Semiconductor Supply Chains, AI and Economic Statecraft

2.3 Strategic AI-semiconductor partnerships and collaborations

The convergence of AI and semiconductor technologies has spurred a wave of strategic partnerships and collaborations. The best partnerships combine expertise of AI companies in algorithm development and software optimisation with semiconductor companies’ capabilities to design and manufacture at cutting-edge process nodes (Korean collaborations of this nature are discussed in Section 4).

• Intel and NVIDIA: Intel is a leading semiconductor manufacturer, while Nvidia is a pioneer in GPU computing. They have indicated possible collaboration on “Confidential Computing for AI” which would signal greater integration of platforms, although details remain scarce.
• Google and TSMC: Google, a key player in AI research and development, has a manufacturing partnership with Taiwan Semiconductor Manufacturing Company (TSMC), the world’s largest dedicated semiconductor foundry, to develop custom AI chips. Google’s Tensor Processing Units (TPUs), designed specifically for deep learning tasks, will be manufactured by TSMC in 2025 using advanced process nodes. These chips are used for so-called “edge” compute to be deployed in smartphones.
• Apple and TSMC: Apple, renowned for its integration of AI technologies in products like Siri and Face ID, collaborates closely with TSMC to produce custom-designed chips, including the A-series processors used in iPhones. TSMC’s advanced manufacturing processes enable Apple to achieve significant performance and energy efficiency gains. As Apple moved away from Intel-designed chips to design their own chips (“Apple silicon”), they partly drove TSMC’s rise as the world’s largest manufacturer. 
• AI start-ups and semiconductor foundries: Many AI start-ups often collaborate with semiconductor foundries to design and manufacture specialised AI chips. Foundries like TSMC, GlobalFoundries, and Samsung Foundry offer access to advanced process technologies and design services, enabling start-ups to bring their AI innovations to market faster and more cost-effectively. 

As the AI landscape continues to evolve, we can expect to see deeper integration and collaboration between AI and semiconductor companies, leading to even more ground-breaking advancements (and concentration of market dominance among a small number of companies).

2.4 Looking ahead: A ‘frontier AI’ dominated landscape?

The relationship between frontier AI models and the semiconductor ecosystem could shape the future of both industries in the following ways:

1. Evolution of AI accelerators: as AI models become more sophisticated, there is a growing demand for specialised chips to handle AI workloads. Major semiconductor manufacturers are expected to develop AI accelerators, poised to compete with GPUs for market share.
   
   
2. Critical role of semiconductors in AI advancement: semiconductor innovation has enabled further development and proliferation of AI technologies. The demand for high-performance chips continues to rise as AI applications become more data intensive.
   
   
3. Technological advancements: HBM chips are crucial components in AI systems, especially for training large language models (LLMs). Companies like SK Hynix and Samsung Electronics are scaling up production of HBM chips to meet the growing demand from AI applications. An emerging trend here is in-memory compute, a technology primarily espoused by IBM and emerging photonic computing platforms.
   
   
4. Compute infrastructure challenges: the demand for high-performance chips, such as Nvidia’s H100 and A100, has led to unconventional arrangements and strategies to secure compute resources. Governments and companies are investing heavily in compute infrastructure to support AI research and development.
   
   
5. Technological diversity and innovation: newer chip designs, such as Graphcore’s Intelligence Processing Units (IPUs), offer potential alternatives to traditional GPUs for AI applications. However, challenges exist in breaking lock-in effects and promoting technological diversity in national AI compute infrastructure.
   
   
6. Compute intensiveness and energy efficiency: the exponential growth in AI compute demands poses significant challenges in terms of energy efficiency and resource consumption. Efforts are underway to address these challenges, but the industry faces substantial hurdles as it navigates the limitations of Moore’s Law.

In summary, AI requires an entire system to work in concert to deliver computational advances. This entails close interplay between the software, piping and core hardware technology, so-called System Technology Co-Optimization (STCO).

As AI continues to evolve and permeate various industries, collaboration between software developers and chip designers will be essential to driving innovation and shaping future AI architectures. Strategic partnerships between leading companies in both sectors will also dictate the direction of travel: by combining expertise in AI algorithms, software development, and hardware design, these partnerships will push the boundaries of what is possible in AI-powered applications.

3. The UK Semiconductor Industry: Punching Above its Weight?

The UK semiconductor industry has seen a hive of activity building up to the National Semiconductor Strategy and following its publication. On the international front, this involved semiconductor-related agreements signed with Japan and Korea in 2023. 

To build on these agreements, the UK must double down on a clear offering in the semiconductors space. Bilateral and multilateral partnerships are proliferating at speed, and the UK cannot afford to be left behind. If it is unwilling to leverage industrial policy in the same way as some other countries, it must be even more aggressive in leveraging existing semiconductor strengths, and adjacent domains of expertise in areas like AI and cybersecurity. 

3.1 Recent developments

In August 2023, then Technology Minister, Paul Scully MP, said that the UK will not “recreate Taiwan in South Wales” and will not join the race across Asia, Europe and the US spending billions on advanced chip-building facilities. Instead, as per the National Semiconductor Strategy published in May 2023, the UK would invest up to £200m in the semiconductor sector between 2023-25, and up to £1bn in the next decade, focusing on the following three objectives:

1. Grow the domestic sector by establishing the Semiconductor Advisory Panel; ensuring that applied research is oriented to industry needs; improving access to infrastructure; and better attracting talent.
   
   
2. Mitigate supply chain disruptions by helping industry better understand risks from future shortages; improving resilience in critical sector product manufacturers; working with international partners to prioritise access to chips; and diversifying supply.
   
   
3. Protect UK national security by utilising investment screening and export controls; and leveraging existing hardware strengths so that cybersecurity is prioritised at the chip design stage across the world. 

The total £1bn funding will support the ‘UK Semiconductor Infrastructure Initiative’, with some indications that this could morph into a UK National Semiconductor Institute. A central coordinating function for the UK semiconductor industry was seen as sorely lacking in the current landscape by some interviewees, and it is imperative that a new National Semiconductor Institute is established this year.

UK fabs do not produce the most advanced silicon semiconductors. The lack of investment into building semiconductor facilities in recent decades has led to specialised clusters forming around certain universities. The benefits of this will be highlighted later, but it also means that each fab has its own manufacturing method and end-product, with knowledge, skills, and manufacturing equipment less likely to be cross-purpose.

It also means that there may be a mismatch between the output from UK fabs and the requirements of UK manufacturing or technology firms which rely on semiconductors. Further upstream, it is generally a struggle for small companies and research organisations to access the infrastructure needed to develop designs or manufacture prototypes to attract investment (although the reasons for these struggles usually vary across companies).

International partnerships are increasingly being recognised as an important route to addressing domestic shortcomings and reinforcing UK strengths. The UK-Japan Digital Partnership coincided with the launch of the National Semiconductor Strategy, and there is more focus across government on identifying ‘Tier 1’ countries with which to prioritise bespoke agreements, also seen in the UK-Republic of Korea framework for semiconductor cooperation announced in November 2023. Yet given geographical disparities, time zone differences, and more limited day-to-day cultural engagement with partners in Asia, UK stakeholders need to continually nurture these relationships in order for them to bear fruit. 

Fundamentally, future UK semiconductor resilience will be dependent on the success of industrial policy in countries like the US, Korea, Japan and the EU: in the case of a China/Taiwan crisis, the UK manufacturing base will need to be able to draw on chips made in allied countries, or otherwise face the prospect of seizing up entirely. This is why it is important for the UK to invest in the design capability that can exploit the available friend-shored manufacturing capability, coupled with strategic agreements for reserve chip capacity in partners’ fabs. 

Nonetheless, there was also a view from some interviewees that the UK’s relative financial prudence could, perhaps paradoxically, end up working in its favour as US-China trade disputes escalate: “this might give the UK a mild competitive edge because we are not going hot and cold on the money – simplicity could be working to our advantage.”

Because of semiconductors’ position at the bottom of the tech stack and their facilitatory nature for other technologies, it is also informative to consider how the semiconductor industry is impacted by a wide range of ‘adjacent’ government strategies, reviews, and associated funding pledges, separate to the National Semiconductor Strategy. One of the main upshots of this is that there is a large amount of money in the telecoms, quantum, and AI industries that could be better aligned – making these strategies and reviews fit together as part of a coherent whole will underpin UK strategic advantages in the long run.

Table 2. Adjacent UK strategies and investments 

3.2 Areas of strength 

The UK’s National Semiconductor Strategy emphasises that ‘building on our strengths will help us achieve our ambitions elsewhere: to lead the way on AI, to enable advances in quantum computing and telecommunications, to power high performance computing, and to facilitate progress towards net zero and in life sciences’. It is therefore worth profiling these existing strengths and some of the barriers to further consolidation. 

3.2.1 Core IP and chip design

Core IP consists of reusable, modular design blocks that chip design firms license for use in their designs. This represents the main part of the supply chain where the UK is close to market leadership. According to the Emerging Technology Observatory Supply Chain Explorer, the UK has 43% of global market share in core IP, largely due to ARM being the top core IP vendor, providing an instruction set architecture (ISA) and underpinning most of the world’s smartphone processors.

Section 2 touched on the centrality of ARM to UK semiconductor competitiveness, and this was also underscored by interviewees. One commented, “we must have a seat at the table. If we lose our position in the design of chips, via the likes of ARM, we become irrelevant in terms of working with the big three global companies (TSMC, Samsung, Intel).” It is therefore easy to see why the eventual failure of the Nvidia-ARM merger was met with relief in UK policy and industry circles. The prospect of ARM being restricted to developing on-chip AI functions tied to Nvidia’s proprietary hardware, along with the physical relocation of ARM’s headquarters to the US, would have significantly undermined UK leverage in the sector.

Nonetheless, ARM is not the whole story – global companies like Apple and Intel also have UK offices focused on semiconductor design and GPU development, while Graphcore are a Bristol-based company which specialise in designing hardware for AI. Yet Graphcore arguably offers an example of where the government’s strategic approach falls short. While Nvidia dominates the GPU market, Graphcore focuses on IPUs which are designed to be better suited to the specific requirements of AI. Because IPUs are so different to the Nvidia GPUs familiar to users, gaining a foothold in the market is a tall order. This was made even more challenging when the UK Government’s Exascale funding explicitly specified GPUs, excluding IPUs from the tender. The fact that the US Department of Energy’s National Labs have made IPUs part of their infrastructure makes this decision even more surprising – forsaking technological diversity in the national AI compute infrastructure will further entrench applications that suit GPUs and limit techniques made possible by new AI systems.

The UK Government could also be doing more to address the high costs of EDA tools which restrict production timelines for new designs and smaller companies’ ability to scale up. These costs often result in companies outsourcing and missing out on commercial opportunities. One promising solution that has been recommended in this regard is a government-sponsored Design Competence Centre offering design flow support, design services and EDA/IP training. Working with international partners like Korea on open-source EDA tools would also be a positive way forward.

A longer-term threat to the UK’s chip design strengths comes in the form of growing competitiveness of Chinese chip design companies, which have not received as much media attention as Chinese foundries (like SMIC) and memory chip manufacturers (like YMTC). In areas not targeted by US export controls, China is likely to strengthen its design capabilities, benefiting from their strength in adjacent industries like consumer electronics and EVs. This symbiotic relationship between design and end-customer industries could create new dependencies on Chinese chip design, which is especially concerning given that Chinese chip design capabilities could have defence and espionage-related implications. The upshot of this is that the UK must retain sovereign control as a design authority, while also maintaining the capability to use the complete global design and manufacturing supply chain in service of national security. 

3.2.2 Compound semiconductors

Similar concerns about China potentially undercutting UK strengths are relevant to compound semiconductors. These are semiconductor materials made from two or more elements, for example, silicon and carbon resulting in silicon carbide. South Wales is home to the CSconnected  cluster, the world’s only dedicated compound semiconductor cluster. Although currently only 20% of chips produced are compound semiconductors (the remaining 80% are pure silicon), they do possess three properties that outperform silicon:

• Power – i.e. electronics for EVs
• Speed – i.e. radio frequency for 5G and RADAR
• Light – i.e. photonics for optical fibre communications

The interest in compound semiconductors is partly driven by the limits of continued transistor scaling and the need for performance gains through novel technologies. There is a clear logic to the UK focusing on new architectures and materials, given that it is not a winner in the current silicon-dominated paradigm. There are high value areas of the compound semiconductor industry where giants like TSMC are not active: including silicon prototyping, photonics, quantum chips and silicon carbide. Indeed, one interviewee commented that “even Taiwan sit up when we talk to them about compound semiconductors, Japan too. That could be a strength for us to play internationally.” Samsung are showing interest in building out their compound semiconductor line, and the UK should be leveraging its strengths here to ensure that any prospective plant is built in the UK rather than other options in Europe, such as Germany. 

Another driving factor are the numerous promising applications of compound semiconductors in the defence and security context. For example, indium phosphide chips are used for facial recognition and high-speed optical communications, particularly in demanding sub-sea environments, while gallium nitride and gallium arsenide play vital roles in devices with lasers, LIDAR, and sensors, helping connectivity and enabling use in satellite applications.

Recent investments indicate a vote of confidence in UK companies’ potential. The UK Infrastructure Bank co-invested £182m in Pragmatic Semiconductor at the end of 2023. Theoretically, the approach adopted by Pragmatic could result in fabs which are 10-100 times cheaper than a standard manufacturing facility, taking only 48 hours to go from the start of the process to a finished wafer product. This encapsulates a trend of more diversification in how semiconductors are made and what they are used for. 

Despite strengths in compound semiconductor R&D, there is still a recognised gap in the subsequent route to commercialisation. This is why the Institute for Manufacturing – following a DSIT-commissioned study – called for an uplift in compound open-access foundry capability, proposing intervention options ranging between £100m and £250m in cost.

There is currently limited overlap between the UK’s research strengths in compound semiconductors and commercial ambitions in AI. Building advanced logic chips using compound semiconductor capability is still likely to be 10-15 years away, although there may still be features of an AI system (such as image sensors) where compound semiconductors could have a role. The UK compound semiconductor industry is showing great promise, but the reality is that it comprises a broad mixture of processes that build very different chips, so will not be a silver-bullet for any single application. More strategic alignment with other UK strengths moving forward will be key in attracting larger investment and proving economic value. 

3.3 Hardware and cybersecurity 

A resilient semiconductor industry is largely dependent on security throughout the supply chain. This requires designing mitigations into chip devices to protect against both supply chain side-channel attacks and regular operation attacks. This is important because the cybersecurity risk surface is usually broadened when software interacts with hardware, rather than a vulnerability being due to software or hardware individually. Vulnerabilities built into hardware may be exploited by malicious software to undermine the security of a chip and the critical applications it enables. 

The previous section discussed UK chip design strengths, but from a security point of view, these strengths can be undermined once a design is sent for manufacturing, where attackers could install a backdoor or cause deliberate failure in the device. New functionality added at the packaging stage can change the workings of a chip – if flaws are introduced into enough computers, then backdoors could compromise whole networks in a single attack. Moreover, the threat increases as chip production and packaging gets more complex – Chinese packaging innovations in chiplets and heterogenous integration layered on top of a UK chip design could amplify security risk significantly.

These risks may also be inadvertent, as with the ‘Meltdown’ and ‘Spectre’ vulnerabilities that hit microprocessors in 2018, which allowed attackers to exfiltrate data. But they reflect the need for engineering design practices to identify code or IP that has not been part of the original design specification. Side-channel and fault-injection attacks also allow for chips to be targeted in the field – the type of attacks described above cannot typically be patched immediately like pure software vulnerabilities.

AI also adds an expanded attack surface through interventions in the model training and deployment chain, such as trojan attack insertion into the training loop and malicious model deployment. An AI application running on an AI chip thus has an extended attack surface over and above something that does not use trained models.

When interviewees were asked about plausible worst-case scenarios, many spoke of “kill switches” implanted by adversaries on circuit boards in critical infrastructure, although the likelihood of this happening at scale is very hard to determine. The limited number of processor designs and their ubiquity across most embedded devices was seen as a critical risk factor:

‘One generic processor design could be used in the brake controller of every car in the world (…) if you can get into every car in the country, turn them all left or stop them all at once, that would be disruptive. The example is just illustrative but is theoretically possible because of the ubiquity of a small number of products.’

This reinforces the challenges the UK faces meeting sensitive semiconductor requirements. The National Semiconductor Strategy commits to ‘evaluate the UK’s future domestic semiconductor manufacturing needs to see where the baseline level of manufacturing could provide a low volume of chips for critical infrastructure.’ In the immediate term, focusing on technical solutions for hardware verification and security may be more useful due to the difficulty of avoiding reliance on overseas suppliers. 

There are also numerous market incentives challenges which are inhibiting better security outcomes. The National Semiconductor Strategy references this, saying that chip security is ‘not seen as an independent end goal’ for designers and manufacturers, and this was reaffirmed by interviewees: 

‘People buy capabilities because they are fast or shiny (…) often there is the expertise to help a company fix a problem, but the cost and the cascading effects of fixing it would mean they go out of business.’

Building a secure chip needs to be the default option for the market, and there are lessons to be learned from recent history: 

‘20 years ago, multicore processors basically had no security, and now a CPU is pretty good security wise (…) now that we have had the AI explosion, developers cannot get away with not thinking about security, we cannot wait 20 years for security to catch up.’

The challenge for the UK is to leverage its status as a global authority on hardware and cybersecurity. The Digital Security by Design (DSBD) programme is a public-private partnership which aims to ensure memory safety of software without rewriting trillions of lines of code, and block cyberattacks from over 70% of today’s ongoing vulnerabilities. The programme has already delivered DSBD technology through ARM’s Morello project, which has boosted the UK’s credentials in this domain according to overseas experts: 

‘When we talk about R&D collaboration on secure computing architectures, the UK are a partner of note (…) the ARM and Cambridge-run programme is a great case study, so we have to find tangible, well-scoped ways to cooperate around particular challenges.’

The essential IP to adopt the DSBD approach is freely available, allowing it to become a standard feature in chips worldwide, but there is more to be done to further expand international outreach and ensure that countries like Korea are fully engaged in these endeavours as well as the US. This was also recognised by UK-based interviewees who said that the UK’s DSBD and cybersecurity leverage has still not been used to full effect globally, though some also highlighted that the Morello project is only useful for some side-channel attacks, not all.

Nonetheless, regarding Korea specifically, one interviewee believed this was a particularly fruitful avenue of collaboration, because of recent alarm in the country regarding cyber preparedness, referencing recent incidents such as data centre fires (heavily impacting NAVER and Kakao services) and government IT shutdowns. The UK has recognised competency in this space, and there is a clear opportunity to make it a stronger bargaining chip and enable deeper collaboration with trusted allies. 

3.4 Leveraging UK R&D 

Underpinning the areas where the UK excels is a globally competitive R&D ecosystem, which takes a keen interest in AI research, as well as deep technical research in next-generation compute architectures. 

Compute to realise AI goals

A significant driving force for the UK Government’s approach to semiconductors and investments in compute is the provision of hardware needed to achieve breakthroughs in technologies like AI and quantum. Using non-state-of-the-art AI chips increases both energy consumption and training time significantly, translating directly into cost increases – Nvidia’s cutting-edge H100 GPUs have a chip-to-chip interconnect bandwidth speed of 600 GB/s compared to 300 GB/s for the H800 GPUs sold to China. In recent years, OpenAI and DeepMind made decisions to be invested in or acquired by Microsoft and Google respectively primarily due to compute costs.

One interviewee speculated that training runs could reach Apollo Programme levels and eat up large portions of countries’ GDP, while another referenced industry discussions about new power stations being needed for the data centres that drive AI compute. The total energy consumption by general-purpose computing continues to grow exponentially and is doubling approximately every three years, while the world’s energy production is growing only linearly, by approximately 2% a year.

The UK’s Advanced Research and Innovation Agency (ARIA) has launched a £42m moonshot programme seeking to meet this challenge, focusing on principles found ubiquitously in nature to create more efficient options for AI infrastructure at 1/1000^th the cost. This is an example of ambitious R&D on next-generation compute architectures which the UK must leverage globally.  

It is not just training AI models but also evaluating them which depends on reliable access to compute. This is central to the UK’s ambitions to be a global leader in AI safety research and evaluation, and the recently established AI Safety Institute will receive priority access to the £300m AI research resource (AIRR). These developments are especially important considering the recent UK-US partnership on science of AI safety.

AI to realise chip goals

At the same time, the potential of AI to produce better chips is becoming clearer. More companies are rolling out internal AI tools to improve chip design (Nvidia has been doing so for several years), while AI is also being used to determine equipment and process failure in fabs, which can often occur unexpectedly. This is noteworthy in the context of the government’s new Advanced Manufacturing Plan. Additionally, generative AI may help to reduce chip design times and enable simpler natural language inputs instead of needing specialised programming languages.

Algorithmic progress is a significant component of performance improvement, and has played a more important role in Nvidia achieving 1000x gains in GPU performance over the last 10 years than pure hardware efficiency (see Figure 8 below). Given both the fact that UK AI strengths are more pronounced at the software/algorithm level rather than the hardware level, and the importance of the chip design sector to the UK’s semiconductor positioning, there is a clear logic to the UK leaning into this dynamic more aggressively.

Figure 8. Nvidia GPU improvement over 10 years

Source: Nvidia, 2023.

In 2019, South Korea’s then president, Moon Jae-in, said that if the country could “link AI primarily with the sectors in which we’ve accumulated extensive experience and competitiveness, such as manufacturing and semiconductors, we will be able to give birth to the smartest yet most humanlike artificial intelligence.” For the UK, the reverse may be more instructive: utilising its world-leading research base in AI and novel forms of computing may be vital in strengthening the UK’s position across the semiconductor supply chain.

4. The South Korea Semiconductor Industry: A National Treasure

In an age of hyper-uncertainty, Korea and the UK can work closer together to secure competitive advantage in critical technologies. The semiconductor-AI lens presents an excellent opportunity to leverage the complementarities between both countries. This section focuses on the strengths that Korea brings to this equation.

4.1 Recent developments

4.1.1 Semiconductors 

Korea has maintained the second largest market share in the global semiconductor industry and has been the market leader in memory semiconductors since 2000. Korea occupies 18.4% of the global semiconductor industry (56.9% of memory (DRAM/NAND), 3% of non-memory (DAO)).

Yet despite considerable strengths, Korea has experienced several crises due to its high dependency on materials, parts, and equipment (MPE) and vulnerability to geopolitical risks. Three recent cases are informative here:

• Japanese export controls on three key chemicals (hydrogen fluoride, photoresist, and fluorinated polyimide).
• US export controls on equipment (such as Extreme Ultraviolet Lithography (EUV)).
• Chinese export controls on gallium and germanium.

To overcome structural vulnerabilities, Korea has developed several national strategies to support semiconductor competitiveness, the most significant of which are the ‘National Advanced Strategic Industry Act (2022)’ and the ‘National Strategic Technology Act (2023)’.

The National Advanced Strategic Industry Act features extensive financial and regulatory incentives promoting strategic industries across 12 key technology areas, as well as designating strategic industry clusters. In January 2024, Korea’s Ministry of Trade, Industry, and Energy (MOTIE) and Ministry of Science and ICT (MSIT) announced a plan to build the world’s largest semiconductor mega-cluster, which will have the capacity to produce 7.7 million wafers per month by 2030. The Act also designates specialised research institutes, establishes regional technology innovation hubs and strengthens mission-oriented R&D through a National Critical Technology (NCT) laboratory system.

At the same time, Korea’s private sector is committing large-scale investments. For example, Samsung Electronics announced its ‘Semiconductor Vision 2030’ which included $55 billion investment in R&D and $45 billion in infrastructure in 2019. Additionally, Samsung Electronics has initiated world-leading production based on the 3 nm foundry process incorporating Gate-All-Around (GAA) technology.

4.1.2. Artificial Intelligence

The Korean AI landscape is characterised by the growing number of companies across the entire AI value chain. Korea benefits from a sovereign data ecosystem: NAVER, not Google, leads the search engine market, and domestic social media platforms also dominate. The Korean Government is also supporting these data ecosystems by constructing a national integrated database called the ‘Public Data Portal’.

Figure 9. Korea AI Landscape

Source: Author, Seungjoo Lee, 2024. 

Korea’s AI capabilities are dispersed across several established companies, such as NAVER, LG, and KT, and promising start-ups across diverse fields like healthcare (Lunit), LLMs (Upstage), and education (Quanda). Since 2019, the Korean government has implemented a comprehensive strategy to build on this foundation, focusing on boosting competitiveness, expanding AI adoption across all sectors, and ensuring ethical implementation.

Table 3. Korea AI strategies

Source: Summary of MSIT documents and press releases.

These efforts have already yielded tangible results, including major advancements in flagship AI projects, AI-powered healthcare systems, and nationwide AI literacy programs. Yet challenges remain in balancing innovation with ethical considerations: the National Human Rights Commission recently argued for the elimination of the 'principle of first accept and later regulate'.

The Korean Government is proactively leveraging strengths in cutting-edge fields like semiconductors and AI in its international partnerships. Collaborations are already underway with the UK, Singapore, and the US. Additionally, large-scale projects in Saudi Arabia offer promising opportunities for Korean companies, backed by government support. 

Table 4. Recent international cooperation initiatives

4.2 Areas of strength

On the semiconductors front, Korea’s strength is based on two of the world's top-tier IDMs – Samsung Electronics and SK Hynix. By maintaining a dominant position for over 20 years, Korea has succeeded in possessing the most advanced technology and production capacity, along with the most sophisticated ecosystem.

Furthermore, Korea’s competitiveness in relatively weaker areas such as foundries is improving. In the foundry sector, Samsung Electronics is investing aggressively in its capabilities. In the fabless sector, several innovative start-ups such as Rebellions and Furiosa are growing rapidly. The active participation of a domestic digital platform/cloud provider (NAVER) possessing advanced LLM capacity also enhances Korea’s competitiveness in chip design.

Meanwhile, Korea also holds impressive strengths in AI: the Global AI Index rankings place Korea 3rd in development, 6th in government strategy, and 7th in infrastructure, while the Government AI Readiness Index places Korea 6^th. Global AI patent applications are also high (4^th) although only 7% of patents fall within the top 10% of the patent citation index, falling short compared to the 14% average of the leading 10 countries.

Several major Korean companies have established dedicated research organisations, including Samsung AI Center, NAVER AI Lab, Kakao Brain, and LG AI Research Institute. These companies are increasingly focused on building and commercialising their internal LLM capacity. 

4.3 SK Hynix/Samsung strategic pivots

4.3.1 SK Hynix: Mission-Oriented Semiconductor-AI innovation

To achieve twin transformation, the SK Group is pursuing three strategies at the intersection of semiconductors and AI:

• Developing next-generation semiconductors (e.g. HBM) for generative AI. 
• Introducing AI into its core industries to enhance manufacturing and IT service competitiveness.
• Refreshing its international strategy considering heightened geopolitical risk.

Enhancing Future Technology Competitiveness with Open Innovation

SK Hynix’s Revolutionary Technology Center (RTC) conducts next-generation semiconductor research, focusing on key areas such as future DRAM, NAND, new memory, quantum computing and neuromorphic computing. SK Hynix has successfully developed the 5th generation of HBM and initiated sample supply for customers. HBM vertically connects multiple DRAMs to enhance data processing speed and is recognised as a high-value, high-performance product capable of meeting the data processing needs of generative AI.

SK Hynix’s future technology strategy also prioritises international collaborations with prestigious institutions like the Interuniversity Microelectronics Center (IMEC), the National Institute of Standards and Technology (NIST) and the Semiconductor Research Corporation (SRC). The open, international positioning of the RTC has enabled the exploration of cutting-edge technologies and the pursuit of new approaches to overcome the limitations of existing memory semiconductors.

Two-track domain driven AI-semiconductor strategy by Gauss Labs and Sapeon

The SK Group has established two independent AI companies, Gauss Labs and Sapeon, to enhance its AI-semiconductor capacity. The former is pursuing ‘AI for manufacturing innovation,' while the latter is concentrating on 'AI for IT service innovation.’

Gauss Labs uses data generated from SK Hynix plants to optimise manufacturing at several stages, including process management, yield prediction, equipment maintenance, material measurement, and defect inspection and prevention. Meanwhile, Sapeon aims to develop its own neural processing unit (NPU) and process in memory (PIM) for advanced IT services – Sapeon successfully developed and commercialised Korea’s first AI chip for data centres (X220). The ongoing development of the Sapeon X300 series supports AI model training and is being developed for various fields such as automotive, security, and media.

Refreshing international strategy: the geopolitical risk approach

The SK group has historically maintained strong business ties with China. However, recent geopolitical tensions have motivated the SK Group to adopt a new international strategy with a more cautious approach to operations in China. The SK Group is establishing offices in the US and investing heavily in manufacturing plants there, while also seeking closer alignment with the US government on economic security matters. This closer proximity is helping to deal with requirements for accessing CHIPS Act subsidies and to mitigate the impact of scandals like SK Hynix products being installed in Huawei’s new smartphones.

4.3.2 Samsung Electronics-NAVER: AI-semiconductor collaboration for mutual competitive advantage

Global technological competition is intensifying, meaning that companies lacking a unique offering are at risk. To overcome this, NAVER and Samsung Electronics (hereafter Samsung), Korea’s leaders in AI and semiconductors respectively, launched a taskforce in December 2022 to develop semiconductor solutions optimised for AI systems. The main objective was to solve AI data bottlenecks and maximise power efficiencies.

Mutual complementarity for sustainable growth

Samsung faces limitations in the fabless and foundry fields despite its dominant position in the memory field. Commercialising AI semiconductors presents a strategic move to not only strengthen its weaker presence in the non-memory market but also to secure a leadership position in an AI-driven future. Samsung wants to both reinforce existing technologies like HBM, compute express link (CXL), and PIM, and explore novel hardware structures specifically designed to accelerate AI model compression techniques. These tailored structures would bring significant efficiency boosts and reduce the amount of data transferred between processing units and memory. 

Developing such chips requires expertise in both semiconductor design and manufacturing and AI algorithm design and verification, with a specific focus on optimising hardware structure for specific AI architectures and model compression techniques. Therefore, collaborating with AI specialists like NAVER is crucial for Samsung's success in this new market. 

On the other hand, NAVER possesses a near-monopoly on the Korean IT industry. Based on strong network effects, NAVER has accumulated large amounts of high-quality data, which has fed into their superior ML capabilities. Server costs and availability are a huge challenge to providing hyper-scale AI services, which has meant several major AI companies have started developing their own AI chips. Indeed, in January 2024, OpenAI CEO Sam Altman visited Samsung and SK Hynix to discuss AI-semiconductor collaboration – this link was reaffirmed when Samsung and SK Hynix were suggested as potential partners in reports of a multi-trillion-dollar fundraising vision.

Samsung’s capabilities across the supply chain make them a compelling partner for many global companies seeking to develop specialised AI semiconductors. However, it is closer to home through NAVER where the strongest partnership has emerged so far.

Progress and further cooperation

The partnership works along the following lines. NAVER identifies key technical challenges for their AI services, while Samsung leverages its hardware expertise to develop solutions. NAVER then validates these solutions with their software know-how. Both companies prioritise enhancing computational and inferential capabilities while maximising power efficiency. 

A year into the partnership, they unveiled their first AI-semiconductor prototype, demonstrating an 8x or more improvement in power efficiency compared to current commercial semiconductors. Design is now under construction aiming for mass production by the end of 2024.

Samsung has set ‘hardware solutions for on-device AI’ as an important goal. Aligned with this vision, the Samsung-NAVER partnership is known to have adopted low power DDR DRAM instead of HBM. The chips will be manufactured at Samsung’s foundry and deployed in NAVER’s data centres. If successful, Samsung and NAVER plan to offer the chip to other global tech giants.

4.4 Leveraging semiconductor strengths for a globally competitive AI industry

The simultaneous pursuit of a domestic AI-semiconductor ecosystem and proactive international cooperation is crucial to Korea’s overall approach. Korea must strengthen areas of existing competitive advantage and build buffers where it is weaker.

The Korean semiconductor industry benefits from the rise of AI chips, as evidenced by the increase in demand for SK Hynix’s HBM. But mechanisms are still needed to ensure market stability and coordination. To this end, MSIT has launched the ‘AI Semiconductor Scaling Network’, which aims to support links between AI chip buyers, suppliers, and researchers to create a highly networked domestic ecosystem. This is something that DSIT should take a close interest in and consider mirroring in the UK.

The Korean Government is also developing a more joined up approach with key semiconductor allies. One example of this was an agreement during the August 2023 trilateral US-Korea-Japan summit to launch a pilot project focusing on supply chain early warning systems. The warming of relations with Japan has seen Samsung invest 40 billion yen (£210m) in an advanced semiconductor R&D centre in Yokohama, Japan, which will facilitate joint research between Korea and Japan on semiconductor packaging technology. At the same time, SK Hynix is working more closely with Taiwan, outsourcing some processes for HBM4 to TSMC.

The domestic knowhow that Korea is developing through the Samsung-NAVER partnership is positioning it as a key country in the global AI-semiconductor nexus. This is also reflected in government policy: in January 2023, the Korean Government decided to invest KRW 260 billion (£153m) in the development of next-generation AI, and in June 2023, MSIT announced an investment of KRW 1.02 trillion (£600m) to create an industry-academia-research institute collaboration ecosystem. This brings together (1) multinational companies like Samsung and SK Hynix, which are at the core of the semiconductor-AI ecosystem, (2) start-ups such as Furiosa AI, Rebellions, Sapeon, Telechips, and DEEPX, and (3) research institutions like The Korea Advanced Institute for Science and Technology, The Electronics and Telecommunications Research Institute, and Seoul National University. The Korean government will allocate KRW 2 trillion (£1.2bn) over the next five years to support the development of AI semiconductor libraries and compilers. Moreover, data centres, which are one of the largest sources of demand for AI semiconductors, will be built to generate initial demand.

These measures could ensure Korea is better positioned to capitalise on increased global demand for AI systems outside of the US or China, in the event of an escalating trade war. The UK could occupy a similar position in this regard, potentially working closely with Korea on joint ventures in third countries.

5. Semiconductors and Geostrategic Alliances: Opportunities for the UK and Korea

Geopolitical risks are creating a dichotomy between technological competition and cooperation. On one hand, individual countries are driven to prioritise their own national interests, leading to fragmentation rather than integration of technologies and industries. On the other hand, the need for cooperation is increasing because of the importance of an open innovation ecosystem to securing competitive advantages in high-tech industries. 

This section seeks to bring together the individual analyses of the semiconductor-AI landscapes in the UK and Korea, to demonstrate why the two countries are an ideal case study in bilateral technology and security partnership. 

5.1 A high-point in UK-Korea relations

The November 2023 Korean State Visit to the UK coincided with the celebration of 140 years of diplomatic relations between the two countries. Last century, the UK contributed around 90,000 soldiers to fight in the Korean War; the focus today is more on finding ways to bolster mutual technological and economic security.

The State Visit, which resulted in the signing of the Downing Street Accord, was a landmark moment in formalising both countries’ shared intention to strengthen and deepen bilateral collaboration across several critical sectors. In a speech to the UK Parliament, President Yoon spoke of their bilateral relations being “reborn as true ‘Global Strategic Partners’” and the need for them to “stand in solidarity and respond to many of the world’s challenges.” Prime Minister Rishi Sunak said, “as two nations focused on innovation, harnessing new technologies and defending the international rules-based order, the UK and ROK are natural partners”, recognising the value of the nexus between technology and global security. This is also timely given that Korea is a member of the UN Security Council from 2024-25.

In a recent report, Imagination Technologies laid out general criteria for an ideal UK strategic interdependence partner. These are informative for thinking about UK-Korea complementarities vis-à-vis security and technology.

Table 5. Criteria for a strategic interdependence partner: Korea

Source: Adapted from Imagination Technologies and Global Counsel, 2022.

5.2 Complementarities in semiconductors and AI

The table above demonstrates high-level suitability, but there is more to be gleaned at the specific level of semiconductors and AI. The fact that the UK and Korea have high ambitions in multiple technology areas means that there is an additional benefit to focusing partnership efforts on semiconductors and AI, technologies which are mutually reinforcing, and which provide benefits cascading through to other critical technologies.

AI innovation, policy and regulation has been a site of shared activity and aspiration for both countries. Korea were important stakeholders in the UK’s inaugural AI Safety Summit in November 2023, presenting their Digital Bill of Rights at the summit and expressing their support for transnational AI governance bodies. The next edition of the summit will be co-hosted by Korea and the UK in 2024. The UK’s leading role in AI safety is regarded positively in Korea, providing an immediate platform for closer coordination and joint thought leadership. 

On the semiconductors front, numerous interviewees recognised the natural complementarities between Korea’s relative strengths in manufacturing (hardware) and the UKs in core IP and chip design (software).

‘The UK has talent in the chip design space and Korea has capability in advanced manufacturing – there are interesting areas of collaboration there.’

‘We’ve got the chip design capability and Korea have got the manufacturing. Already, ARM work closely with Samsung and SK Hynix, and Samsung is an investor in Graphcore.’

‘Samsung have a historic presence in the UK with their design bases, the fact that they have facilities here already makes things easier.’

However, the software-hardware lens was seen by many as a starting point rather than the whole story. As per recommendations 5-10, there are six specific priorities for deepening joint R&D and investment between the two countries, three of which are discussed below:

• The UK has expressed a desire to develop advanced packaging capabilities to help build on strengths in compound semiconductors, photonics, and chip design. Meanwhile, Korea is motivated to steer new assembly, testing and packaging investment away from China. While countries like Japan, Vietnam, Malaysia, and India are often referenced in this regard, there is a distinct opportunity for Korean investment into a 2.5D/3D hybrid advanced packaging centre in the UK.
   
• UK research into novel forms of compute, such as neuromorphic computing, quantum computing, and reversible computing is highly regarded. None of these are ready to revolutionise the semiconductor industry today, but the longer time horizon for development and commercialisation is in line with Korea’s long-term investment planning. For example, the $230bn investment in the Yongin chip cluster will not come online until 2042, by which point newer compute architectures may have proved their viability. 
  
  
• There are specific sub-domains of hardware and systems security research which the UK excels in, and Korea needs more of. Mixed signal design capability is not as prevalent in Asia, nor is the knowledge base around power and high voltage applications. Moreover, the UK’s experience in radio frequency and advanced mixed signal technology would also likely be seen as desirable in Korea.

5.3 Focusing efforts

Hitting the right challenges and achieving buy-in from UK-Korean industrial and academic bases is vital. One interviewee emphasised that researchers and government officials will be further away from the gaps and bottlenecks in the market than industry, and that devising bespoke mechanisms for exchange between academia and industry in both countries could yield practical solutions more quickly. Lots of progress has been made at the government-to-government level via the Downing Street Accord, but this now has to be reflected in rates of FDI, dedicated funding for academic partnerships and joint ventures. 

Nonetheless, both governments will still have a crucial role in setting the tone: “the market can be reluctant to take on early-stage risk, so government has to play that bridging role to kickstart the bigger, longer-term projects.” On the UK side, speeding up the development of semiconductor coordination capability through a new UK National Semiconductor Institute will be integral for broader strategic discussions among shared value chain partners.

Governments usually set different priorities for agreements with different partners. Korea’s agreements with the US and EU respectively are an example of this: with the US there is a dedicated Export Control Working Group, whereas with the EU, members ‘only exchange views on export controls’ and there is greater emphasis on preparedness and crisis response. Moreover, priorities can shift over time. The EU-US Trade and Technology Council initially focused on short-term supply chain issues, moving into subsidies and export controls, before more recently landing on collective research to address the environmental footprint of semiconductor manufacturing. The lesson is that the structure of bilateral partnerships need tailoring and constant revision.

Both the UK and Korea share membership of various international organisations or alliances, and cooperating through these vehicles must complement bilateral discussions. One interviewee commented that, “the UK has a big influencing voice in the tech sector, and you do not leverage that just through bilateral engagements. You need to bring the bilateral engagement through with you into the bigger forums.”

Finally, numerous interviewees believed that the current golden period in UK-Korea relations provides an opportunity to advocate for each other’s inclusion in new multilateral fora. For example, one interviewee commented that now the UK has joined the Comprehensive and Progressive agreement for Trans-Pacific Partnership (CPTPP), it would be a timely and strategic move to advocate entry into the Indo-Pacific Economic Framework led by the US. On the other hand, the UK could put more energy into expanding the G7 to include countries like Korea and Australia: “Korea is at the centre of semiconductors, batteries, EVs, the whole range of emerging technologies. The G7 needs Korea’s cooperation in these sectors.”

In summary, the UK and Korea must now establish a virtuous cycle between collaboration at the R&D stage and industry-based cooperation. Both Governments have made clear their intent for closer technology and security alignment, but targeted action is needed to ensure their academic and private sector communities buy in to the same vision and deliver on implementation at the ground level. If successful, the UK-Korea model of semiconductor-AI cooperation could become the blueprint for others to follow. 

Extended Recommendations

The table below builds on the Summary Recommendations by containing additional detail on the rationale behind each recommendation and category of partnership. 

References

For the full list of references, please see the report PDF.