicon Carbide

US-based startup Hard Blue Si-Carbons makes recycled silicon carbide by converting agricultural residues into semiconductor abrasives.

While companies conventionally convert pure biomass to fuels, the startup recovers silicon carbide from agricultural remains. Further, chip makers chemically fuse the obtained material and carbon to get industrially valuable silicon carbide.

Digitho improves Chip Traceability

Canadian startup Digitho provides die identification as a service for chip traceability. Since current semiconductors contain subcomponents from multiple supply chains, it is challenging to track them.

To overcome this, the startup offers digital lithography that enables manufacturers to identify each chip at the wafer level. It collects historical data for material recycling by combining digital verification technologies.

This way, Digitho promotes a circular economy and mitigates the need for virgin raw materials.

Discover all Semiconductor Industry Trends, Technologies & Startups

The semiconductor industry is integrating digital tools and fabrication technologies, as well as novel materials and design solutions. Further, more and more companies are integrating in-house chip production and re-shoring to address the chip shortage. This will drive innovation towards easily scalable and deployable fabrication units.

The semiconductor industry trends and startups outlined in this report only scratch the surface of the trends that we identified during our data-driven innovation & startup scouting process. Identifying new opportunities & emerging technologies to implement into your business goes a long way in gaining a competitive advantage.