Chip Giants Are Pulling Back From These Key Segments

Recently, the World Semiconductor Trade Statistics (WSTS) organization raised its outlook for global semiconductor market growth.

The market is now expected to expand 22.5% year over year in 2025, far above the previous 11.2% forecast. In 2026, growth could accelerate again by 26.3%, pushing the market to US$975 billion—just a step away from the US$1 trillion milestone.

What makes this rebound more meaningful is its quality and breadth.

In 2024, the boom was largely driven by Nvidia-led AI chip momentum and a cyclical rebound in memory, creating a “top-heavy, single-structure” recovery. By 2025, the upturn is becoming more balanced. AI and memory remain the primary engines, but other segments are now showing strong growth as well.

Yet, in the middle of this expansion, semiconductor giants are behaving with surprising restraint.

TSMC is exiting GaN foundry services. NXP is shutting down a once-promising GaN wafer fab. Micron is walking away from consumer storage altogether. Instead of chasing an “everything empire,” the largest players are systematically shedding weight—doubling down on businesses that can endure cycles and deliver long-term returns.

Traveling lighter is becoming a shared strategy across the industry.

The Exit List: Who’s Walking Away?

NXP: Shutting Down ECHO Fab and Exiting GaN 5G Power Amplifiers

In December, NXP announced it will close the ECHO wafer fab in Chandler, Arizona.

In an email to Light Reading, NXP said that due to weak carrier ROI, the pace of 5G deployment has slowed, and global base station rollouts have come in far below early expectations. Given the market reality and a dim recovery outlook, the RF business no longer fits the company’s long-term strategy. As a result, NXP will gradually scale back its RF power product line.

ECHO Fab officially launched in September 2020, when 5G was still surging. NXP invested over US$100 million to upgrade the facility, using advanced 6-inch SiC substrate processes to produce GaN RF devices.

The company positioned it as one of the most advanced facilities of its kind—built specifically to manufacture high-efficiency, high-power-density GaN PA chips, aiming to replace traditional LDMOS technology and become the new “gold standard” for 5G base stations.

But market reality shifted.

Omdia data shows 5G equipment revenue fell from US$45 billion in 2022, declining for two consecutive years—down roughly US$5 billion in both 2023 and 2024. In the 4G era, NXP’s LDMOS PA was close to an industry default, widely supplied to major vendors like Nokia’s Flexi series and Huawei. In the 5G era, however, its response speed and competitiveness clearly lagged.

The closure of ECHO Fab signals NXP’s full retreat from RF power—once a core strength. The fab is expected to complete its final wafer production in Q1 2027.

TSMC: Exiting GaN Foundry Services

TSMC also adjusted its GaN strategy this year.

In July, GaN company Navitas announced that its 650V device products would gradually move foundry production from TSMC to PSMC over the next one to two years. TSMC responded that after a full evaluation—and considering market conditions and long-term business strategy—it plans to phase out GaN business within the next two years.

This is a notable pivot, especially since TSMC previously voiced strong confidence in GaN’s future—at times even more optimistic than for SiC.

TSMC R&D leadership had emphasized a long-term focus on GaN development in compound semiconductors, arguing that market acceptance was improving and more applications would open over the next decade. In 2023, TSMC reportedly held around 40% of global GaN wafer foundry share, forming a “one superpower, two strong followers” landscape alongside Germany’s X-Fab and Taiwan’s Hanlei.

So why exit?

The answer is profitability and prioritization. TSMC is famously margin-driven and maintains long-term targets centered on high gross margin performance. GaN foundry volumes are relatively small, with current 6-inch monthly capacity estimated at only 3,000–4,000 wafers, and a large portion tied up by a top customer like Navitas—limiting revenue contribution and making it hard to meet TSMC’s return expectations.

Even so, TSMC’s step back creates room for second-tier foundries. As TSMC fades out, Navitas quickly partnered with PSMC to push mass production of 8-inch GaN-on-Si.

Micron: Ending the Crucial Consumer Brand to Focus on Enterprise Storage

In December, Micron announced it will exit the Crucial consumer business, including selling Crucial-branded products through major global retailers, e-commerce platforms, and distributors. Micron said it will continue supplying Crucial consumer products through consumer channels until February next year.

Historically, the Crucial brand served DIY users and laptop upgraders through memory modules and SSDs.

Micron EVP and Chief Business Officer Sumit Sadana explained the decision: AI-driven data center growth is generating strong demand for memory and storage, and Micron is making a difficult choice to leave the consumer Crucial business so it can better supply and support its larger strategic customers in faster-growing segments.

Two drivers sit behind the move.

First, consumer storage is low-margin and brutally competitive, dragging down profitability over time. Crucial has brand recognition among enthusiasts, but its positioning is awkward—squeezed between premium niche brands and low-end mass brands. In contrast, data center and enterprise products come with longer contracts, higher ASPs, and more predictable demand.

Second, the AI boom is accelerating demand for high-value products like HBM. Micron is reallocating limited capacity toward higher-return segments. By Micron’s estimates, the global HBM market could grow from US$4 billion in 2023 to US$25 billion in 2025, with explosive compounded growth. Meanwhile, the HBM capacity bundled into each AI server is rising sharply—from 128GB in traditional servers to over 1TB—turning HBM into a major cost and value center in AI infrastructure. On its September 2024 earnings call, Micron CEO Sanjay Mehrotra said quarterly HBM revenue was nearing US$2 billion, implying an annualized run-rate around US$8 billion.

SK hynix: Officially Shutting Down Its CIS Division

Micron isn’t the only memory heavyweight cutting side businesses. SK hynix announced in March that it would shut down its CMOS image sensor (CIS) business division and move several hundred staff into AI memory.

That decision ended a 17-year CIS journey—and marked another contraction by Korea’s semiconductor industry in mid-to-low-end chip categories. SK hynix’s CIS story began in 2008 with the acquisition of Korean design firm Siliconfile. It launched the “Black Pearl” sensor lineup positioned against Sony and at one point entered Samsung Galaxy foldable supply chains.

In 2021, its mobile CIS revenue peaked, with market share reportedly exceeding 7%. But Sony and Samsung’s patent barriers made it difficult to break into high-end CIS. From 2022 onward, revenue declined, hitting a low point in 2024. Under margin pressure, SK hynix couldn’t scale shipments profitably.

With AI fueling massive demand for memory, SK hynix choosing to refocus is hardly surprising. Its exit also opens space for players like Omnivision and GalaxyCore to grow.

Wingtech: Fully Divesting Its ODM Business

This year, Wingtech announced a major asset sale plan. It will transfer 100% equity in multiple entities—along with business asset packages in Wuxi and India—to Luxshare Precision and Luxshare Communications (Shanghai) in an all-cash deal.

Once completed, Wingtech will formally leave behind the ODM business that powered its growth for more than a decade.

ODM had been Wingtech’s revenue backbone. In 2023 and 2024, its product integration segment generated RMB 44.315 billion and RMB 58.431 billion, contributing 72.39% and 79.39% of total revenue.

But since 2022, that business slipped into losses. Net losses were RMB 447 million in 2022 and RMB 1.569 billion in 2023. In the first half of 2024, revenue rose 26.68% to RMB 26.12 billion, yet net loss widened to RMB 850 million.

Wingtech attributed the losses largely to being placed on an entity list, which created major uncertainty in winning new ODM projects and triggered impairment charges on related assets.

After divesting ODM, Wingtech’s semiconductor value becomes far more visible. By Q3, its semiconductor segment delivered RMB 4.3 billion revenue, up 12.20%, with 34.56% gross margin and RMB 724 million net profit. China market revenue hit a record high, up about 14%, with automotive revenue up over 26%, and strong growth in AI servers, AI PCs, computing devices, and industrial applications. China contributed 49.29% of global revenue.

Exiting Is Another Way to Attack

Looking back at 2025, this isn’t simply “the tide going out.” It’s more like strategic recalibration.

These exits are being driven by four structural forces.

1) Structural demand slowdowns.
5G base station construction is cooling, and smartphone camera proliferation is reaching a ceiling. Some tracks are shifting from high-growth into adjustment.

2) A worsening competitive landscape with no profit in sight.
In consumer storage, ODM, and CIS, leaders with cost advantages, stronger supply chains, and local execution are squeezing competitors fast. Latecomers without deep technical or ecosystem moats are stuck with razor-thin margins—or outright losses. Rather than running an endless “supporting role,” companies are choosing to step off the stage.

3) Technology paths are diverging, forcing focus.
In wide-bandgap semiconductors, GaN fits high-frequency fast charging, while SiC is better suited for high-voltage traction and powertrain systems. In memory, HBM and CXL are becoming core AI infrastructure requirements. As paths become clearer, companies must place bets—not cast nets.

4) Capex pressure is forcing strategic contraction.
2nm processes, HBM4, and other frontier bets require massive investment. In an era where capital efficiency dominates, divesting low-return businesses becomes necessary to free cash flow and support core R&D.

Exiting is another way to attack.

In 2025, logic chips are projected to grow 37.1% in revenue year over year. Memory follows at 27.8%, making them the two strongest growth categories. Sensors are expected to rise 10.4%; microprocessors 7.9%; analog 7.5%; optoelectronics 3.7%. Discrete devices may slip 0.4% due to weaker automotive demand.

Looking into 2026, logic and memory are still projected to lead the pack with growth above 30%. WSTS’s forecast points to the new center of gravity: AI servers, data centers, and high-performance computing are driving a structural growth cycle.

In a semiconductor era approaching one trillion dollars, the most decisive offense often begins with the courage to exit.