Testimony: Supercomputer Export Controls – 1999

Testimony of Gary Milhollin

Professor Emeritus, University of Wisconsin Law School and
Director, Wisconsin Project on Nuclear Arms Control

Before the House Committee on Armed Services

October 28, 1999

I am pleased to appear before this distinguished Committee to discuss supercomputer export controls. I direct the Wisconsin Project on Nuclear Arms Control, a research project here in Washington that is devoted to tracking and inhibiting the spread of nuclear weapons.

I would like to begin by congratulating the Committee for taking the lead on what is now known as the “NDAA process.” I am proud to say that back in the spring of 1997, I recommended to the Committee that it change the law to reinstate the controls on high performance computers that the Clinton Administration had dropped in early 1996. The Committee then amended the National Defense Authorization Act to accomplish that purpose in late 1997. The existing notification process was the result.

It is clear beyond any doubt that the Committee did the right thing. The notification process has worked brilliantly. It has stopped a number of dangerous exports without imposing any significant burden on American industry–which is the very definition of a good export control system. It is an example of government regulation at its best.

The testimony by the General Accounting Office today, and the study it released last month, prove that the present control level should be retained at 2,000 MTOPS. If it is increased next year, as the Administration proposes, we can expect hundreds of American high-performance computers to wind up in foreign programs to build weapons of mass destruction.

To see how successful the NDAA process has become, the Committee should consider what would have happened without it.

Let’s start with Digital Equipment Corporation, recently acquired by the Compaq Computer Corporation. Digital applied for permission in 1998 to sell a high-performance computer to the Nuclear Power Corporation of India. The Nuclear Power Corporation builds and runs a number of reactors outside international inspection that produce plutonium available for atomic bombs. As we all know, India tested a number of atomic bombs last year. The application was wisely denied after objections by several federal agencies. The question therefore is this: would Digital Equipment Corporation, or its successor Compaq, be happier today if its products were helping India make plutonium for atomic bombs? That is what would have happened without the NDAA process.

Digital also applied for permission to sell a supercomputer to the Harbin Institute of Technology in China. The Harbin Institute is overseen by the China Aerospace Corporation, China’s principal missile and rocket manufacturer. Harbin itself makes rocket casings and other components for long-range missiles. This application too was denied after objections by the Arms Control and Disarmament Agency and the State Department. Again, the question is: would Digital or Compaq be happier today if their equipment were helping China make long-range missiles?

In a third case, Digital applied for permission to sell a supercomputer to the Weizmann Institute in Israel. The Weizmann Institute performs research on the high-energy physics and hydrodynamics needed for nuclear bomb design, and was the birthplace and scientific center of Israel’s nuclear effort. The application was blocked by being “returned without action.” The question is the same: would Digital or Compaq be happier today if their equipment were helping Israel’s nuclear weapon program?

If we had not had the NDAA process, Compaq Computer Corporation would now look out across the world and see machines made by a company it acquired operating at nuclear weapon or missile sites in Asia, South Asia and the Middle East. Is that something that Compaq wants to be known for?

And how much money are we talking about? The sale to India was valued at $250,000, the sale to China at $348,000, and the sale to Israel for even less. Compaq has an annual revenue of about $31 billion. Why would a $31 billion corporation want to supply A-bomb and missile plants in three sensitive regions of the world for a few hundred thousand dollars? That makes no business sense at all. Compaq’s reputation is worth a lot more than that. Compaq should be praising the NDAA and lobbying hard to preserve it.

We should also consider Silicon Graphics, Inc. The members of this Committee no doubt remember that Silicon Graphics illegally sold a set of high-performance computers to a leading Russian nuclear weapon laboratory in 1996. That sale was one of the scandals that led the Committee to amend the NDAA in 1997. After the NDAA took effect, SGI applied for permission to sell a high-performance computer to India’s Tata Institute of Fundamental Research, which is part of India’s nuclear program and on the British government’s list of entities linked to mass destruction weapon programs. It is also listed by the U.S. State Department as “involved in nuclear or missile activities.” The application was returned without action.

SGI also applied for permission to sell a high-performance computer to India’s Space Applications Center. The Center is part of the Indian Space Research Organization, which develops India’s largest rockets, and is listed by the British government as linked to mass destruction weapon programs and by the U.S. State Department as involved in nuclear or missile activities. The application was wisely denied after an objection by the Arms Control and Disarmament Agency.

For these two sales combined, SGI would have received only $500,000–a pittance compared to its annual revenue of $2.7 billion. Would SGI be happier today if its computers were operating at Indian nuclear and missile sites? SGI too should be praising the NDAA for saving it from being embarrassed over economically insignificant sales.

The third company to be considered is Sun Microsystems, Inc. Sun applied for permission to sell a high-performance computer to the Indian Institute of Technology. The Institute is helping to develop India’s biggest rockets–ones that could carry India’s new nuclear warheads to intercontinental ranges–and it supplies the Indian defense industry with high-technology electronic components. The application was returned without action.

Sun also applied to sell a high-performance computer to Israel’s Rafael Armament Development Authority. Rafael helped develop Israel’s largest nuclear-tipped missile and also developed the missile’s re-entry vehicle, which is designed to carry a nuclear warhead safely to its destination. In addition, Rafael produces a wide range of advanced conventional military equipment, including missiles. The application was returned without action after objections by the State Department and the Arms Control and Disarmament Agency.

Sun would have received only $52,000 for these two sales, compared to its annual revenue of more than seven billion dollars. Thanks to the NDAA, Sun was saved from having its products help build missiles in South Asia and the Middle East. And all Sun lost was the price of one luxury automobile.

I have brought out this sales information to make a simple point: the NDAA process helps exporters. It saves them from embarrassing sales that are simply bad business. The revenue is not worth the risk. Exporters should be grateful for the help they are getting, and they should be the first ones to ask that the present process be preserved.

According to the most recent information I have, the CEO of Compaq Computer Corporation is Mr. Michael Capellas. SGI’s CEO is Mr. Robert Bishop. Sun’s CEO is Mr. Scott McNealy. I recommend that this Committee invite these three men–and the CEO’s of other computer companies whose sales were blocked under the NDAA–to appear in a hearing, and to state whether they believe that their companies would be better off without the NDAA process. Do they believe that the sales of their products that were blocked should have gone through? Do they believe that similar sales should happen in the future? Do they believe that American citizens will be more secure or less secure if American high-performance computers go to foreign companies that help make nuclear weapons and the missiles to deliver them?

If these men agree that their sales were rightly blocked, they should be willing to stand up and admit that the present system works and should be retained as it is. They should also be willing to endorse a request by this Committee that the Administration rescind its decision to weaken the NDAA by raising the threshold for control next year.

According to the information I have, every one of the computers cited in the examples above operated at less than 6,500 MTOPS. That means that if the control level in 1998 had been at 6,500 MTOPS–where the White House now wants to set it–all of these American computers would now be aiding foreign nuclear and missile programs. That is also true of virtually all of the other computers that were blocked under the existing rules. A total of 85 were blocked.

Thus, if the control level is raised to 6,500 MTOPS early next year, as now planned, we must expect that at least 85 high-performance American computers will be exported to foreign nuclear and missile programs during the year 2000.

How important will these computers be? Could countries such as China, India or Pakistan make use of computers with processing speeds up to 6,500 MTOPS for nuclear weapon development?

The U.S. Department of Energy has found that they could. In a study released last year, DOE found that for these countries to improve their nuclear weapon designs, they will need high-performance computers able to perform about 4,000 MTOPS. Machines operating at that level, of course, are scheduled to be decontrolled next year. DOE also found that access to high-performance computers “would have the greatest potential impact on the Chinese nuclear program.”

Thus, if the decontrol goes through, we have to expect that high-speed American computers will help design nuclear weapons in China, India, Pakistan and possibly other countries.

This does not need to happen. The present system can be kept in place with practically no burden to exporters. In fact, the present system simply creates a paper trail. It asks the exporter to fill out a form and wait ten days to find out whether the buyer is dangerous. That is a small price to pay for keeping one’s products out of mass destruction weapon programs. A company that refuses to pay such a small price is really saying that it prefers to be duped. It doesn’t really want to know what its products are going to be used for.

The GAO report shows how light the burden really is. According to the GAO, there were 938 notifications under the NDAA between February 1998 and March 1999, a period of about thirteen months. That is about 70 per month, or between three and four cases per workday. That is not a heavy burden for export control officers.

Almost ninety percent of the notifications sailed through the process with no objection. And even the 85 cases that were blocked do not amount to a significant number of lost sales. I recommend that the Committee ask GAO to add up the value of the applications that were blocked. I estimate that the total will not exceed $30 million. That amount is insignificant compared to the multi-billion dollar annual revenues of the American computer industry.

The Committee should keep one additional figure in mind. The Tier III countries–the only countries that the NDAA affects–account for only five percent of the market for high-performance computers. Thus, the NDAA process is catching only ten percent of the exports to just five percent of the market, which translates into only one half of one percent of all high-performance computers sold. Such a tiny slice of the economic pie is not going to affect the competitive position of any major computer company.

The industry and the White House claim nevertheless that if the control level is not raised next year, American companies will lose business. That is simply not true. There is no evidence to sustain such a position. If the present control level of 2,000 MTOPS were retained, what would happen?

As chip speeds increase, more computers would exceed the control level and more forms would have to be filled out. The number of notifications would therefore grow. How fast? If the number doubled next year, the Commerce Department would be receiving about seven notifications per day by January 2001. Could the Commerce Department handle that many? There is no evidence that it could not. If the objection rate stayed about the same–at, say, the present level of ten percent–then handling the additional notifications would mean that roughly 170 American high-performance computers would not go to nuclear, chemical, biological, missile or military sites around the world next year.

Would the benefit be worth the cost? The answer is “yes.” It is more economical to pay export control officers to stop dangerous sales than to pay troops to defend against the weapons that the sales produce.

Industry, however, is claiming that chip speeds are increasing so fast that export controls will soon be overwhelmed. But where is the proof that this is true? There simply isn’t any.

We should remember that in January 1996, the Clinton Administration abolished controls on computers operating at less than 7,000 MTOPS to most countries. It did so on the strength of industry predictions (contained in a government-sponsored study in 1995) that computers operating at 7,000 MTOPS would become so common by 1997 that it would no longer be feasible to control them.

That prediction was flat wrong. It turned out that there was no evidence to support it. GAO found explicitly that the 1995 study “lacked empirical evidence or analysis.” Computers operating at 7,000 MTOPS are not commonly available even today. Another government-sponsored study by the same author found last year that machines operating at much lower speeds could still be controlled until the year 2000. And a study last year by the General Accounting Office came to the same conclusion. The industry predictions in 1995 were a mistake and so was the decontrol in 1996.

The result was a series of scandals, as this Committee knows. IBM and Silicon Graphics outfitted Arzamas-16 and Chelyabinsk-70, Russia’s key nuclear weapons laboratories, and Silicon Graphics supplied the Chinese Academy of Sciences, a leading Chinese nuclear and missile research institution. In addition, IBM and Digital sold high-performance computers to the Indian Institute of Science in Bangalore, one of India’s main missile research sites.

If the industry’s last set of predictions about chip speeds were so wrong in 1995, why should they be right today?

The real issue is foreign availability. Are high-performance computers readily available from foreign sources? All the most recent studies say “no.” Indeed, if our industry could prove foreign availability, it would do so. It cannot, so it has thrown up a smoke screen to hide that fact.

The smoke screen is domestic availability. The industry is now estimating, according to a White House press release in July, that computers performing 2,000 MTOPS “will be available in the tens of thousands during the coming months.” But available where? From foreign countries or only from the United States? The answer is: the United States. If they are available only here, of course, they can still be controlled for export. Industry is glossing over that essential point. And how many months are we talking about? Twelve, twenty-four, or thirty-six? And how many of these computers will be ordered by Tier III buyers? These are the real questions, and neither the industry nor the Administration is providing the answers.