Focusing on “speed” as the core metric of performance can lead to some own-goals in the world of tech. Intel became a category-king and dominated its era when its unalloyed religiosity on sustaining Moore’s Law was the nucleus of its business model. But as it focused on speed, pushing down the feature sizes of transistors, it forgot the “besides speed, are you checking other metrics”. From the flanks, new species of companies like Nvidia, Qualcomm, ARM, and AMD 2.0 emerged, and punted the dominance of Intel.
Yes, if your phone is fast, are you open to trade that speed if the phone battery can last longer? In electronics, that I^2R in CMOS transistors and broad power dissipation will define the usability of products, well ahead of the speed you can get things done. So, when those new ARM-anchored processors came out with better power efficiency, markets changed in the mobile internet era.
Then we moved into the AI fledgling era, and Nvidia, AMD 2.0, etc became the new Intel signing the songs of speed, processing capacity, etc as they make the case why high compute is the holy grail to unlock the heart of the AI age. But with that fixation on computer power, the crusaders like Microsoft, OpenAI and Google concluded that to win, more computing power is the denominator. Yes, efficiency and optimization were not really vital.
But from the flank, a small Chinese company named DeepSeek went for that metric of optimizing everything to use less computing power. Magically, we’re learning again that speed, computing power cannot be the only core metrics if one expects the system to be symphonically architected like an orchestra that delivers nice music-of-AI-services.
In the industry, one of my core roles used to be creating the electronics for MEMS systems. In pacemakers, when you engineer the MEMS-based gyroscope and accelerometer, you have to be really great on power management so that the electronics can be implanted in a human and stay there for more than a decade! That disciplined approach of designing for usage where scarcity defines the operating environment is the mindset which must be used for this AI age.
Simply, from the $500 billion Stargate to building nuclear power plants, AI leaders must chill and say, “could there be another way”. DeepSeek has provided a reset to rethink the future of AI at the foundation model, large language model and generative AI levels.
Tech stocks recovered Tuesday, with the Nasdaq rising 2%, after China’s DeepSeek AI model sent shockwaves across the sector and cast doubt about U.S. dominance in the artificial intelligence race. Nvidia shares rebounded, climbing almost 9%, following Monday’s 17% rout, which erased a record $600 billion in market value. With U.S. tech giants reporting earnings this week, one analyst sees a “real potential shift in the underlying narrative” as investors assess the high valuations of many AI-linked stocks.







