Intel Offers Korean AI Chip Firms Custom Design, Packaging and Foundry Routes

TL;DR: Intel’s 2026 AI Summit Seoul, held at COEX on 12 November 2026, was an Intel-hosted event, not a Korean government summit. Marcus Lawrence, vice president in Intel’s Central Engineering Group, offered Korean AI chip companies five routes: turnkey development, co-design around customer IP, migration from another foundry to an Intel process, advanced packaging, and custom variants of existing Intel products. Intel disclosed no process node, packaging technology, capacity quota or price for Korean customers. Korea’s own on-device AI chip programme, announced the same day by MOTIE, routes prototype priority to Samsung Foundry.
Some circulating summaries describe the event as a government or third-party AI summit at which Intel opened its advanced packaging and foundry capacity to Korean chipmakers. The reporting does not support that framing. Intel organised the summit itself, the offer was a service menu rather than a capacity allocation, and the edge-AI supply-chain argument is an interpretation of Intel’s separate Korea messaging, not a stated objective of the foundry offer. This article separates what Intel and Korean officials said from what analysts and summaries have added.
What did Intel offer Korean AI chip companies in Seoul?
Intel offered service routes rather than fab capacity: turnkey chip development, co-design around the customer’s own IP, migration of an existing design from another foundry to an Intel process, advanced packaging, and custom variants of existing Intel products. Lawrence said Intel can join a project at the concept stage and combine different processor architectures in one package.
- MoneyToday reported on 12 November 2026 that Intel presented five cooperation models and cited Korea’s memory position: Samsung Electronics and SK hynix together make about two-thirds of the world’s memory, and Korea’s October semiconductor exports rose 93.7% year on year to a record.
- Businesskorea quoted Lawrence, head of custom silicon market strategy at Intel’s Central Engineering Group: Intel integrates processors of different architectures with advanced packaging and provides design and foundry services from the concept stage to the product.
- TheElec reported that Intel Korea president Tae-Won Bae launched the K-AI Full-Stack Alliance with 23 initial Korean participants, and that Intel Korea signed a memorandum of understanding with Seoul AI Hub.
Which five collaboration models did Intel put on the table?
The five models are turnkey development, joint design using customer IP, process migration from another foundry, advanced packaging, and custom derivatives of existing Intel products. They differ in who owns the architecture and how much of the design Intel performs. Intel published no pricing, minimum volume, node menu or schedule for any of the five.
| Model | Intel’s role | Customer input | Public terms |
|---|---|---|---|
| Turnkey | Architecture, design, manufacturing, packaging, test | Product specification and market requirements | Not disclosed |
| Co-design | Joint design around customer IP | Proprietary IP and architecture decisions | Not disclosed |
| Process migration | Port an existing design to an Intel process | Design database, licensed IP, verification data | Not disclosed |
| Advanced packaging | Assemble die into a package and test it | Die, known-good-die data, interface definitions | Not disclosed |
| Custom derivative | Modify an existing Intel product | Application requirements and volume estimate | Not disclosed |
Intel’s public packaging portfolio includes EMIB, which connects chiplets inside one package, and Foveros, which stacks die vertically. The Seoul presentation did not say which of these technologies would be used for Korean customers, and Intel’s own third-quarter remarks point to EMIB as the line where demand is running ahead of supply.
Did Intel open fab capacity to Korea, or only a service route?
Intel opened a service route, not a capacity allocation. No executive quoted in Korean coverage named a node, a packaging technology, a wafer or panel quota, a price or a start date for Korean customers. Korean firms can contract Intel’s design, packaging and foundry teams, but each project remains subject to commercial negotiation and Intel’s own capacity.
- Intel’s third-quarter 2026 prepared remarks, published 29 October, say 18A yields improved at a double-digit rate and wafer output exceeded expectations, 18A-P remains on track for launch by year-end, and the 14A 0.9 PDK was released to select customers in October. CEO Lip-Bu Tan repeated that Intel will not disclose customer details.
- The same remarks say an industry-wide shortage of advanced organic substrates for EMIB packaging is limiting Intel’s ability to scale output, and that New Mexico and Malaysia are adding capacity to support advanced packaging ramps next year.
- The Edge Malaysia reports that Intel’s US$7 billion Penang complex contains a 710,000 sq ft cleanroom, and that its first phase is expected to begin advanced packaging operations later in 2026 after earlier delays.
That combination matters for judging the Seoul offer. Intel is selling access to a packaging line that it says is supply-constrained, and to process nodes whose external customer list it refuses to publish. A Korean start-up that signs a design agreement therefore buys a place in a queue, not a reserved block of wafers.
How does Intel Foundry compare with Samsung Foundry for a Korean AI chip team?
On public information, Intel offers a wider menu of design and packaging services outside Korea, while Samsung offers proximity, a 2nm-class process and priority access to Korea’s government prototype programme. Neither company has published a Korean AI customer contract from this summit. Node names are marketing labels, so 18A, 18A-P and SF2 cannot be ranked by name alone.
| Dimension | Intel Foundry | Samsung Foundry |
|---|---|---|
| Public leading-edge position | 18A in production; 18A-P due by year-end 2026; 14A 0.9 PDK released to select customers in October 2026 | SF2, a 2nm-class process; 21 government-backed prototype runs allocated for 2027, three at 2nm |
| Advanced packaging | EMIB, EMIB-T and Foveros family; capacity being added in New Mexico and Malaysia | 2.3D and 3DIC packaging offered alongside memory and logic integration |
| Korean policy access | No allocation announced; summit participation was commercial | Priority for 21 prototype production requests in 2027 under the MOTIE programme |
| Named Korean AI customer | None announced | None announced at the summit |
| Quota for Korean AI firms | Not disclosed | Not disclosed beyond the prototype programme |
Samsung’s advantage is structural rather than technical: a Korean fabless company can walk a design into a domestic foundry, use the government prototype slots, and keep packaging and memory supply in the same time zone. Intel’s advantage is the breadth of the menu, because few foundries will co-design a chip, assemble a package from die built elsewhere and sell a modified catalogue product through one contract.
Is the offer really about edge and on-device AI processors?
Not in Intel’s own wording. The custom-silicon and foundry offer was presented for AI chips generally. Intel’s Korea executives separately named sovereign, personal and edge AI as local opportunities, and a Korean ministry programme funds on-device AI chips. Linking Intel’s foundry routes to edge supply-chain resilience is analysis, not a disclosed Intel commitment.
- TheElec reports that Santosh Viswanathan, Intel’s Asia-Pacific regional head, named sovereign AI, personalised AI and edge AI as three areas where Korea could capitalise, and that Dan Rodriguez, executive vice president and general manager of Intel’s Edge Computing Group, presented Core Ultra Series 3 as an SoC for edge customers.
- ChosunBiz reported Intel’s case study with Samsung Medison, where ultrasound image measurement using Core Ultra and OpenVINO raised AI inference throughput 4.2 times against the previous method without a separate GPU. The figure is Intel-supplied and has not been independently tested.
- The Ministry of Trade, Industry and Energy announced a separate on-device AI chip programme on the same day, which is the clearest Korean policy signal in this story rather than anything said on the Intel stage.
Read together, the two events describe a market rather than a deal. Korean fabless companies are being funded to build on-device AI chips, and Intel is offering them a way to design, package and manufacture those chips without building a full internal flow. Whether any of them chooses Intel over Samsung is unknown, and no company at the summit announced such a decision.
What is Korea’s own on-device AI chip programme doing?
MOTIE announced an 800.2 billion won programme on 12 November to develop eight on-device AI chips and two design IP assets by 2030. Samsung Foundry gets priority for 21 prototype production requests in 2027, three of them on a 2nm process. The programme gives Korean fabless firms a domestic prototype route that does not depend on Intel.
- Businesskorea reported the 800.2 billion won budget, roughly US$600 million, for eight on-device AI chips and two semiconductor design IP assets by 2030.
- Aju Press reported that Samsung Electronics’ foundry division will handle 21 prototype production demands in 2027, including three at 2nm and four at 4nm, aligned with its existing production schedule. Businesskorea adds three runs at 5nm, three each at 8nm and 14nm, and five at 28nm.
- Etnews reported that the government and public bodies at the Intel summit were the Ministry of SMEs and Startups, NIPA, the Korea Startup Promotion Agency and Seoul AI Hub, not MOTIE. The two events were separate.
The policy detail explains why the Seoul offer is narrower than it first appears. Korea is already paying for prototype silicon and directing that work to Samsung. Intel’s opening is therefore aimed at companies that want something the domestic programme does not cover: custom architecture work, packaging of die from multiple sources, or a second-source path for products intended for export markets.
What does this change for plant-level edge systems?
Nothing changes this quarter. Intel’s packaging expansion depends on substrate supply and on capacity arriving in New Mexico and Malaysia next year, so a new Korean foundry route affects processor roadmaps before it affects installed equipment. Existing edge controllers, gateways and industrial Ethernet modules keep their qualification cycles and spare-part lists unchanged.
For plants already running Intel-adjacent edge hardware, the practical questions are still interoperability and spares. A Modbus TCP gateway is qualified against the controller firmware on the machine, not against a foundry roadmap. The same applies to an 8-port industrial Ethernet switch, a fixed-mount barcode scanner on a packaging line, or a PROFINET RT communication module inside a drive cabinet. Intel’s summit changes what may be designed in 2028 and later; it does not change acceptance testing for equipment installed today.
Two existing Autronexa articles cover the same industrial pressure from different angles: servo loop timing and fieldbus jitter on the machine side, and Emerson’s semiconductor and software results on the process-automation side. Both show that edge performance is decided by integration and qualification, not by the process node of the processor inside the box.
What remains undisclosed?
Six items were missing from every account of the Seoul session, and each one determines whether the offer is material for a Korean chip company.
- The process node offered to Korean customers, and whether early projects would run in Oregon, Arizona, Ireland or another site.
- The packaging technology and package size limits, including whether EMIB, EMIB-T or Foveros would be available and at what yield data.
- Any wafer, panel or engineering-support quota, and the queue position a new Korean customer would receive.
- Pricing, minimum volume, mask and design-service costs.
- The names of companies that have signed, beyond the 23-member alliance list.
- Export-control treatment for chips designed in Korea, packaged in Malaysia or New Mexico, and sold into other markets.
Until those points are published, the correct reading is that Intel has opened a conversation, not a production line.
Did Intel promise packaging capacity to Korean customers?
No. No quota, reservation or delivery date was announced. Intel’s third-quarter 2026 remarks say the company is working with its supply chain to secure advanced packaging capacity, and that advanced organic substrate shortages are limiting EMIB output. The Seoul presentation offered a contract route, not a capacity guarantee, so any Korean customer still has to negotiate volume and timing project by project.
Can a design built for Samsung’s process move to Intel 18A without a redesign?
No. A process port needs a new PDK, re-targeted analogue and interface IP, a new physical design and full verification. Intel’s migration model implies joint engineering work rather than a database hand-off, and Intel has published no schedule, cost or supported-IP list for such ports. Treat migration as a new tape-out programme with its own mask set and qualification budget.
Does Intel’s advanced packaging accept chiplets made at other foundries?
Intel sells packaging as a service and says it can integrate processors of different architectures in one package, but it has not published which external nodes or die formats it will accept. Feasibility depends on die-level interfaces, test access and known-good-die data, all of which are negotiated per project. A packaging quotation should list the source fab, bump pitch, interposer or bridge type and the test coverage assumed for each die.
Is there Korean government money for using Intel Foundry?
No. MOTIE’s 800.2 billion won on-device AI programme assigns prototype priority to Samsung Foundry for 21 runs in 2027, including three at 2nm. No Intel allocation appeared in the programme announcement or at the summit. A Korean fabless company can still use Intel commercially, but it would not receive the prototype subsidy attached to the domestic programme.
Which Intel nodes can an external customer book today?
Intel says 18A is in production, 18A-P is on track for launch by year-end 2026, and the 14A 0.9 PDK was released to select customers in October 2026. The company has not named external customers or stated which node is open to Korean firms. Any enquiry should ask for the specific PDK revision, IP availability and risk-production date rather than a node name alone.
Does the K-AI Full-Stack Alliance require members to use Intel chips?
No public volume commitment was announced. The alliance lists 23 initial Korean participants across hardware, software, cloud and services, and Intel Korea signed a separate memorandum with Seoul AI Hub. Membership is an ecosystem arrangement for joint marketing, validation and reference designs, not a supply contract, so it does not commit a member to an Intel foundry or packaging order.
Sources
- MoneyToday: Intel’s five cooperation models for Korean AI companies — 12 November 2026; five models, Korea memory share, October export growth.
- Businesskorea: Intel Unveils New GPU, Seeking Korean AI Deals — 12 November 2026; Marcus Lawrence quotation, Crescent Island details, 23-member alliance.
- TheElec: Intel Forms AI Alliance With 23 Korean Companies — 13 November 2026; Tae-Won Bae, alliance list, Seoul AI Hub MOU, Dan Rodriguez and Core Ultra Series 3, Intel GPU D-Series figures.
- ChosunBiz: Intel and SK Hynix Present Solutions to AI Inference Bottlenecks — 12 November 2026; CXL DRAM first-token latency, Samsung Medison case, Intel custom-silicon model.
- Etnews: K-AI Full-Stack Alliance launch — 12 November 2026; 52 participating companies, four government and public bodies.
- Intel third-quarter 2026 prepared remarks — 29 October 2026; 18A, 18A-P, 14A PDK, EMIB substrate constraint, New Mexico and Malaysia packaging capacity.
- The Edge Malaysia: Advanced packaging is Malaysia’s next AI test — Penang complex, 710,000 sq ft cleanroom, first-phase timing.
- Businesskorea: South Korea to Develop 8 AI Chips and 2 Semiconductor IPs by 2030 — 12 November 2026; 800.2 billion won budget, MOTIE roadmap, Samsung Foundry’s 21 prototype allocations and node split.
- Aju Press: South Korea to develop 10 domestic AI semiconductors by 2030 — 12 November 2026; Samsung Foundry’s 21 prototype runs, three at 2nm and four at 4nm.
