The ASCI Red supercomputer first broke the teraFLOPS barrier in December 1996.
On May 25, 2008, an American military supercomputer built by IBM, named ‘Roadrunner’, reached the computing milestone of one PetaFLOP by processing more than 1.026 quadrillion calculations per second The budget was $100 million.
A 20 PetaFLOP supercomputer is being completed in 2012 that uses 6 megawatts of power. The budget is believed to be over $100 million.
Normally 12 years is expected between supercomputers that are 1000 times faster. The NSA appears to be shaving 4 to 5 years off the schedule by spending about nine times as much money and increasing electrical power by ten times.
A similar time shaving could be possible for the ZettaFLOP supercomputer by further increasing budget to about $9 billion and power supply to 600 Megawatts. The ZettaFLOP supercomputer could then be brought forward from 2030 for a $100 million system to 2022 using $9 billion and 100 times the power. There would be a lot more project risk and construction schedule risk and the need for construction innovation.
600 Megawatts would need to be a dedicated nuclear powerplant with backup power and other electrical grid and storage support.
It would still be very challenging and there would still be technical hurdles. However, more budget and more power could enable brute force to surmount the last two orders of magnitude in improvement. A concern would be whether time would be lost on a project of that size which would wipe out the hoped for acceleration in capability. It would be depending upon improvements in managing logistics and project efficiency. The project risks would increase.
The kinds of modular construction innovations that are being introduced by Broad Group for fast construction of buildings would need to be used to accelerate the construction of a computer facility of substantially larger scale.