U.S. Says “No” to Supercomputers for Russia’s Nuclear Weapon Labs

In mid-October, the Clinton Administration turned down a bid by Russia’s nuclear weapon laboratories to buy U.S. supercomputers. The sale was opposed by experts inside the U.S. government who feared that the machines would be used to improve Russia’s nuclear arsenal after the agreement this year to end nuclear testing. Because supercomputers can simulate the conditions within an exploding nuclear warhead, they can help maintain, refine, and develop nuclear weapons without testing.

The October decision occurred at a White House meeting called to respond to media inquiries. The inquiries were triggered by the Subcommittee on Military Procurement of the U.S. House Committee on International Security, which had released a report by the U.S. General Accounting Office in early October on the proposed exports. The report concluded that one of the computers, the model Convex SPP 2000 (recently renamed the “X-Class”) made by Convex Computer Corporation, a subsidiary of the Hewlett-Packard Company, could operate at a speed of 35,000 MTOPS (million theoretical operations per second), at least ten times faster than any machine currently available in Russia. The Subcommittee had requested the GAO report in March after Russia’s interest in supercomputers was revealed by a February editorial in the New York Times by the Wisconsin Project on Nuclear Arms Control.

The SPP 2000 posed a thorny policy question for the Clinton Administration. In a September 9 letter to Energy Secretary Hazel O’Leary, Russian Atomic Energy Minister Victor Mikhaylov admitted that the computer would be used to “confirm the reliability” of the Russian nuclear arsenal. Previously, Mikhaylov’s Ministry had claimed that it wanted supercomputers to do environmental modelling. “Reliability” testing insures that a nuclear weapon will explode with its intended force. “Safety” testing insures that the weapon won’t explode accidentally. The Department of Energy had previously assured Congress that reliability testing was outside the scope of U.S. nuclear cooperation with Russia. Representative Duncan Hunter (R-CA), chairman of the Subcommittee, said in a press release that he was “astounded and dismayed” by the possibility that American computers could go to “Russian nuclear laboratories for the purpose of improving the quality of Russian nuclear weapons.”

Mikhaylov’s letter also made it more difficult to approve the export of two other U.S. computers to the Russian Atomic Energy Ministry. The first, a Convex Model SPP 1200 XA-32, operating at 4,564 MTOPS and priced at $3.8 million, was destined for the Institute of Experimental Physics in Moscow, a Ministry of Atomic Energy site partially closed to outsiders. The second, an IBM 9076 Model 304 (known at the SP 2) operating at 780 MTOPS, was destined for Arzamas-16, a closed nuclear weapon laboratory where Moscow’s first atomic and hydrogen bombs were built. According to the Russian news agency Itar-Tass, scientists at Arzamas are still using simulations to develop new nuclear warheads. U.S. experts familiar with the export license applications tell the Risk Report that the IBM machine was of concern because it could be scaled up to a speed of 7,000 MTOPS or even higher by simply adding more processors. The stated use of both machines was environmental modelling and nuclear safety.

Convex had first applied in November 1995 to export the SPP 1200 XA-32 to Chelyabinsk-70, a Russian laboratory that claims credit for developing most of Russia’s nuclear warheads, including the world’s most powerful hydrogen bomb. That application was submitted with two others to export Convex machines to Arzamas-16, one operating at 1,630 MTOPS and costing $1.5 million, and a second operating at 1,870 MTOPS and costing $2.6 million. According to a company spokesman, Convex withdrew the two latter applications in May 1996 and changed the destination of the SPP 1200 XA-32 from Chelyabinsk to the Institute of Experimental Physics in Moscow. The spokesman tells the Risk Report that Convex was ready to adopt a security plan to prevent the computers from being misused and was “willing to do anything necessary to protect the national security of the United States.”

Apparently, IBM first learned of the Convex applications from the February editorial in the New York Times. The Times article was the first public disclosure that Convex was hoping to make computer sales to the Russian laboratories. Soon afterward, IBM filed its application to export the SP 2 computer.

By the time the computer cases reached the White House in October, several U.S. federal agencies had taken positions on the Convex and IBM applications. A senior U.S. official who surveyed the various agencies’ views tells the Risk Report that the Departments of Defense, Energy, State and the Arms Control and Disarmament Agency (ACDA) had voted to deny the Convex machine, whereas the Department of Commerce took no position. The Departments of Defense and Energy also had voted to deny the IBM machine, whereas the Department of Commerce and the Arms Control and Disarmament Agency took no position and the Department of State failed to respond when asked to state its views.

The result of the White House meeting was to instruct the Commerce Department not to approve the applications. The applications were then “returned without action” to the applicants. The Commerce Department issued a statement saying that it had returned the applications “due to the inability of the United States Government to obtain adequate information at this time from the end user and the Russian government to insure that the commodity will be used in accordance with U.S. law….” The decision by Commerce leaves open the possibility that the applications will be resubmitted at a later date, probably after the November elections.