In 2024 Japan's AIST bought a superconducting quantum computer through an open, scored tender for about ¥6B. This autumn it added 192 qubits from the same supplier for ¥1.67B with no tender, on a sole-source contract citing "compatibility". Very important to secure initial system sales, even if crap margins
@Sujeet Worth adding from the GAO report: a lot of agencies said testing was premature because vendors still hadn't built the new algorithms into their products. Hard to fund a migration when you can't price the tools yet!
Two European universities awarded quantum hardware contracts last week - each got exactly one bid. One paid about €450k for qubit control electronics. The other bought a 10-qubit chip and controls to go into its own lab cryostat. A lot of real quantum spending looks like this: parts and lab kit, from a very short list of suppliers.
The quantum computer unveiled in Chicago last week shows how these deals get stitched together. The state put in a $25 million grant, paid to the University of Illinois trustees, for equipment and installation. The University of Chicago hosts the machine for now, and it moves to a state-backed park once that's built.
https://t.co/WGT2NpBEdC
@ben_j_todd I guess the challenging thing is that the timelines are getting shorter but the required infrastructure to keep up the pace is damn expensive
@Hesamation Fascinating on the AI side too. The simple takeaway is that quantum needs a hidden pattern to be dramatically faster on these random yes/no problems. Being quantum alone doesn’t do it.
@LukeGromen This is fascinating. In plain English: quantum doesn’t get a huge speedup on random yes/no problems just by being quantum. It needs a hidden pattern to exploit, like the repeating pattern Shor uses. This doesn’t overturn Shor
The result concerns whether quantum computers can be exponentially faster on random inputs. For yes/no questions measured by how many times you read the input, the reported answer is no: classical can approximate with small error on nearly all inputs using polynomially more reads. Structure still matters, as in Shor.
This is fascinating, especially coming from someone who has spent time on the problem. The eventual commercial reading I take is similar: if it holds, dramatic speedups of this kind need structure in the input, as with Shor's periodic pattern. It does not tell us what a particular machine can do on a particular commercial problem, but it does sharpen the question every claimed speedup should answer - is that a fair way of describing it?