# Asianometry: Cisco's Real-Time Dashboard Was Measuring an Echo

Source: Realpha Blog (blog.getrealpha.com)
Original article and charts: https://blog.getrealpha.com/en/blog/asianometry-2026-11-08-how-cisco-got-bullwhipped/

> Notes on Asianometry's 'How Cisco Got Bullwhipped', published 8 November 2026, on how Cisco in 2000 mistook an amplified order signal for demand: lead times stretched to four to six months in early 2000, customers double- and triple-ordered, salespeople added their own buffer, Cisco committed to suppliers on a 60% growth forecast, inventories rose 87% year on year and another 58.8% the following quarter, and in April 2001 the company took a US$2.2 billion inventory charge and cut 8,500 jobs. This piece makes the episode's case at its strongest, then argues the other side, then says which one I think holds and on what condition. Educational listening notes only — no investment advice, no price targets, and the judgements here may be wrong.

Published: 2026-11-09
Locale: en
Tags: bullwhip effect, supply chain, demand forecasting, telecom bubble, podcast-notes

TL;DR: Place one batch of demand with three resellers and the order book shows three times the demand. The more real-time Cisco's dashboard was, the faster it pushed that three times upstream.


![A long warehouse aisle at night, sealed cartons of networking equipment stacked to the ceiling on both sides receding into depth, a loading dock door at the far end letting in a shaft of cold white light, one box slit open in the foreground showing a stack of custom circuit boards](/covers/asianometry-2026-11-08-how-cisco-got-bullwhipped-cover.png)

> Long ago Zengzi lived in Fei. A man of Fei who shared his name and clan killed someone. Someone told Zengzi's mother, "Zeng Shen has killed a man." She said, "My son does not kill," and went on weaving. A while later another man came: "Zeng Shen has killed a man." She kept weaving. Then a third came and said it again — and she took fright, dropped her shuttle, climbed the wall and fled.
>
> —— Zhanguo Ce, "Strategies of Qin II" (compiled by Liu Xiang, Western Han, from Warring States sources; my own translation)

Asianometry's "How Cisco Got Bullwhipped", published 8 November 2026, is about one mistake the networking equipment maker Cisco made in 2000. It had what the business press called the best supply chain in the industry: CEO John Chambers could read a daily report of the previous day's worldwide sales, and the CFO could close a quarter's books in hours. When component lead times stretched to four to six months in early 2000, Cisco committed to long-term orders on a forecast of 60% revenue growth; inventories rose 87% for the fiscal year and a further 58.8% the next quarter; in April 2001 it took a US$2.2 billion inventory charge and cut 8,500 jobs. The host's reading is that the order book was measuring a signal amplified by every layer beneath it. That reading depends on two conditions — lead times long enough that customers hedge, and several layers of middlemen who can each add to the number — and the next two sections take both sides of it.

## The case at its strongest

Cisco's system was taught in business schools at the time. From 1992 onward it outsourced production to Jabil and Solectron, two contract manufacturers that assemble electronics for other companies; by the late 1990s 70 to 80% of its products were built outside, saving an estimated 30% on cost. It built an information channel it called an extranet, letting those manufacturers read Cisco's order book directly, build to it, and ship finished product straight to the customer. Carl Redfield, Cisco's SVP of manufacturing, put it this way: "We've cleaned up the supply chain. You don't add value by having multiple people touch a product." Management ran daily "tactical operating movements" off it, and the press compared the arrangement to a Japanese keiretsu.

Then the shortage arrived. In early 2000 Cisco saw lead times on some networking equipment stretch to four to six months; shortages were estimated to have cost it 10% of the prior year's revenue. Customers who could not get the quantities they wanted protected themselves: they put in double and triple orders, and some placed the same requirement with several resellers and suppliers at once. Cisco's own salespeople added a buffer on top, hoping to win allocation for their accounts. The order book looked magnificent, Cisco took it at face value, passed it faithfully upstream, and locked in long-term commitments with manufacturers and component suppliers on a 60% growth forecast for the fiscal year beginning July 2000.

![There are four increasingly tall bars, each built on the same dark base block representing true demand, with gray blocks stacked above as duplicate echoes, growing from one layer to four.](/figures/bullwhip-one-demand-four-orders-en.svg)

Solectron was the first on the chain to raise a hand. Its own internal forecasts, reading a weakening economy, said caution: it warned Cisco that it was buying more communications chips, optical lasers and assembly boards than appeared necessary. Cisco answered that the business would keep growing, that the failing dotcoms were under 5% of its revenue, and that the components would be needed. The Wall Street Journal later quoted Chambers: "To not plan to meet that growth is the quickest way to lose customers."

The record shows up in the filings. The fiscal 2000 10-K reported inventories up 87% from the prior year, attributed to "increased purchases to secure the supply of certain components"; the 10-Q for the quarter ending October 2000 showed another 58.8% increase. Management had one more cushion in hand: a US$3.8 billion sales backlog at the end of September, worth eight to ten weeks of sales. As cancellations came in, that cushion went quickly.

On 4 December, Chambers told analysts, "I have never been more optimistic about the future of our industry as a whole or of Cisco." A week or so later he ran the daily sales report and found the day short of projection; when the trend held for a week, he started calling customers himself. In April 2001 Cisco faced it: revenue would fall 30% from the prior quarter, 8,500 jobs went, and the company took a charge against the inventory sitting at Jabil and Solectron, including material it had committed to without a contract. Some components were sold. The rest went into a warehouse to wait for the market to come back, and when it did not, that material had to be destroyed, because it had been custom made for Cisco.

![Three horizontal bars show the revenue growth forecast as the shortest, inventory growth extending far beyond it, with a dashed box marking the excess distance between them.](/figures/inventory-outruns-revenue-en.svg)

## The other side: the bullwhip may be too cheap an answer

The first objection comes from the signals themselves. For the quarter ending October 2000, Cisco reported record revenue growth of 65%, and its executives said they had seen no sign of a slowdown. Morgan Stanley's Christopher Stix ran his own channel checks and still found three-month waits for Cisco gear. Tens of millions of people were joining the internet every month, and corporate networking and data switching both looked to have runway. Pointing the other way were a stock down 32% from its March peak and a few scattered warnings — Xilinx, which makes the programmable chips inside routers, cut its internal growth forecast from 16–20% to 10% in September. Two or three negatives are a small thing at a company that has grown for dozens of consecutive quarters. Twice before, including through the Asian financial crisis, management had hit the gas and been right. From where they stood, pressing on was a calculable bet.

The second objection cuts deeper: the disease was in the capital structure upstream, and Cisco's purchasing decisions could not have cured it either way. US telecom capex was US$47 billion in 1995; at the 2000 peak the San Francisco Fed estimated roughly US$120 billion. Most of that money went into the easy part — long-haul fibre loops through rural country — while the last mile into homes went largely unbuilt, and every dollar of long-haul cable needs twenty dollars of last mile behind it. The financing was fragile too. Vendor financing was standard: Lucent had committed US$8.1 billion, Nortel about US$3 billion, and Cisco US$2.4 billion with US$625 million drawn, which made it one of the more restrained players. The extreme case was Winstar Communications, which took US$500 million from Lucent in October 1998, funding over 40% of the price of the Lucent equipment it bought, when the norm topped out at 10 to 15%. When that structure cracked it cracked fast: telecoms raised an average of US$5.6 billion a month from stock and bond markets over the first eight months of 2000, then US$1.2 billion in September, US$900 million in October, US$700 million in November. ICG Communications in Colorado went bankrupt owing Cisco US$32 million for equipment. On this reading, the bullwhip explains why the charge was US$2.2 billion rather than US$500 million, and the bubble explains why there was a charge at all.

![A zero-to-50% scale shows the normal range as a narrow gray band from 10% to 15%, while the Winstar mark sits far away near 40%.](/figures/vendor-financing-over-the-line-en.svg)

The third objection stings most: the conservative side did not win either. Solectron, which had warned Cisco, announced revenue growth of 85% in March 2001 and cut 8,200 jobs the same day. JDS Uniphase, which supplied the optical lasers, cut 16,000 jobs that year, and its stock went from above US$130 in 2000 to US$7 a year later. People did say it out loud in advance: Andrew Odlyzko of AT&T Labs published a paper in late 1998 dismantling the claim that internet traffic doubled every 100 days, by then fundraising gospel; the McKinsey Quarterly in 1999 calculated that if all the fibre planned by US telcos were fully used, backbone capacity would rise 200-fold in three to five years, and noted the oversupply — then added that there was no technological reason the capacity would not fill up. The people reading it correctly left themselves a door too. The host's line about optical valuations in those years is the one I will remember: as hilarious as a donkey eating your figs. Cisco paid US$7 billion for Cerent, a fibre optic startup with US$15 million of revenue. What that money bought was the position that put Lucent and Nortel on their back foot in packet networking, so the same bubble was both the disaster and the weapon.

![Four bars show a tall first bar for the first eight months’ monthly average of 56, followed by three bars that nearly touch the floor, with a dashed line plunging from the high point to mark the cliff.](/figures/telecom-funding-cliff-en.svg)

## Which side holds, and on what condition

Put the two sections side by side and the question I care about is the one anyone spending their own money runs into: everyone is scrambling, my supplier tells me to take a number, revenue is at a record — how much of the demand in front of me is an echo?

The dividing condition this episode offers is when an order book stops measuring end demand. Two things make it fail together: lead times long enough that customers must hedge (four to six months will do it), and several layers who can each add to the number — resellers, suppliers, your own sales force. With both present, the order book is the same demand counted repeatedly, like the three men reporting that Zeng Shen had killed someone: if all three heard it from one source, three reports carry as much information as one. Both conditions held at Cisco in early 2000, so on this one I side with the bullwhip reading — and I keep the bubble reading too, because it answers a different question, namely why downstream demand vanished rather than merely slowed.

![A four-quadrant chart plots lead time on the horizontal axis against the number of middle layers that can each add their own markup on the vertical axis, and only the top-right cell is flagged as one where the order book no longer measures end demand, while the other three cells still measure it.](/figures/when-the-order-book-stops-measuring-en.svg)

Then comes the part I found least intuitive: when those conditions hold, a more real-time dashboard is more dangerous. Cisco's extranet pushed the amplified orders to its manufacturers immediately, and the virtual close let management read the same numbers every day. A report that tells you the same wrong thing daily starts to feel like confirmation. John Morgridge, the former CEO who had moved up to chairman, recalled in an oral history for the Computer History Museum that he watched internal trend data deteriorating in the autumn of 2000; Chambers was optimistic by nature and chose the move that had worked twice before. Starting from the wrong measurement, looking more often gets you wrong faster.

Two habits I now carry back into my own reading. First, keep "supply is tight" separate from "has pricing power": a longer lead time reads as strong end demand and as constrained upstream capacity equally well, and telling them apart means watching whether unit prices and long-term contracts move with it, which the lead time alone will not show. Second, watch the gap between inventory growth and revenue growth — 87% against 50% in Cisco's case. That gap is a forecast bought with cash, and the filings show you who is carrying it. I have fallen into this one myself: I treated a supplier's shortage notice as evidence of end demand and added to a position, and only later found the same end customers queuing at three suppliers.

![Two lines start from the same point and diverge: the upper line bends upward with dense daily-check dots, while the lower line bends downward, and the gap between them widens to the right.](/figures/the-faster-the-dashboard-the-faster-the-error-en.svg)

## Worth a look

- Asianometry, "How Cisco Got Bullwhipped", 8 November 2026 — the source for this piece
- Cisco's fiscal 2000 10-K and the 10-Q for the quarter ending October 2000 (inventory movements and the stated reasons)
- Larry Carter's Harvard Business Review article on Cisco's virtual close
- Andrew Odlyzko's 1998 paper on internet traffic growth rates
- The 1999 McKinsey Quarterly estimate of US backbone capacity and fibre supply
- San Francisco Fed data on late-1990s telecom capital spending
- The Computer History Museum oral history with John Morgridge

## The one thing to take with you

The more hands a "everyone wants this" has passed through, the bigger the discount it deserves. This has nothing to do with whether the information is accurate — every order Cisco received was placed by a real customer, every forecast Solectron received was sent by Cisco, nobody on the chain lied, and the number still came out three times too big.

Here is something I have tried that works outside investing too. Next time I hear "everyone is grabbing this", "you can't get a table there", "everybody is applying to that school", I ask one question back: how many people did you hear that from, and do they know each other? If all three heard it from the same person, what I am holding is one person's account — and I was about to make a decision sized for three.

![One dot on the left splits into three arrows leading to three report boxes, whose arrows then converge into a box on the right labeled one information unit, forming one into three and back into one.](/figures/three-reports-one-source-en.svg)