Samsung Foundry May Turn Profitable as Early as 3Q26; Nvidia RTX Spark Bets on AI PC
N.R. Finch
Samsung's foundry unit could turn profitable as early as Q3 2026, ahead of market expectations. Meanwhile Nvidia launched RTX Spark to bring AI onto laptops — but ASIC shipments may surpass GPUs next year, signaling a structural shift in the inference market.
Why could Samsung break even ahead of schedule?
The most direct driver: 2nm orders rose 130% year-on-year, while 4nm line utilization jumped on HBM4 — high-bandwidth memory designed for AI chips — base-die orders from Nvidia and AMD.
This means → as orders fill the fabs, depreciation costs spread across more chips. Revenue rises, unit cost falls — the breakeven point pulls forward naturally.
In plain terms = the busier the factory, the less "rent" each chip has to carry. Samsung's fabs are finally busy.
Why are customers coming back to Samsung?
Samsung has improved yields on its 2nm GAA process — gate-all-around, a transistor design where the gate wraps the channel on all four sides, cutting leakage and boosting performance. It is now ramping production for Tesla and other major clients.
Some companies are actively seeking alternatives to TSMC's advanced nodes, giving Samsung a "backup supplier" dividend.
This reflects a deeper signal: AI chip demand is so intense that one foundry's capacity is no longer enough — customers must split orders.
How much can the Texas fab help?
Samsung's $37 billion fab in Taylor, Texas is ramping capacity and should improve cost efficiency over time.
But accounting treatment and early-stage operating costs remain uncertain — new fabs typically front-load depreciation, dragging short-term margins.
This means → Samsung's near-term breakeven hinges on filling existing lines with orders. The new fab's contribution comes later.
What problem is RTX Spark trying to solve?
At Computex 2026, Nvidia unveiled RTX Spark, a processor platform designed to run AI agents — autonomous programs that execute tasks on their own — locally on PCs, without relying on the cloud.
The strategic intent is clear: Nvidia is extending its reach beyond data centers into personal AI computing, while leaning on MediaTek and Intel's PC expertise so it can focus on AI accelerators, autonomous driving, and robotics.
In plain terms = Nvidia is saying AI should not live only in the cloud — your laptop deserves its own "AI brain."
Can RTX Spark actually sell?
Analysts are cautious: laptops with RTX Spark are expected to carry premium prices, potentially limiting mass-market adoption.
The core challenge is that most users already rely on cloud AI services such as ChatGPT and Copilot. The differentiated value of on-device AI still needs market proof.
This means → RTX Spark is more likely to gain a foothold first in high-performance niches — developers, creators — with broader adoption waiting on price declines.
What does the ASIC rise mean for Nvidia?
DIGITIMES analyst Joyce Chen forecasts that AI inference workloads will accelerate ASIC adoption — application-specific integrated circuits, chips custom-built for a single task, highly efficient but not general-purpose. Shipments could surpass GPUs as soon as next year.
Beneficiaries include MediaTek and the Google TPU ecosystem. Separately, network-transmission bottlenecks are creating openings for Marvell and Broadcom in CPO — co-packaged optics, which places optical communication modules right next to the chip to shorten data paths.
This reflects a structural reality: Nvidia's dominance in AI training remains firm, but its GPU share in inference is facing structural dilution. The pace of ASIC substitution and RTX Spark's market traction are the key variables testing Nvidia's next-phase growth thesis.
Content is for reference only, not financial advice.