The hum of the servers is a constant thrum, a low-frequency heartbeat in the otherwise silent data center. Engineers at a major chip manufacturer, let’s call them ‘NovaTech,’ are hunched over screens, poring over thermal tests. It’s early 2024, and the pressure is already on. Their M300 chip, slated for release in late 2025, needs to outperform the competition, especially those from across the Pacific.
Marco Argenti, Goldman Sachs’ CIO, has a front-row seat to this. He anticipates significant shifts by 2026, as he recently told Fox Business. “We’re going to see a ratcheting up of the U.S.-China competition, particularly in the AI space,” Argenti explained. He points to rising costs, not just in chip manufacturing, but also in the training of large language models (LLMs). The race to build the most powerful AI, he suggests, is becoming a financial arms race.
The implications are far-reaching. The workforce will need to adapt, with Argenti predicting the rise of “personal agents” – AI assistants that manage tasks and information, potentially reshaping job roles. Meanwhile, export controls and domestic procurement policies are becoming battlegrounds. The U.S. government, for instance, has placed restrictions on the export of advanced chips and related technology to China, aiming to slow its AI development. Beijing, in response, is pushing for domestic chip development, pouring billions into companies like SMIC.
This isn’t just about chips, though. It’s about the entire AI ecosystem. Training an LLM requires massive computational power, which translates to a high demand for GPUs. The supply chain is strained, with companies like TSMC operating at full capacity. Analysts predict that the cost of training a state-of-the-art LLM could reach tens of millions of dollars by 2026, if not more, depending on the model’s complexity and the data used.
The NovaTech engineers are acutely aware of these realities. They’re not just designing chips; they’re navigating a complex web of geopolitical tensions and market forces. One engineer mutters, “If we don’t hit the performance targets, the entire project gets pushed back, or maybe that’s how the supply shock reads from here.”
The competition is fierce. The expected capabilities of the M300 are ambitious, but so are the timelines. The race is on, and the stakes are higher than ever, by 2026.
About the Author
