IBM reckons it’s finally nailed the architecture. In June it laid out a new blueprint that plugs the holes in its earlier plans and claims it can build a proper quantum computer before the decade’s up.
IBM quantum boss Jay Gambetta said: “It doesn’t feel like a dream anymore. I really do feel like we’ve cracked the code and we’ll be able to build this machine by the end of the decade.”
Google isn’t taking that lying down. It claims to have bulldozed one of the last major hurdles and says its hardware is also on track.
Google Quantum AI Julian Kelly said: “All the [remaining] engineering and scientific challenges are surmountable."
But just because the quantum physics bit might be sorted doesn’t mean it’s all downhill from here. There’s still a mountain of tedious, soul-destroying engineering to do.
Amazon Web Services executive Oskar Painter doesn’t expect anything useful for at least 15–30 years.
“We should not underestimate that engineering effort to scale,” he said, pouring cold helium on the hype.
Making this work means going from a couple hundred qubits to a million or more. These twitchy units only stay useful for fractions of a second before falling apart into noise and gibberish.
IBM’s experimental Condor chip, which packed in 433 qubits, ended up suffering from “crosstalk” between components.
Rigetti Computing chief executiv Subodh Kulkarni, called it “a bizarre effect we can’t control anymore... that’s a nasty physics problem to solve.”
IBM says it was ready for the interference and has moved to a different qubit coupler to reduce the mess. Manually tuning qubits, which worked at small scale, clearly isn’t viable once you scale up.
Google wants to slash component costs by 90 per cent to get its total machine budget to $1 billion (€920 million). It’s also made the most noise about error correction, which creates redundancy by copying data across multiple qubits.
“If you scale without error correction, you get a very expensive machine that outputs noise... and does not provide any value at all,” Kelly said.
IBM isn’t sold on Google’s surface code approach, which connects qubits in a 2D grid and needs more than one million of them to work. Microsoft has already walked away from that idea, calling it too messy.
Instead, IBM is going for low-density parity-check codes, which supposedly require 90 per cent fewer qubits. The catch is they need longer-range connections between distant qubits, which are a bastard to make stable.
Kelly said the new design “adds new levels of complexity” to an already unmanageable beast. IBM insists it’s finally cracked those long connectors, though.
Gartner analyst Mark Horvath said IBM’s machine design might work, but right now it’s just drawings. “They need to show they can manufacture chips that can do that,” he said.
Beyond the qubits, the real fight is untangling the spaghetti of wiring, fitting components onto a single chip, then stacking those into modules. All of it needs to be shoved into massive fridges hovering near absolute zero.
While superconducting qubits show the most progress, they’re also fragile and hard to manage. Alternatives using trapped ions, photons, or neutral atoms are more stable but face other problems like slower speeds and poor connectivity.
Universal Quantum chief executive Sebastian Weid said governments will end up backing specific tech stacks, narrowing the field to a few survivors who can go the distance.
US defence tech outfit Darpa is already sniffing around trying to spot winners. Meanwhile, several companies have started showing off exotic new qubit designs, some even claiming new states of matter might be the key.
That’s years off, but the old-school superconducting crew aren’t slowing down. “Just because it’s hard, doesn’t mean it can’t be done,” said Horvath, summing up the grim optimism driving the whole show.