Ai's quantum leap: computing power explodes beyond human comprehension
The relentless march of artificial intelligence is no longer a gradual progression; it’s a detonation. Mustafa Suleyman, the driving force behind Microsoft’s AI ambitions, isn’t exaggerating when he describes it as an ‘explosion’ – a shift so profound it’s rewriting the very rules of technological advancement.
The numbers tell a terrifying story
Suleyman’s observations, gleaned from a decade immersed in the field, paint a stark picture. Computing power, he explains, has increased exponentially, leaping from 1014 FLOPS in early systems to over 1026 FLOPS today. This isn’t simply scaling; it’s a discontinuous rupture. It’s akin to watching a single step transform into a freefall.
He illustrates the process with a vivid image: a crowded room of people, each equipped with a calculator, desperately trying to process data. Initially, simply adding more calculators – more ‘people’ – yielded incremental gains. But modern AI, he argues, isn’t about brute force. It’s about orchestrating that entire room of calculators into a single, unified intelligence. A frighteningly efficient, and rapidly accelerating, collective.
Engineering breakthroughs fuel the fire
This dramatic shift isn’t happening in a vacuum. It's underpinned by three converging technological breakthroughs. Nvidia’s chips, for instance, have surged from a modest 312 teraflops in 2020 to a staggering 2.250 teraflops today – a 7x increase. And Microsoft’s Maia 200 chip boasts a 30% performance boost per euro compared to rival silicon.
Then there’s High Bandwidth Memory (HBM), effectively stacking chips into dense, vertical structures, tripling the bandwidth of its predecessor, HBM3. This eliminates the frustrating bottlenecks of the past, ensuring that GPUs are perpetually fed with data. Finally, technologies like NVLink and InfiniBand are stitching together hundreds of thousands of GPUs into colossal supercomputers – structures that dwarf even industrial ships – forming a single, massively parallel processing unit. It’s a demonstration of sheer, overwhelming scale.
A warning from the top
Even Eric Schmidt, the former CEO of Google, acknowledges the profound implications. “We’re experiencing between 10 and 15% of the effects of all this,” he cautions. The implications are already visible: a 50-fold improvement in processing speed for a 2020 task, achieved in under four minutes with current hardware – a figure that would have been wildly optimistic just three years prior, based on Moore’s Law.
The projections are nothing short of apocalyptic. Labs are currently expanding their computing capacity at nearly 4x annual rates, with a fivefold increase since 2020 anticipated by 2027. That’s a staggering 100 million H100 chips – a tenfold surge in just three years. Microsoft, Nvidia, and the looming specter of Artificial Superintelligence (ASI) are locked in a desperate race to harness this unprecedented power. And with the cost of renewable energy plummeting – a 97% decrease in battery prices over three decades – the energy constraints that once limited progress are rapidly dissolving.
Suleyman concludes with a chilling assessment: “We’ll continue to be surprised. The boom in computing is the technological narrative of our time, and it’s only just begun.”