AI Chips: The New Gold
Shree Priya Thakur | 5 July 2024
Summary
AI chips, the latest advancement in semiconductor technology, are driving the next wave of digital innovations by outperforming legacy chips in the domain of memory, logic, and accuracy.
As the driving force behind generative AI, autonomous vehicles, robotics, and Edge AI, AI chips will serve as the foundational elements of AI digital infrastructures, with the market projected to reach a value of USD 258 billion by 2033.
The concentration of the chip supply chain in a few companies like TSMC and ASML has sparked geopolitical concerns and fear of weaponised interdependency, necessitating a global consensus and strategic alliances.
As the world begins to understand the rise of semiconductor chips, the tech world’s latest silicone slice is already on the market. Artificial Intelligence (AI) chips are the micro elves making macro decisions, elevating the next set of digital superstructures. Branded as the world’s most advanced technology which “no one has heard of”, what makes these chips a silent revolution?
AI chips are specialised hardware used to develop artificial intelligence systems. It is an umbrella term for different types of chips including Field Programmable Gate Arrays (FPGAs), Graphics Processing Units (GPUs), and Application Specific Integrated Circuits (ASICs) among others. Simply put, chips or microchips are typically manufactured at a microscopic scale and embedded with transistors ie. minuscule switches that control the flow of electrical current in a circuit. With the help of these transistors, the chips can perform two very important functions: 1) memory ie. storing data, and 2) logic ie. processing the data.
What makes AI chips different and necessary? AI chips in particular, specialise in the function of logic. By performing the function of logic ie. synthesising data, they will be instrumental in handling intensive data processing. AI chips are equipped with smaller transistors rather than legacy chips, allowing them to deliver more computations, consequently increasing their speed and reducing their energy consumption. They mark a departure from the legacy chips in two other realms 1) intelligence and 2) accuracy. These chips are intelligent due to parallel processing, meaning they can multitask and perform multiple calculations simultaneously. They are accurate as they are tailor-made for intelligence, for instance, they can be customised to meet the requirements of specific AI models by performing cognitive tasks like image recognition and natural language processing. Commanding a market of USD 258 billion by 2033, AI chips will be the core aspects of future Telecom, Banking, and Financial systems.
Currently, these chips are powering AI breakthroughs like ChatGPT, autonomous vehicles, robotics, and Edge AI ie. making AI processing possible on smart devices like cameras and kitchen appliances. As the basic building block of AI systems, these chips provide the “brains” by facilitating neural networks ie. the fundamental algorithm structures behind AlphaGp, Midjourney, etc. As important as they are, these chips have only recently started to make headlines following Nvidia’s announcement of its Blackwell series of AI chips which can power data centres, of core importance in training the next generation of GPT, Anthropic’s Claude, and Google’s Gemini models.
Yet, the recent buzz around AI chips is courtesy of a deeper geopolitical concern. This emerging technology forming the nucleus behind the rest of the 21st century’s tech advancements is almost exclusively produced in a single facility. Rewording the definition of leverage, the asymmetric power held by the Taiwan Semiconductor Manufacturing Company (TSMC) is signalling a shift from state to MNCs ensuring all eyes are on the Ilha Formosa ie. Taiwan. TSMC today makes all of the world’s advanced chips, catering to Google, AMD, Microsoft, Amazon, SambaNova and other credible transnational corporations. Not only is there a monopoly, but TSMC is located on the western coast of Taiwan (approximately 180km away from mainland China) where China is conducting military exercises at an unprecedented scale. On this account, Chris Miller, author of the “Chip War” holds that “no other facet of the economy is so dependent on so few firms.” Though parts of the semiconductor supply chain are scattered across the world, they are equally concentrated. For instance, 100% of the world’s extreme ultraviolet lithography machines (used to make the advanced chips) are manufactured in ASML, a company in the Netherlands. Further, there are only three companies in the world, TSMC, Samsung, and Intel which can build advanced chips. TSMC, in this context, becomes a global force.
What facilitates this monopolistic dynamic? Thinking critically, this is more than a “right place at the right time” fluke. The chip fabrication industry is inherently built around an economies of scale model coupled with a heavy dose of “Moore’s Law” (the number of transistors on a microchip will double every two years at minimal cost). This makes a “winner takes all” dynamic possible as there is high risk and high return; even a single speck of dust can ruin a chip production cycle, wasting millions of dollars. This leads to the concern of high upfront capex. Manufacturing chips requires periodic investments, for instance, TSMC is expected to invest USD 100 billion in only three years. Previously, it spent USD 20 billion on the “Fab 18” to create Nvidia’s H100 chips. For budding startups, this level of expenditure is difficult to initiate and even harder to match. Furthermore, what sustains a company of this magnitude is a grand alliance. Its investments and reputation have cemented long-term partnerships across the chip supply chain, allowing TSMC to influence industry standards. It has the power to determine how products of Cadence, ASML, and Nvidia’s can be made compatible with the TSMC processes, facilitating supremacy in the chip ecosystem.
Barring all complexities, the world’s new gold has an intricate mix of opportunities and challenges; opportunities that will usher in a new era of intelligence and challenges that require TNCs to be given a seat at the global roundtable. AI Chips- the crunchiest tidbit where every byte is out of this world.
Umberto/Unsplash
Forecast
Short-term
Politically, the manifestation of a “silicon shield” is highly likely. China’s heavy reliance on Taiwan for chips (where 70% of all chips in China are made by TSMC), can deter Sino-expansionism and the prediction (by Washington) that China will invade Taiwan by 2027. Any politico-economic paralysis of Taiwan can jeopardise the global progress of AI, thus the United States is highly likely to have a vested interest in Taiwan’s security. In economic terms, AI chip start-ups will likely emerge despite the odd chances of ever replacing TSMC. Startups are inherently more agile and risk-taking, facilitating greater innovation.
Medium-term
Given the pre-eminence of AI chips, public-private collaborations between states and MNCs are highly likely (on the lines of an Apollo space program). For instance, Open AI is trying to raise USD 7 trillion for chipmaking, a breeding ground for US economic interest. Furthermore, as technologies mature, we will likely shift to AI chip alternatives such as ‘chiplets’ which break the different components of an AI chip into different blocks. In the tech domain, experimental use of Lidar sensors has started suggesting an eventual replacement as light is more efficient in transmitting information than electrical signals.
