Global “Leading-Edge” Semiconductor Supply Chain Imbalance Is Placing Data Center Sustainability Objectives under Threat

Sustainability and energy efficiency are some of the key drivers in modern developed economies and this trend is increasingly impacting stakeholders within the Data Center Market (DCM). From a societal perspective, data centers (including cloud computing services) currently emit around 2.5% of global Greenhouse Gas (GHG) emissions, and although these services are essential to fueling the digital world, it is clear that greater regulation needs to be imposed to lower this market’s environmental impact. Meanwhile, from a financial perspective, recent energy price growth as a result of Russia’s War in Ukraine has made it fiscally important for operators to cut energy usage to reduce Operational Expenditure (OPEX), while ensuring competitive performance.

So far though, governments have done little to implement effective policy, leaving sustainability in the hands of the private sector. Cloud service providers (i.e., Amazon Web Services (AWS), Google Cloud Platform (GCP), and Microsoft) and other data center operators have introduced corporate governance that has a strong focus on sustainability metrics. In addition, Sustainability Linked Loans (SLLs) are becoming increasingly prominent, especially among colocation data center providers. SLLs provide operators with financial incentives to improve energy efficiency by offering lower borrowing rates and tax incentives, and penalizing companies when “green goals” are not met. EdgeConneX announced that it secured US$150 million in SLLs to support its Latin American expansion, while Flexential previously completed a new US$2.1 billion securitization offering with the requirement to demonstrate Power Usage Effectiveness (PUE) of 1.4 or below, as well as zero water usage in cooling.

Meeting these sustainability objectives will not be easy. Compute workloads are becoming more intensive with Artificial Intelligence (AI) and High-Performance Computing (HPC) increasingly becoming prominent. Stakeholders must channel substantial investment into energy saving technologies, such as cooling and construction. But to maximize PUE, stakeholders must deploy a larger proportion of “leading-edge” processors. Leading-edge processors (although not continuing to follow Moore’s Law) have a much higher transistor density than the “trailing edge,” reducing heat dissipation, consuming less power, and providing faster compute power. The leading edge is often classified as any processor less than 10 Nanometers (nm). This results in a far more energy-efficient chip, which is reflected in Taiwan Semiconductor Manufacturing Company’s (TSMC) claim that its 7 nm processor (leading edge) has 20% better performance and 40% power efficiency gains, as compared to its trailing edge 14 nm chip.

Although the European Union (EU) controls the leading-edge Semiconductor Manufacturing Equipment (SME) market, with ASML’s monopoly in Extreme Ultraviolet Lithography (EUVL) machines, and the United States dominates leading-edge chip design, neither the EU, nor the United States has leading-edge manufacturing capacity. Asia-Pacific dominates leading-edge manufacturing, with TSMC and Samsung controlling the majority of market share. This imbalance makes Western fabless and even Integrated Device Manufacturers (IDMs) completely reliant on Asia-Pacific for leading-edge manufacturing. TSMC is the global leader and has more than 1,000 fabless clients, including the majority of U.S. leading-edge chip designers.

NVIDIA is one of the key players within the U.S. fabless market designing leading-edge Graphics Processing Units (GPUs) for data center AI workloads. In 2021, NVIDIA shifted its HGX A100 GPUs to TSMC’s 7 nm, making TSMC its only foundry partner. In addition, even Intel (which, as an IDM, usually manufactures its own semiconductors) has turned to TSMC to manufacture its ARC GPUs on its 6 nm process. This decision reflects the key challenges within the leading-edge market—capabilities and cost. Intel highlights that in-house manufacturing would not be feasible and highly inefficient given the complexities, knowledge barriers, and foundry requirements to produce leading-edge chips.

However, this reliance on TSMC for leading-edge manufacturing capacity is becoming increasingly risky. China-Taiwan tensions continue to increase at a time of increasing U.S.-China trade disputes (focused on advanced semiconductors and manufacturing equipment, such as NVIDIA’s HGX A100 GPU). These concurrent factors could place substantial long-run risk on the relationship between U.S. fabless companies and TSMC. But, at present, Western players, given their own domestic manufacturing deficiencies, have very little option. The only other leading-edge foundry competitor, Samsung, is still “ramping up” its foundry business and would currently be unable to provide sufficient leading-edge manufacturing scale to accommodate Western leading-edge foundry demand.

Supporting Western sustainability efforts means ramping up leading-edge processor manufacturing now. So far, governments have begun to act by introducing the U.S. CHIPS and Science Act, as well as the EU Chips Act. These acts are spearheading a regional strategy to incentivize ecosystem development and manufacturing growth with Europe looking to meet its goal of 20% of global semiconductor manufacturing by 2030. But they’re not doing enough to ensure that long-term sustainability ambitions in the data center market can be ensured. Below, ABI Research provides some practical strategies to support leading-edge manufacturing development:

• Domestic Cooperation: Cooperation between policy makers and market stakeholders is vital to assess challenges and provide funding/structure legislation to loosen bottlenecks constraining domestic leading-edge manufacturing growth (e.g., access to equipment, manufacturing skills, and capital deficiency). This will also avoid “over-balancing,” as excess capacity can be equally dangerous to Return on Investment (ROI).
• International Cooperation: Governments should look to “friend shore” by building smaller supply chains with key allies to increase leading-edge manufacturing availability in the medium term.
• Diversify the Supply Chain: Fabless companies (especially in the United States) must work now to diversify the supply chain by working with Samsung (the only other leading-edge foundry). Early agreements will ensure that key companies can secure access to limited manufacturing capacity at potentially preferential prices.
• Leverage Regional Competitive Advantage in Design and SME: the EU and the United States have near monopolistic considerable control over two (of three) key elements. Ensuring that these resources are managed effectively to promote domestic ecosystem development will be a key part of rebalancing strategies. ASML’s monopoly in EUVL will be particularly important (as we have seen in the U.S.-China Trade war), given its complexity, price point, and 2 to 5-year supply cycle.
• Accelerate Government Timelines: Especially in the EU, companies have already set forth substantial investment in the leading-edge ecosystem (e.g., Intel’s investment in a mega fab in Germany). However, without EU approval of the Chips Act, these investments cannot proceed. Policy makers must be quicker to adopt legislation and make subsidies and grants available if they wish to make a tangible impact on supply chains in the near term.
• Don’t Forget about Skills: Western economies not only lack capacity, but they also lack access to human capital required to support manufacturing operations. Governments must look to support Science, Technology, Engineering, and Mathematics (STEM) programs with direct funding, as well as partnerships with companies, to ensure a linear path into this industry.