Jeffrey LewisChina's Nuclear Stockpile, Revisited

How big is China’s nuclear stockpile?

Benn Tannenbaum sends along this quote from Ed Corcoran who provides, er, a rather large range of estimates:

China is neither ally nor adversary, but something in between. It is a nuclear power with a relatively modest arsenal of perhaps a couple hundred nuclear warheads, though some estimates do run over two thousand, together with some fifty missiles capable of reaching the United States.

“Two thousand?” Benn asked.

The 2,000 reference is based on an essay by a college student in Singapore that I thought I had thoroughly debunked more than four years ago.

Since it doesn’t seem to go away — despite a pretty rough treatment on the blog and in the pages of the Bulletin of the Atomic Scientists (when the Bulletin had pages) — I thought I would revise that essay.

The short answer is that 2,000 is ridiculous.

How Many Chinese Nuclear Weapons Redux?
Originally published, December 2004 • Revised, May 2009

In 1996, a Singapore University student, Yang Zheng, posted a short essay online, entitled China’s Nuclear Arsenal, that concluded that “it is very likely that China is making 140-150 nuclear warheads a year and she has accumulated 2,350 nuclear warheads so far.”

By contrast, unclassified U.S. intelligence assessments at the time suggested that China “has over 100 warheads deployed operationally on ballistic missiles. Additional warheads are in storage.” Declassified documents from the 1990s place classified estimates of the total stockpile, including a small stockpile of aircraft -delivered gravity bombs, between 200 and 250 warheads.

The fact that Yang’s numbers were completely outside the bound of the possibility didn’t stop certain conservative “defense intellectuals” from citing the hell out of it:

• Richard Fisher, writing for The Heritage Foundation, cited it not once but twice — (the latter co-authored with Baker Spring). Fisher even claimed that “one U.S. government expert told this author that its estimates are plausible.”

• David Markov and Andrew Hull use Yang’s fissile material prodution and warhead estimates in an Institute for Defense Analyses study entitled The Changing Nature of Chinese Nuclear Strategy.

• David Tanks, then with the Institute for Foreign Policy Analysis, called the essay “convincingly argued”.

• The Center for Defense and International Security Studies also cites the essay to argue that an estimate by the Natural Resource Defense Council of approximately 400 warheads was “too low and thus inaccurate.”

Every single one of these people should be red-faced with shame. (And, I regret to add, this is not an exhaustive list of citations.)

The core of Yang Zheng’s argument is that estimates placing the Chinese arsenal at 2,000 plus warheads are “are reasonable” because “data from various U.S. intelligence agencies show that, in the mid-1980’s, China was producing at least 800 kilograms of U-235 and 400 kilograms of Pu-239 per year.”

That’s great, except the declassified data doesn’t show fissile material production at that level.

Just take a look at plutonium, production. A classified DOE estimate of Chinese plutonium production, leaked to the press, places Chinese Pu stockpile at 1.7-2.8 metric tons. This is consistent with unclassified estimates by Gronlund and Wright (2-5 metric tons) and Albright et al (4.8 metric tons). For more information, see my post Guangyuan Plutonium Production Reactor, November 9, 2006.

Assuming 3 to 5 kilograms of plutonium per warhead, 1.7-2.8 tons of plutonium could support a force of 340 to 930 weapons. If China uses substantially more than 5 kilograms per warhead, its stockpile might only support a few hundred weapons.

In fact, Yang’s dicussion of Chinese Pu and U-235 production facilities is wonderfully incompetent. Here is Yang’s table of Nuclear Explosive Material (NEM) facilities:

Facility Pu or U235 kg/year
Lanzhou Gaseous Diffusion Plant U235 400
Helanshan Centrifuge I U235 400
Helanshan Centrifuge II U235 ???
Yumen Breeder Reactor Pu 250
Baotou Breeder Reactor Pu 150
Guangyuan Breeder Reactor Pu ???

Reproduced from Yang Zheng, China’s Nuclear Arsenal, March 16, 1996. Reproduction does not imply endorsement.

These sites are simply wrong:

• Baotou isn’t a plutonium reactor but rather a Nuclear Fuel Element Plant. The Baotou Plant resembled a French-designed plutonium reactor from the air, leading US intelligence agencies to mis-identify it during the early years of the Chinese nuclear program (William Burr and Jeffrey Richelson discuss this error in “Whether To ‘Strangle the Baby in the Cradle’: The United States and the Chinese Nuclear Program, 1960-64.”)

• The list is missing a second gaseous diffusion plant at Heping.

• China didn’t have centrifuges at Lanzhou when Zhang wrote his post. (China did subsequently install leased-Russian centrifuge modules at Han Zhong and Lanzhou to enrich LEU to fuel reactors, but the first did not become operational until after 1996.)

An accurate list here.

Moreover, Yang Zheng doesn’t consider the operating histories of the facilities, which stopped producing enriched uranium in 1987 and plutonium in 1991. Serious problems with the facilities could have substantially reduced output—substantial operating problems are detailed in the official Chinese history of the program and probably account for the DOE classified estimate being lower than either unclassified estimate.

Examing the operating history of the reactors, of course, would require actual analysis. What Yang Zheng did, on the other hand, was to simply clean up and repost the uninformed rantings of an anonymous commenter on a message board in cyberspace.

That doesn’t suprise me — plenty of the online debate is uninformed and biased.

What does surprise me, however, is the number of so-called “experts” at allegedly respectable institutions — Heritage, IDA, IFPA, CDISS and others — who further cleaned up Yang Zheng’s rantings and presented them as fact.

These people should find a new line of work.


  1. Major Lemon (History)

    Ambiguity is a useful factor in nuclear deterrence because a potential enemy never knows exactly what to expect. The same with arms control negotiations. The ‘experts’ that Jeff rightly calls into question are just pawns in this game of Chinese Whispers.

  2. Emily (History)

    Re: Baotou, the FAS/ (Global Security and FAS) entries for this facility seem to contribute a lot to the confusion. John Pike has updated the facility’s entry at the newer site, but it only says ‘In fact, there appears to be to no reactor at this facility’ at the end of the entry. It is tough to update such open-source information, but removing those erroneous entries from highly-referenced websites could be a start in reducing misinformation.

  3. FSB

    Did you just mention “intellectual” and Richard Fisher in the same sentence?!


  4. steve (History)

    Thankfully, the practice of appropriating postgrad-level work on WMD capability by those with a specific agenda is isolated and…oh, that’s right.

  5. Smith (History)

    Ah, the dunce cap.

  6. V.S. (History)

    “…line of…” Redundant.

  7. AWR (History)

    Just one question for the experts, who have done great work in their estimates, even if they differ numerically: what’s the real difference between 200 and 2000? It’s not a math problem.

  8. Bill (History)

    You wrote this piece in 2004 and most of the authors you cite wrote theirs in 1996/7, so of course a snarky piece written in 2004, with the advantage of hindsight makes you look insightful. The collective body of knowledge on China’s nuclear infrastructure has grown considerably with each passing year, while in 2004, it is easy 8 years later to shake your snarky finger of indignation and denounce them as fools and naives. If you were around doing research in the late 90s, and talking or reading assessments of others looking at the issue at the time in Japan, South Korea and yes even Singapore, there were scholars and government representatives who voiced a growing concern about the potential for China to, if it decided, to quickly expand its nuclear weapons inventory.
    China was alarmed by the results of Desert Storm in 1991 and was discussion in military journels in the mid to late 90s at expanding their tactical nuclear weapon inventory as well as using enhanced radiation warheads much the same way as conventional weapons would be used. In fact recent writings by members of the Chinese military see little difference between tactical nuclear weapons being used to achieve conventional effects. These same authors even cite examples of using these devices to neutralize ports, airfields and bunkers in a Taiwan scenario.
    So the question is when did China, if at all, begin their road to building tactical nuclear weapons? In 1996, the idea that an open source was discussing the possible expansion of China’s nuclear inventory (tactical and strategic) was an intriguing one when coupled with all the rest of China’s nuclear weapon and delivery modernization efforts. Was it all the result of an effort by a nation to simply build a more assured nuclear delivery platforms/weapons or the result of a military-industrial cultural that was seeking to have resources spent on it so it could remain financially viable?
    Looking at all the delivery programs that were rumored at the time, it doesn’t seem that far fetch in 1996/7 to have wondered out loud, if China was looking to grow their inventory to match the growth in their R&D in new delivery platforms and nuclear warheads. The point being, were the numbers portending a growth in new delivery platforms that were being research in 1996/7 (i.e., since cancelled DF-41 and work on DF-31 / DF-31A) with corresponding warheads (was China going to MIRV)?
    Given all this, it appeared that the number of warheads was going to grow at a rate potentially greater than many of the estimates at the time (1996/7) because of the expansion of China’s strategic delivery platforms and their efforts to expand their tactical nuclear weapon applications and inventories. Were analysts forecasts at the time (1996/7) taking too conservative of a look at the inventory of 2006/7?
    Was the number higher than 400 or was there the possibility that China, as Alistair Johnston’s wrote in “China’s New “Old Thinking”: The Concept of Limited Deterrence” International Security, Vol. 20, No. 3, Winter 1995/96, considering a change from a countervalue to counterforce targeting strategy? Was China beginning to signal a paradigm shift in their nuclear strategy and were Chinese notions of limited deterrence beginning to rest on the notion of a limited warfighting capability?
    So while the various persons you cited used what was a poor reference, it appeared that China’s inventory had all the signs it was going to grow and was not a bad point of departure to begin to explore what the implications of such growth would be for the future. Further, any topic such as nuclear weapons and China is a difficult one, to have a truly fact based discussion given China’s efforts to hide even the most trivial of details on its nuclear weapons infrastructure is fraught with debate on any number a Western analyst wishes to put forward whether in or out of the government. I agree the number is not 2,000 but it is certainly more than most analysts speaking on the topic state with any certainty today. Unless of course, Jeff has a secret Beijing source, who provides him unique insight, that is not worthy of civil individuals trying to make sense of a topic which is murky at best, even with what is attributed to be the “right” information. And by “right” one means things that come from the keyboard of Jeff.

  9. Joseph Logan (History)

    I thought this had been put to bed long ago, as it appears you did. Maybe it’s just my ignorance of the field, but you seem to be applying heavy logic to what you yourself have made into a simple problem. Doesn’t seem like Corcoran is solidly behind the 2k number, but he doesn’t seem bashful about citing it, either.

  10. Ray (History)

    Let me do a little “piling on.” Snarky Jeff’s constructed his 2004 assessment of the Chinese nuclear infrastructure and targeting doctrine virtually without benefit of any primary source material. The original 2004 article in the Bulletin of Atomic Scientists contains only one reference to a Chinese source – and that source made his comments 16 years before the article was written! By contrast, many of the authors you cited relied heavily on open source Chinese military doctrinal writings. Instead of using Chinese military writings, Snarky Jeff quotes Bill Gertz – that imminent scholar. Snarky Jeff also asserts (as I remember the article) that the Intelligence Community is never wrong in its assessments of foreign nuclear weapons material issues. Lastly, the source he so despises and seeks to discredit did yeoman work in the mid-1990s trying to understand Chinese ability to produce nuclear materials, based almost solely on exploiting original open Chinese sources. He may have been wrong, but his research approach was much more rigorous than any used by Snarky Jeff. Finally, Snarky Jeff chooses to disregard all of the other CORRECT predictions that many of the authors of those works made in their 1996-7 assessments.

  11. Jeffrey Lewis (History)

    Bill and Ray:

    Gosh, just a tad defensive, aren’t you?

    Bill, it is very simple — accurate information about Chinese NEM facilities was available in the mid-1990s.

    Had it not been, that still wouldn’t explain simply copying any old thing you found on the internet.

    The misidentification of Baotou, for instance, was really quite sloppy. Look, it is one thing to disagree with the IC estimates for a reason. It is another to not know what they are in the first place.

    As for Ray, I don’t say the IC is “never” wrong — see Baotou, above.

    But what else would you start with? Random internet postings?

    The only sensible thing to do is to start with IC estimates for most quantitative data, then compare that with qualitative statements by the Chinese. (There are also some non-IC sources of quantitative data, including seismic data and Swedish radio-chemistry data.)

    As for Chinese source material, when Beijing starts publishing missile numbers and giving tours of NEM facilities, I am there. But I will still check it against IC estimates. Folks in the IC are human, much-maligned, subject to political pressure and just generally underpaid. Their subjects often deliberately conceal information from them. Despite all this, they produce serious work. I really admire them, even if I don’t always agree with their conclusions.

    As for my work, there is some Chinese source material in the book, both documents and interviews. (Did either of you even read the book?) I also made the effort to schlep out to Haiyan, for instance, took pictures in the museum and brought back video of China’s early nuclear tests. I used that video in making mass estimates of the warhead design that ended up in Pakistan and Libya. That’s pretty damned rigorous, if I do say so myself.

    As a closing thought, when I pointed out that some other researchers have committed what amounts to academic incompetence — and that is what uncritically citing internet postings amounts to — I don’t expect a thank you note.

    I knew people would be embarrassed and angry, and that I would paint a giant target on myself.

    I take some pride, Bill and Ray, that you concede that my estimates are accurate. Your argument is not that I am wrong, but that I am right with the benefit of hindsight.

    I am not a perfect scholar — we all make mistakes. But there is a difference between honest differences of opinion or errors in search of the truth, on one side, and incompetence or dishonesty on the other.

    I know on which side of that line those who uncritically cite “Yang Zheng” are standing.

  12. AWR (History)

    Wow. When I had access to intel many years ago, I asked how many nuclear weapons a certain Near East country had. The answer was not in any way definitive, and I was given a range that was pretty wide. It made me very curious, but no one could or would tell me what it meant. I asked the same question years later about China, and was also given a wide range – approximate to 200 to 400. Bottom line is nobody knows, with all info available, how many nukes the Chinese have, any more than Jeffrey or anyone else what Hu Jintao had for dinner last night. I think the probability of dumplings is quite high, but I’ve never dined with him.

    Estimates are just what they are, nothing more, and they help us to understand, in the absence of information that is as secret as what was in the dumplings, what a country’s intentions and preparations may be.

    I am much more appreciative of deductions made about minimum deterrence – if that’s what Chinese policy really is – than I am about whether it’s 200 or 2000. But those distinctions do make a difference, and any non-perfect scholar really deserves more respect than what has been bruited about on this blog. As was said long ago, the difference between zero and one (or five) is far more important than one and one hundred. Once you establish a number greater than zero, you can only go to an estimate like all of you have produced to establish some meaning that the government in question will never provide you. But keep up the good work!

  13. bobbymike (History)

    200 or 2000 it really doesn’t matter as the big picture shows the Chinese are doing what they please and developing nukes for their defense REGARDLESS of what the US does – who have not built or tested a nuke since 1988-1991 respectively.

    I have never bought into the “what the US does the world will follow” line. Certain countries want nukes for their defense or national prestige or whatever. In fact evidence is the exact opposite as every country with nukes is modernizing EXCEPT the US, how does that square with the “US should show leadership” argument?

    The US should continue to modernize – warheads and delivery systems – and most importantly have a robust R&D program and “breakout” capability to avoid strategic surprise.

  14. Yale Simkin (History)

    While I don’t see any reason for China to have built more than a few hundred warheads, what could they build?

    Zheng’s numbers (+2000) appear to mostly be derived from the fissile stockpile.

    What do the data show?

    Jeffery’s essay stated an “An accurate list” of Chinese fissile production facilities was at the NTI site. That page links to another NTI page of various expert estimates of actual production.

    The range of HEU was 4 – 20 tons.

    10 kilograms of HEU per boosted warhead yields a range of possible warheads of 400 to 2000.

    The DOE estimate of 1.7 – 2.8 tons of plutonium, at 3 kg per boosted device yields 567 – 933.

    This results in an outside number of +2900 potential warheads.

    Again, I am not promoting the view that China has that many devices or that they are efficient in per weapon fissile requirements, but Zheng may not have been way off in what they can do.

    I am interested in Chinese tritium production. That would say a lot about their actual and potential stockpile.

  15. Jeffrey Lewis (History)

    Hi Yale. Since you actually do the research, let me give you a somewhat more detailed response with some links to explain my assumption that the size of the stockpile is based on Pu production divided by slightly more than 3 kg per warhead.

    The number of Chinese designs is rather small and, until, 1980, detailed information about the designs were available through data from atmospheric testing. (That means probably only the 470 kg DF-31 Type RV is not known through debris fallout, depending on what arms the DF-21.) Two decent primers are this one and
    this one.

    The pits for the deployed warheads (viz the HEU warheads tested in the 1960s) are mostly Pu and use a rather lot of it.

    So, when I do stockpile estimates, I size the force to the number of Pu pits the Chinese could make.

    I think I figured 3-5 kg of Pu spread across a 1.7-2.8 t stockpile for a total of 340-930 warheads.

    If the Pu usage is more intense, on the order of 7 kg per primary (see De Geer for evidence that the Chinese might have more Pu per bomb), then, well, that number drops pretty low — to about 240-400.

    I am still unsure of DeGeer’s estimate — he’s quite competent and there is no reason to doubt him, but that is a lot of Pu per bomb. Tom Bielefeld and I are trying to work on a paper considering this precise question.

  16. Yale Simkin (History)

    De Geer’s numbers are based on pulling a quite small number from a great deal of noise. He also makes some big assumptions as to the elemental and isotopic composition of the primary, secondary blanket (and possible fissile sparkplug in the secondary – which he appears not to address), nor the possiblity of a failed tertiary stage with Pu sparkplug, nor fusion boosting.

    That being said, he does make his point, which is not demonstrating the quantity of fissile in a Chinese nuke, rather it was showing that the H-Bomb explosion was a Teller-Ulam design rather than the Teller classical (or Sakharov layercake).

    But what about the actual numbers he did derive. What are their utility in stockpile analysis?

    De Geer calculated that the Chinese test left 3.09 +/- 1.01 g of primary Pu unburned per kiloton of fission. He then calculated the fission yield of the explosion was 2,500 +/- 200 kilotons.

    From that you get a range of UNBURNED (not total) primary Pu of 4.784 Kgs to 11.07 kgs.
    De Geer tightened the range to a likely probability of 5.1 kgs to 10.3 kgs.

    This is not the amount of Pu in the primary. A reasonable yield for the primary is, lets say, 25kt. At 17.3kt/kg of fissioned Pu, this adds about 1.4 kg of Pu. (The primary may have been more powerful, but I leave it at 25kt)

    So, our range is now 6.2 to 12.5 kgs. (Fat Man was 6.3 kg)

    There is also the very significant quantity of Pu burned by fusion neutrons even with a primary separate and shielded from the secondary.

    The bottom line is that De Geer’s numbers are very high. His lowest numbers are possible – but very inefficient. His highest numbers are very weird. They are higher than the bare-sphere critical mass quantity for Pu239!

    It would mean a extremely low-density core geometry and real quick and powerful assembly.

    I dunno…

    Does any of this matter?

    The quantity of Pu used in a primary for a test of 4000 kilotons 33 years ago is not necessarily identical to the current stockpile designs.

    Again, none of this is relevant to the point I was trying to make in my post.
    I was not claiming that China is using both HEU and Pu, nor that they are using them in efficient designs.

    My post was about what is their POTENTIAL capability? – the necessary starting point for considering China’s future stockpile.

    To quote myself: While I don’t see any reason for China to have built more than a few hundred warheads, what COULD they build?

    And: Again, I am not promoting the view that China has that many devices or that they are efficient in per weapon fissile requirements, but Zheng may not have been way off in what they CAN do.

    While we may consider their intentions and needs, good policy requires that we must not lose sight of their capabilities. (See ACW discussions like here )

    If China used both its HEU and Pu, and used it in high-tech efficient designs, then without any further fissile production, it is sitting on more than 2000 bombs worth of SNM – a fact of at least passing importance.

  17. Jeffrey Lewis (History)


    I agree that De Geer’s numbers are somewhat confusing. That’s why Tom and I are trying to take a closer look at them.

    Regarding the Chinese stockpile, the DF-5 is armed with a warhead atmospherically tested in 1980.

    The only “new” warhead designs to enter the stockpile are those that arm the DF-21 and, now, the DF-31/DF-31A.

    That is about half the stockpile.

    I do wonder how much more efficient the “miniaturized” DF-21 and DF-31 warheads will be.

    At issue is whether the Chinese have made a design choice or no.

    I agree that the fact that the Chinese might have made relatively inefficient use of the NEM stockpile is of at least passing importance.

    Send me an email. There are some things I’d like to add, offline.