History doesn’t repeat but it rhymes. At Metis we are partnered with some of the savviest hedge fund managers in the world. They have a keen understanding of the lessons of history and the impact on the future of businesses and markets… and how to profit from it.
So when I got an early morning text with this article “ No more Moore? So, what then for microchips? And for China? “, by Michael Power, with the simple “worth the read, matters for the AI trade, and reminds me of your strategy”.
I read Michael’s whole dissertation-length piece twice, because guys with a long track record of minting money for a living don’t generally share their printing press ideas.
Not surprisingly, the manager was right, there are implications for the AI trade in the article with a detailed case made for the potential global winners and losers. Metis’ core design philosophy and strategy is to empower our customers to be agnostic to the outcome and benefit from the accelerating pace of innovation in AI.
Power describes his view of diminishing returns on capital and performance of Moore’s law and the current trajectory of silicon-based Nvidia GPUs supported by EUV lithography capabilities that pushing the bleeding edge of physics…perhaps over a cliff. I’ve read about the death of Moore’s law for a while, and the technical debate around GPUs, TPUs, 3D chiplets, photonic processors, and AI ASICs rages today, which Power weaves into his narrative.
However, Power then turns his focus to the difference in strategy and philosophy for China versus the West in scaling AI.
Strategy inevitably trumps technology, in our mind, particularly in the mass adoption phase.
Power argues the AI arms race where Silicon Valley is trying to make the central brain ever more powerful, whether it’s the chip makers or the foundation model providers could be disrupted.
The West is catering to the most demanding customers and use cases, while ignoring the emerging new vectors of performance that mass market adoption needs and values, which are power efficiency, cost, and scalability.
He quotes sources that compare AI scaling to a “war of factories”, akin to WW II, where the American arsenal of democracy prevailed over the Axis powers by producing more tanks, planes, and ships that overwhelmed the adversary, despite the latter having superior technology in some of those categories.
Power explains China has chosen to compete on a different playing field for scaling in AI, using “Distributed Intelligence” to compensate for the lack of access to Western tech, while the West is focused on “Concentrated Intelligence” given its incumbent position.
Power’s argument harkens to the disruption of Intel by Apple in the mobile/laptop market where Apple’s RISC chip design philosophy, while initially inferior to Intel’s CISC processors, ended up winning over mobile device makers that valued power efficiency, size, and scalability over raw compute power which eventually Apple outperformed on as well. Indeed, Power highlights the importance of China adopting RISC-V architecture in its future AI scaling strategy.
At Metis our fundamental design philosophy is not to make the central brain smarter with ever greater inference compute or tapping the latest LLMs but to make the problem easier by autonomously breaking down even the most complex task into simpler tasks run recursively in parallel by many smaller, cheaper, readily available LLM or SLM brains.
We’ve chosen to help our customers compete on a different playing field delivering on what they value most: security, efficiency, privacy, and reliability.
We welcome the factories of the AI war producing more material in the race toward mass adoption and commoditization, as we can enable our customer to harness the spoils of war and future proof their technology decisions.