Product1 publisher3 min readPublished
IonQ credits SkyWater's standard lines with cutting its chip design cycle to two months
Superion 256 opens for orders with deliveries due in 2027. The six tapeouts IonQ ran in half a year say more about whether its 2028 fault-tolerance date is real than the qubit count does.
The Product Desk · Product desk
What happened
- IonQ opened orders for Superion 256, a 256-qubit machine whose control chips are made on standard semiconductor lines, with first customer deliveries due in 2027.
- The chips come from SkyWater Technology, the foundry IonQ agreed in January to buy for $1.8bn, where it has now fabricated its first fully integrated 256-qubit processing units.
- IonQ says working with SkyWater's quantum foundry cut its design cycle from nine months to two and yielded 12 times more wafer lots over six months than its previous foundry partner.
- A Superion occupies a standard server rack, draws less power than a rack of GPUs, and connects to ordinary data centre cooling plumbing.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- decision Security owners now have a dated footprint published by a hardware vendor to schedule elliptic-curve migration against, and QuSecure's Buss aims it at code signing and certificate hierarchies rather than at bitcoin, which is a harder thing to defer than an open-ended quantum warning.
- cost IonQ has committed $2.875bn to a foundry and a control-electronics company before one customer machine has shipped, and that outlay sits behind whatever price a 2027 buyer is eventually quoted.
- capability A quantum system that lands inside existing rack, power and cooling budgets can be bought by the team that already procures GPU capacity, with no separate facilities programme attached to the purchase order.
- constraint By Buss's account the 2028 estimate is gated on error rates nobody has demonstrated at that scale, so the roadmap's credibility depends on fidelity results IonQ has yet to show on a full-size system.
IonQ ran six tapeouts on the Superion control chip in the first half of 2026 [5]. On the nine-month cycle the company says it used to live with, six serial tapeouts would have taken 54 months; at two months each they add up to 12, and they landed inside six, so at least two designs were moving through the fab at once [6][17]. That is what sitting on a foundry line buys, and it is a different purchase from a qubit record.
The bill is public. SkyWater cost $1.8bn to agree to [3], Oxford Ionics $1.075bn last year [7], which is $2.875bn pointed at wafer access and control electronics rather than at qubits [18]. Oxford Ionics brought Electronic Qubit Control, which puts qubit control onto the chip instead of into the laser arrays trapped-ion machines have generally relied on [7]. IonQ expects that one swap to take cost per qubit down by a factor of more than 300 across the roadmap [13].
The material supports a claim about repeatability rather than a performance comparison: the same ions and the same control electronics from 256 qubits up to millions, with Superion 10K adding CMOS on the chip and being the first generation built for Walking Cat, the fault-tolerant architecture IonQ published in April [2][11]. Chief executive Niccolo de Masi called it "the first quantum computer platform designed to be built by the hundreds rather than one at a time" [10]. The 99.99% two-qubit gate fidelity record came off the same design in October [8], not off a 256-qubit machine, and the first unit was pre-sold in the first quarter to a customer IonQ has not named [20]. Chris Ballance's dates are the checkable part: fault tolerance in an IonQ lab in 2027, a manufacturable commercial version in 2028 [12].
The figure that will outlive the launch is the resource estimate IonQ published the same day: 19,397 physical qubits and 25.7 days to run Shor's algorithm against secp256k1, the elliptic curve behind bitcoin, a footprint the company puts near the 2028 stretch of its hardware roadmap [14]. That is about 76 times the physical qubits in the machine now taking orders [19], and 25.7 days is elapsed runtime, which makes availability part of the specification in a way qubit count never is. Garrison Buss, chief technology officer of QuSecure, told SiliconANGLE the estimate assumes error rates nobody has demonstrated at that scale, "but the direction is not in doubt" [16], and that the thing worth planning around is elliptic-curve cryptography generally, since most of the internet's public-key math moved onto curves over 15 years and code signing and certificate hierarchies rest on them [15].
So the useful exercise this week has nothing to do with buying a quantum computer. It is sorting the ECC dependencies at hand by the year each one stops needing to be trusted, then noting who performs the migration for each: the buyer, or a supplier shipping on a schedule the buyer does not set. The rows dated past 2028 are the ones IonQ's own roadmap speaks to. Where the date is late and the control belongs to someone else, that is the work. The rest of it can wait for hardware that has not shipped.
What to watch
- Whether the SkyWater acquisition closes as agreed, and whether IonQ's ownership squeezes capacity for the foundry's other customers.
- Any independent confirmation of the two-month design cycle and the 12x wafer-lot figure, both of which currently rest on IonQ's own account.
- Fidelity data from a fully assembled 256-qubit system rather than from the shared design, and whether 2027 customer deliveries hold.