Did Moore’s Law Die with Gordon Moore?
Peter Connolly (PPS ‘23)
In 1965, a young engineer by the name of Gordon Moore hypothesized that the number of transistors on a microchip would double every two years. The enduring veracity of Moore’s Law would usher in the digital age and transform nearly every aspect of our society. The implications Moore predicted came to be true as over the past sixty years, chips have doubled in power and become cheaper despite their increased abilities.
To many, this may sound like a lot of complex jargon, so I’ll break it down for you. A transistor is the tiny node on a chip that relays either a 1 or a 0, the binary language of all digital machines. The more transistors a chip has, the more powerful it is and the more applications and tasks it can compute. So why does this matter?
Building Walls Around China
Relations between China and the U.S. have continued to worsen over the past year, as concerns over China’s robust military modernization effort and aggressive AI development has brought heightened scrutiny over western companies’ exposure to the Chinese market. There are two concerns that have prompted the most stringent and comprehensive export controls on China to date. The first concern is China’s implementation of Digital Authoritarianism. According to Liza Lin, a WSJ reporter and author of Surveillance State China has aggressively developed AI systems to monitor, categorize and track its 1.3 billion citizens using sensors and millions of surveillance cameras. Reporters such as Liza Lin have exposed western companies’ role in building out these systems, including the fact that the AI and surveillance system behind Digital Authoritarianism runs on western designed and manufactured semiconductors. Moreover, China is not just using it at home but selling it to other authoritarian dictators abroad as part of their foreign policy strategy to provide a viable alternative to the U.S. led world order.
The second is China’s Civil-Military Fusion policy that allows for the military to take any commercial innovation and use it to advance and develop China’s military capabilities. The commission has removed all barriers between China’s commercial sector and its defense industrial base, ensuring technology given to Chinese companies that has any military application will be appropriated and implemented by the People’s Liberation Army. This has given the Biden Administration and both parties on Capitol Hill an enormous justification to implement these bans in the name of U.S. national security and China’s aggressive push to domesticate its semiconductor manufacturing process illustrates their dependency on western chips. National Security Advisor Jake Sullivan laid out the Biden Administration’s strategy in a speech to reporters at the White House, stating that China could “use our technology against us or their own people” as justification for the bans. He specifically cited “the foundational nature of certain technologies, such as advanced logic and memory chips” as reasoning for the widespread export bans. However, as the Biden Administration tightens restrictions on China’s access to advanced semiconductors, it’s important to recognize the impact on the global supply chain.
The Geopolitical Impact on Moore’s Law
The globalized semiconductor supply chain that we have today formed out of a result of economics not geopolitics. The global supply chain reduced costs and allowed companies to specialize on specific parts of the manufacturing process, resulting in higher margins for companies up and down the value chain which translated into more investment in R&D. This created a virtuous cycle of investment and innovation that brought down cost and increased computing power, fueling Moore’s Law nearly 60 years after Gordon Moore’s hypothesis. Moore’s Law is perhaps the most direct proof we have of the economics of globalization working and the positive corollary effects it can have on technological innovation as it provides a 60 yearlong path that we can track.
What Gordon Moore did not predict in 1965 but ended up becoming true was that the increased computing power provided the physical infrastructure for the development and creation of AI systems such as Chat GPT. In 1965, the only logic chip, or chips that ‘ran’ computers and processed the algorithms were CPUs (Central Processing Units). However, today there are three other types of chips that are essential for AI development. GPUs (graphic processing units), and AI ASICs (Artificial Intelligence Application Specific Integrated Circuits). All these chips would not exist in their current form without Moore’s Law. GPUs and CPUs are crucial to training AI models such as Chat GPT, helping it understand prepared data so that it can go out and perform its assigned task. The speed and effectiveness of its training depends on the semiconductors powering it. AI ASICs are semiconductors designed specifically for an AI program and represent the intersection of software and hardware. They are essential for advanced AI systems and expensive to make as their specific application purpose makes them difficult to manufacture at scale. China’s market is one of the fastest adopters and users of AI technology and U.S. companies can no longer sell to them.
Reduced profit margins will undoubtedly have an impact on R&D expenditures, as two of the industry leaders in AI logic chips have already felt the impact of the bans (AMD and Nvidia), which will have significant negative implications for the global state of innovation. A former senior executive at Intel confirmed this intuition, stating that R&D innovation would be one of the first things to suffer in a downturn. The source highlighted Intel Labs, an innovative research laboratory exploring advanced semiconductors in new fields such as quantum computing as one of the first departments within Intel that would receive funding cuts in any downturn as company executives seek to implement costing-saving measures across the board.
According to SEC filings, Intel has earned 23% of its annual revenue on average over the past 10 years from China, including 21 billion dollars in 2021. Nvidia had a 21% average annual revenue exposure to China the past 10 years while AMD faced a 38% revenue exposure to Greater China, including Taiwan, in 2022.
There is growing evidence supporting the national security concerns of many policymakers, providing a strong justification for the Biden Administration to enact these bans. However, its impact will be felt by American consumers and the U.S.’s growing demand for AI powered systems.
The CHIPS Act has mitigated the impact of many of these restrictions, providing enormous subsidies to onshore manufacturing within its 52 billion dollars of funds along with 13 billion dollars in R&D funding. However, for a 440-billion-dollar annual market as of 2021, the funds from the CHIPS Act are a drop in the bucket. U.S. policymakers are aiming to both ‘build walls’, or cut off China’s access to advanced chips, and ‘run faster’, or accelerate innovation to retain the U.S. current technological advantages over China. If these policies sound contradictory, it’s because they are. It is difficult to accelerate innovation while cutting off the revenues, and by extension R&D funding, of the companies driving semiconductor advances. Moore’s Law was at the center of the run faster approach. Doubling the number of transistors on a chip ensured that software developers had the hardware to build powerful AI systems and integrate them into our society to the point where Chat GPT is now the most prolific writer of high school English essays in the country.
My source at Intel confirmed that Moore’s Law, by its own definition had around 2017, but Intel was “redefining Moore’s Law on the fly” to match their own level of innovation. Five years ago, Intel’s ability to double the number of transistors on a chip fell to every three years.
This insight aligns with public statements made by senior executives at Intel. When Riccardo Masucci, Global Director of Private Policy at Intel, came to Duke and gave a lecture about the company and the future of the semiconductor supply chain, he was asked about the continuation of Moore’s Law at Intel. Masucci responded by stating that the law is alive and well at Intel and its engineers continue to innovate its chips in line with its overarching principles. There was no mention of the doubts expressed by the former senior executive at Intel.
Implications for AI
The conversation around Moore’s Law has shifted from node size and the number of transistors to other metrics of performance. One of the most important ones, for example, is performance per watt.
These innovations used to be a supplementary and accompanying innovation to Moore’s Law because as chips got smaller and more powerful, they also got more power efficient and cheaper because each chip could do more. However, as Moore’s Law begins to sputter out, those performance metrics will become the singular focus of chip designers. As AI powered systems consume increasing amounts of power to crunch complex algorithms, even a 1–2% improvement in power efficiency across a data center with thousands of chips would have an enormous effect and determine how many systems AI can be integrated into, from a drone to a smartwatch. However, the question of whether in a world without Moore’s Law, chip advancements around power and cost efficiency will be as profitable as producing chips with double the computing power of the previous generation of chip remains to be seen.
What can we do?
A global supply chain that excludes China but reaches a wide range of countries. It is fantasy to think the U.S. can domesticate the entire supply chain in a cost-effective manner that does not hurt American consumers and hamper innovation. Instead of onshoring manufacturing, we should be friend shoring manufacturing. That means placing factories in our partners and allies and that may require U.S. taxpayer funds to accelerate that transition. It’s the job the President and senior administration officials to stress the need for a secure but still globalized supply chain for both American consumers’ wallets and our national security and state of innovation. This will ensure the financial security of leading western semiconductor companies so that they can both reliably supply the U.S. and its allies with advanced chips while also continuing to drive innovation. Moore’s Law may be over or in its terminal phase but that doesn’t mean the U.S. should get complacent.
Peter Connolly (PPS ’23) is a Public Policy Undergraduate at Duke University’s Sanford School of Public Policy.
Click here to read his larger report, “The Battle For Chips: Semiconductors Crucial Role In AI Development And Its Implications For U.S.-China Strategic Competition.”
This content does not represent the official or unofficial views of the Sanford School, Polis, Duke University, or any entity or individual other than the author.
