ADTechnology Lands $29M AI Chiplet Order on Samsung 4nm

ADTechnology, a South Korean chip design services company with deep ties to Samsung Electronics, has secured a landmark $29 million (400 billion Korean won) turnkey order from a US fabless company to develop an AI HPC system-on-chip (SoC) chiplet. The design will be manufactured using Samsung Foundry's 4nm process node, with tape-out planned for 2026 and mass production targeted for 2028.

The deal signals growing momentum for Samsung's foundry business in the AI chip race and highlights how smaller design service firms are becoming critical enablers in the global semiconductor supply chain.

Key Facts at a Glance

• Deal value: $29 million (400 billion KRW) for a full turnkey chiplet design
• Process node: Samsung Foundry 4nm
• Technology stack: Next-generation HBM memory + 2.5D heterogeneous integration advanced packaging
• Timeline: Tape-out in 2026, volume production in 2028
• Client: An undisclosed US fabless semiconductor company
• Additional partnership: ADTechnology also partnered with US firm Kenyi in April 2025 for edge server HPC solutions

Inside the $29 Million Turnkey Chiplet Deal

The order represents what the industry calls a 'turnkey' engagement — meaning ADTechnology will handle the entire chip design process from architecture through physical design, verification, and tape-out preparation. The client, an unnamed American fabless firm, will receive a production-ready AI HPC SoC chiplet without needing to build its own design infrastructure.

This model has become increasingly popular among AI startups and mid-size fabless companies that lack the hundreds of engineers typically required to execute a complex chip design on advanced process nodes. By outsourcing to specialized design service providers like ADTechnology, these firms can bring custom AI silicon to market faster and at lower upfront cost.

The $29 million price tag, while substantial for a design services contract, is modest compared to the total cost of developing and manufacturing a cutting-edge AI chip. Full development costs for advanced node AI processors can easily exceed $500 million when factoring in EDA tools, IP licensing, mask sets, and engineering resources. ADTechnology's turnkey model effectively democratizes access to Samsung's most advanced manufacturing technology.

Samsung 4nm and Next-Gen HBM: A Competitive Stack

The technical specifications of this chiplet reveal an ambitious design targeting the high-performance computing (HPC) segment of the AI market. Samsung's 4nm process node offers competitive power efficiency and transistor density, positioning this chip against designs manufactured at TSMC — the dominant foundry for AI accelerators from NVIDIA, AMD, and Broadcom.

Samsung has been aggressively courting AI chip customers for its foundry division, which has historically trailed TSMC in both yield rates and customer wins. Landing an AI HPC chiplet design through ADTechnology represents a meaningful data point in Samsung's efforts to close the gap.

The integration of next-generation HBM (High Bandwidth Memory) is particularly noteworthy. HBM has become the defining memory technology for AI workloads, with demand surging as large language models and training clusters require ever-greater memory bandwidth. The 'next-generation' designation likely refers to HBM4 or an advanced variant of HBM3E, both of which Samsung is actively developing.

The use of 2.5D heterogeneous integration advanced packaging further underscores the design's sophistication. This approach places multiple chiplets — potentially combining compute, memory, and I/O dies — on a silicon interposer, enabling higher bandwidth and lower latency between components. It mirrors the packaging strategies used in NVIDIA's H100 and AMD's Instinct MI300 series.

ADTechnology's Growing US Partnership Portfolio

The chiplet order is not an isolated event. In April 2025, ADTechnology announced a separate partnership with Kenyi, a US-based technology firm, to develop edge server HPC solutions. This collaboration combines ADTechnology's ADP620 — a high-performance CPU design targeting 2nm process technology — with Kenyi's data processing unit (DPU).

This dual-track strategy reveals ADTechnology's ambitions:

• Cloud/data center AI: The $29 million chiplet order targets large-scale AI HPC workloads in data centers
• Edge AI computing: The Kenyi partnership focuses on bringing HPC capabilities to edge servers
• Advanced node expertise: The company is simultaneously working on 4nm (production) and 2nm (development) designs
• US market penetration: Both deals involve American partners, reflecting the company's westward expansion

The ADP620 CPU design at 2nm is especially ambitious. Only TSMC and Samsung currently have 2nm process roadmaps, and production-ready 2nm chips are not expected until 2025-2026 at the earliest. ADTechnology's willingness to design at this node suggests deep confidence in Samsung's 2nm GAA (Gate-All-Around) transistor technology.

Why This Matters for Samsung's Foundry Ambitions

Samsung Foundry has faced well-documented challenges in competing with TSMC for high-profile AI chip orders. NVIDIA, Apple, Qualcomm, and most leading AI chip designers have consolidated their manufacturing at TSMC, leaving Samsung with a smaller share of the most lucrative advanced-node business.

ADTechnology's order matters for several reasons:

• Validation of Samsung 4nm for AI: A US fabless company choosing Samsung's 4nm for an AI HPC chip suggests improving confidence in Samsung's process maturity
• Ecosystem development: Design service companies like ADTechnology serve as force multipliers for Samsung's foundry, bringing in customers who might not engage Samsung directly
• Advanced packaging capability: The 2.5D integration requirement demonstrates Samsung's packaging services are competitive enough for demanding AI workloads
• Pipeline visibility: A 2026 tape-out and 2028 production timeline gives Samsung a concrete AI chip in its manufacturing pipeline

Compared to TSMC's dominance in AI chip manufacturing — where the Taiwanese foundry produces virtually all of NVIDIA's AI GPUs and the majority of custom AI accelerators for Google, Amazon, and Microsoft — Samsung's wins remain smaller in scale. However, each successful AI chip project builds Samsung's track record and reference design portfolio.

The Rising Importance of Chiplet Architecture in AI

The fact that this order specifically involves chiplet design rather than a monolithic SoC reflects a broader industry trend. Chiplet-based architectures have become the preferred approach for AI processors because they offer several critical advantages over traditional single-die designs.

First, chiplets improve manufacturing yield. A large monolithic AI chip manufactured at 4nm has a higher probability of defects than smaller individual chiplets. By breaking the design into smaller dies and connecting them through advanced packaging, manufacturers can achieve better overall yield and lower per-unit costs.

Second, chiplets enable mix-and-match flexibility. Different functional blocks — such as compute cores, memory controllers, and I/O interfaces — can be manufactured on different process nodes optimized for each function. This heterogeneous approach maximizes performance while controlling costs.

Third, chiplets accelerate time-to-market. Design teams can reuse proven chiplet IP across multiple products, reducing the engineering effort required for each new chip variant. For AI startups racing to bring custom silicon to market, this modularity is invaluable.

AMD pioneered this approach commercially with its EPYC server processors and MI300 AI accelerators. Intel has embraced chiplets with its Ponte Vecchio and Falcon Shores AI chips. NVIDIA's next-generation Rubin platform is also expected to leverage chiplet architectures. The ADTechnology order shows this design philosophy is now filtering down to smaller, specialized AI chip companies.

What This Means for the AI Chip Market

The deal carries broader implications for the AI semiconductor ecosystem. The fact that a US fabless company is partnering with a Korean design house to manufacture on Samsung's foundry — rather than defaulting to TSMC — suggests the market is actively seeking alternatives to reduce supply chain concentration risk.

Geopolitical tensions surrounding Taiwan have made TSMC's dominance a strategic vulnerability for the AI industry. Companies developing AI chips are increasingly motivated to qualify Samsung as a second source, even if TSMC remains their primary foundry partner. Orders like ADTechnology's $29 million deal help Samsung build the process design kits (PDKs), reference flows, and engineering expertise needed to attract larger customers.

For AI startups and mid-tier fabless companies, the availability of turnkey design services at advanced nodes lowers the barrier to custom silicon. Instead of competing with NVIDIA and AMD on general-purpose AI accelerators, these firms can develop specialized chiplets optimized for specific AI workloads — from inference at the edge to training in the cloud.

Looking Ahead: Timeline and Market Impact

The 2026 tape-out and 2028 mass production timeline places this chiplet squarely in the next wave of AI hardware. By 2028, the AI chip landscape will look dramatically different from today:

• NVIDIA will likely be shipping its post-Rubin architecture
• AMD will have iterated multiple times on its MI-series AI accelerators
• Custom silicon from Google (TPU), Amazon (Trainium), and Microsoft (Maia) will be in their 3rd or 4th generation
• Samsung Foundry aims to be manufacturing at 2nm GAA by then
• HBM4 memory will be in volume production, offering dramatically higher bandwidth

ADTechnology is positioning itself at the intersection of these trends — offering design services that bridge the gap between ambitious AI chip concepts and manufacturable silicon. With both a 4nm AI HPC chiplet in its pipeline and a 2nm CPU design in development, the company is betting that demand for custom AI silicon will only accelerate.

The $29 million order may be modest by semiconductor industry standards, but it represents a meaningful proof point: Samsung's foundry ecosystem is capable of attracting and executing on AI chip designs, and specialized design service companies are emerging as vital intermediaries in the global AI hardware supply chain.