I worked at Cisco for 25 years — Norway, the UK, Tokyo, and a lot of time with San Jose — and was selected for exit in a restructuring last year. That is my interest, declared up front. I write in a personal capacity, and my current employer had no hand in any of it. Where this is my own account I flag it. Everything else is from Cisco’s filings, results and press, sources at the bottom.
Three weeks ago, on the fourth quarter call, the CFO said fiscal 2026 had the highest productivity metrics in thirty years: revenue, non-GAAP operating margin, earnings per employee. Record revenue of $63.3bn, up twelve percent. Net income up thirty percent. An all-time high share price in June.
In May it had announced another reduction: fewer than 4,000 jobs, under 5% of employees, while raising its outlook. Three months later it named earnings per employee as a thirty-year high.
I went out in a round of about seven percent of the workforce. The year after, the company reported its best earnings per employee in thirty years. It is their number and I am in the denominator.
Cisco has been running culls since 2001. From inside, the reductions felt recurrent: some years more than one, occasionally a year off. The public rounds: about 8,500 in 2001, 6,500 in 2011, 6,000 in 2014, 4,100 in 2022, two plans in calendar 2024 at about 5% and 7%, and fewer than 4,000 jobs — under 5% — in May 2026. None of it had to be dug up. It was all press releases.
They kept hiring through it, often in the same breath as the cull. The stock market often liked it. You would think that many people going out the door was a sign of decline. Cisco can do financially very well while removing people in large numbers. The cuts have arrived in good years as well as bad ones. That is not proof that none of them were a response to a business problem. There were good colleagues and good years. This is about the other machine.
Headcount still went up. People in, people out, often in the same year. Distress looks like 2001. This is the other kind: cash, cover, and a ratio that still moves if you shrink the denominator.
A list that can arrive any year changes how the people who still have a badge behave. You cannot put a date on it. You do not need a memo. I cannot prove they run it for that reason. I can tell you that is what it does.
The test was never run Link to heading
Norwegian law requires objective grounds for a redundancy dismissal. It requires a consultation, defensible selection, and a route to the courts. Since January 2024, § 15-7 also requires suitable other work, including in other Norwegian companies in the group. Preferential rights attach to the same word. I am not alleging that Cisco acted unlawfully in my case.
A signature records consent. It does not tell an outsider what the alternative looked like when you signed.
Where an individual exit is resolved by settlement, the § 15-7 questions — grounds, selection, suitable other work, proportionality — may never be adjudicated. The employer does not need a judgment that a dismissal would have satisfied § 15-7. Collective consultation is a different machine. I am not saying a settlement makes that vanish.
I am not saying my exit failed that test. I am saying the test was never run.
Both sides can price the risk of an unfair-dismissal claim. Invalidity is a possible remedy — continuation of employment — and damages are whatever the court finds reasonable. There is no statutory two-year ceiling. A sufficiently attractive settlement can still make testing the dismissal economically irrational for one employee, even where the case might have been arguable. The protection has not disappeared. It has become something the parties can price.
Among the cases I personally saw — Norway, the UK, Tokyo, a lot of San Jose, one person’s view, no company-wide data — the people going out skewed old. Over fifty, mostly. I cannot infer motive, and I am not alleging age discrimination. Long tenure and age raise the economic cost of a dismissal through salary, notice and potential loss. That also raises what an employee may rationally require to leave. A spreadsheet will find that on its own.
The individual protection is real on paper. A settlement means it is often never tested.
The number nobody knows Link to heading
In February 2009 Chambers told analysts Cisco would reduce 1,500 to 2,000 positions, and that this did not count: “the definition of a company-wide layoff to me is at least 10% of your workforce.” Bloomberg ran it under the headline “But Just Don’t Call Them Layoffs.” Delivered in pieces: 250 in February, six or seven hundred in July. On the same call, around 1,000 people had already gone the same way over the previous six quarters, which had never been announced at all.
That is the clearest example I have found of an unannounced round surfacing in public, and it surfaced by accident, in an answer to an analyst. There was nothing unusual about it. It was just visible for once.
So the public record is incomplete. From inside I continued to see departures I could not match to any announcement. People were there, and then they were not, and no number was attached to it anywhere. Repeated reductions small enough not to generate a collective-redundancy record or a separately announced company event are much harder to reconstruct than one large announced round.
Net headcount does not tell you either. It went from 38,000 in 2001 to 86,200 in 2025, a net growth of 48,200, and some of that was bought. Cisco has acquired more than 200 companies. Calendar 2024 is the clearest picture: about 84,900 employees at FY2023 year-end, about 90,400 at FY2024 after Splunk, and two restructuring plans the same calendar year, about 5% in February and about 7% in August. Splunk and the people leaving were different organisations. Nobody swapped one for the other. Acquired growth and large-scale exits were happening at the same time.
So how many people went through the process? Nobody knows. That is the point.
The rounds with a published headcount come to about 25,000. Calendar 2024 adds 5% and 7%. May 2026 adds under 5%. That is a floor of press releases, not an estimate of everyone who left.
A crude rate model shows how quickly the total gets larger. Two percent a year against the people Cisco actually employed over those years gives about 33,000. Four percent gives 66,000. That is not an estimate. It is the size of the uncertainty. The method is in the notes.
When Cisco did publish a rate, it was generally several percent: about 9% in 2011, 8% in 2014, 5% and 7% in 2024, under 5% in 2026. For the named rounds to be the whole story, the average would have to sit well below the figures they put on a round when they put a figure on one. I cannot pin it down any tighter from the public record.
Where the money went Link to heading
Buybacks started in September 2001, about five months after the first big cull. From October 2001 to October 2022 the company spent $152.3bn on them, 95% of net income by Lazonick’s count, plus $55.5bn in dividends. The programme has now passed $172bn and 6.2 billion shares. Cumulatively it dwarfs any single Cisco acquisition.
Revenue sat around fifty billion for more than a decade. Returning that much capital to shareholders is a choice about the surplus rather than retaining it for future investment. Lazonick calls that shift from innovation to financialization. From the inside it just felt like nobody was interested.
They also funded good work. IETF participation for years, including mine. VPP, open sourced: work the company could not keep exclusively proprietary. A place that only extracted would not have done it. The coffee was fine too. None of that is the ratio.
Whatever caused the result, Cisco reports the ratio itself as an achievement. Its stated explanation cuts the other way: it expected to reinvest substantially all of the restructuring savings in priority and growth areas, rather than retain them as ongoing cost savings. Earnings per employee rewards a larger numerator and a smaller denominator. A very strong year on the numerator, while reported year-end headcount had already fallen from 90,400 to 86,200.
The layer Link to heading
Cisco did this three times between 2001 and 2013. A small group of senior engineers left, formed a startup with Cisco as the exclusive investor, and Cisco later bought it back. The same four people led all three. Cisco supplied the capital, carried the risk, and paid a premium for what its own money had built.
Andiamo was storage, a market Cisco was not in. Cisco already owned 44% and paid $750m for the rest. Andiamo’s 317 staff averaged $2.3m each in Cisco stock. An irritated former executive, to Bloomberg: “These folks didn’t have to take any risk.”
Nuova was servers, another place Cisco was not. Completed May 2008, with a maximum success-based payout of $678m rather than a simple purchase price. By February 2009 Chambers was discussing 1,500 to 2,000 reductions, in slices small enough that nobody had to call it a layoff.
Insieme closed in November 2013. The following August, Cisco announced 6,000 redundancies. Andiamo and Nuova had at least been bets on new markets. Insieme built switching, which is what Cisco already was, with around seventy percent share. Maximum $863m for a core product from a startup Cisco already controlled and largely financed. The message received at the time: Cisco’s own engineers were not considered capable of the work.
Across the three, the published maximum payouts added up to nearly $2.3bn. Business Insider later reported principals at up to $40m apiece per spin-in, and more than $2bn directed to the team across their years. Treat the last two as reported. A handful of people got extremely rich. The engineers who integrated the results stayed on payroll. The four were not treated as a cost. The people who made the products run were.
David Graeber has a name for one version of this. In Bullshit Jobs he calls it managerial feudalism: executives downsize the productive part of a company and spend the savings on retinues, worse when the profits come from finance rather than from making things. Lazonick has the finance half for Cisco. From where I sat, it looked uncomfortably like the worked example. That capital return did not mean the work felt lavishly resourced. Products remained unfixed. Teams remained stretched.
Many of the acquisitions I worked around remained separate products with separate roadmaps for years. Flagged as inside, not filed. More than 200 acquisitions create real duplication. From where I sat, a lot of that duplication stayed, and some of the people maintaining it were among those going out of the door.
Anybody with 86,000 employees has layers. This one is titled and expensive. A decade of flat revenue, and I never saw anybody at that level go out for it. A cull shrinks the denominator. I never saw the layer above them on a list.
In 2007 the company built a structure of councils and boards, which pushed decisions down to more than 500 executives. It was written up admiringly at the time. In May 2011 the nine councils were cut to three. Chambers told 73,000 employees that Cisco had lost credibility. Analysts said the structure had slowed decisions and blurred accountability. Same year as the 6,500 round. The committee layer was reorganised. The workforce was reduced. Different verbs.
The last spin-in closed in 2013. Robbins ended the practice. Chambers stopped being CEO in 2015. I cannot prove any of it still holds. 2011 is the last public count I have found of the layer. The workforce is in the 10-K every year. The layer above it has been counted once.
The measurable things did not change: culls, buybacks, the shape of the revenue line. From inside, later years went the other way: more layers, more VPs, more process, more politics, more finance. Five VPs between a working engineer and the chief executive. That count was never the argument. The spin-ins are.
Cisco used to sell this. The company that told everyone the Internet changes everything, that ran a consulting arm on how to be more like Cisco, wanted to be the worked example. It still is, just not in the direction advertised.
Whether it works Link to heading
The largest older study is Cascio, Young and Morris, who took 5,479 changes in employment across the S&P 500 between 1980 and 1994. Firms that cut headcount without restructuring anything else did not beat their own industry on return on assets. Return on assets fell from about fourteen percent to about eleven, and the ones that cut deepest fell furthest. An earlier Fortune 100 study found the same direction. Steel and House, in 2024, pooled the later work and found little evidence of long-run gain, mixed shorter-run effects, and accounting performance inside downsizing firms tending to decline. That is not the same as the question being settled.
Cisco’s own numbers do not rescue the business case. Twelve years at about 1.3 percent revenue growth. Cisco spent more than $11bn on Scientific-Atlanta and NDS, and later sold the set-top-box business for $600m. Purchase price minus later sale, not lifetime economic loss. Intercloud, announced in 2014 as a billion-dollar public cloud, switched off in 2017. Flip Video, $590m, closed. Linksys sold off. On price alone, the share passed its March 2000 close of $80.06 on 10 December 2025. Twenty-five years and nine months. Inflation-adjusted that peak is about $155; the quoted price has never been there. It reached $130.37 in June this year and has come off since. Dividends improve total return. I have not calculated a 2000 buyer with reinvestment. Whatever the culls were buying, it was not growth.
In Cascio’s sample those same firms still modestly beat their industry on share price. The newer meta-analysis finds market effects mixed and often short-lived. I have not done a Cisco-specific event study. What I can say is that Cisco named earnings per employee as a productivity achievement in a record year, after a restructuring.
All else equal, the ratio improves when the denominator shrinks. You can do that inside a year and announce it. Building something takes longer and might not come off. Only one of those can be decided in a quarter. A literature that finds little long-run gain has not, so far, slowed the practice down. The metric that moves in a quarter is the one that gets managed.
The wrong shape Link to heading
Amazon, Microsoft, Google, Meta, Telenor. Once the repertoire exists, you do not need Cisco’s particular history to run it. A RIF in a record year is now an ordinary calendar item.
Doctorow, in Enshittification, says competition, regulation, self-help and labour used to hold a company like this in check, and that all four have been taken apart. That matches the view from inside.
Don’t join is easy to say, and hard when that is where the work is. They often weren’t like this when you joined.
Join a union, then. I wasn’t a member. Norway has high density, sector bargaining, and consultation in the statute. I still don’t think membership would have changed the list. The reason is the settlement, as above. Germany: the works council must be heard before every dismissal, can object on specified grounds, and a dismissal without consultation is void. Establishments with one show lower dismissal rates. Still bolted to a dismissal.
Cisco restructures globally. Employee-specific protections are enforced nationally, often at establishment level. It is not that Norwegian unions are organised one employer at a time. A few hundred people per country per year never have to be added up. Unions that matched the employer would organise across borders, and across employment status — contractors and temps were about a third of the 2001 number and then vanished from the count — with a say in the list, not a meeting after it, and the settlement collectively.
Customers shouldn’t buy from enshittified companies either. Try that in a concentrated market with high switching costs. Try ripping a campus network out of Cisco, or leaving Microsoft. Count the operating systems: IOS, IOS-XE, IOS-XR, NX-OS, Meraki, the firewall one, the SD-WAN one. Management platforms: CiscoWorks to Prime to DNA Center were transitions; DNA Center to Catalyst Center was primarily a rename. A CCIE is enormous technical skill, much of it transferable, and also years in one vendor’s architecture, terminology and tooling. An incentive for engineers and employers to stay in the ecosystem they already know.
A way out that doesn’t require heroism means interoperability. Public procurement is the lever we are worst at using. Standards-based interfaces and a documented exit path, and the market would follow. The EU Data Act does some of this for cloud. Nothing much does it for networking.
The individual protections attach to a dismissal. A company that can exit people by settlement, and that can keep each round small enough, nationally, not to generate a collective record, will never have to add the numbers up.
The common theme is that the employer should not be able to make each exit disappear into an individual transaction. If the surplus cannot be returned through buybacks on this scale, some of it might stay in the company. If the employer carries more of the real cost of a redundancy, instead of handing it to the state and to the family, the spreadsheet is less attractive. If the annual total has to be published unsliced, the pieces have to be added up. If consultation, selection and bargaining attach to the settlement as well as to a dismissal, the individual questions get asked.
The protections are real. They are the wrong shape for the machine they are supposed to constrain.
Notes Link to heading
Announced rounds with a published headcount come to about 25,000. Calendar 2024 is percentages in Cisco’s filings; I have not converted them to headcount. Cisco hires in the same years it fires.
- 2001: about 8,500
- 2011: about 6,500 (BBC counted more across that stretch)
- 2014: about 6,000
- 2022: about 4,100
- Calendar 2024: about 5% in Q3 FY2024, then about 7% announced Q1 FY2025 (August 2024, after FY2024 closed)
- May 2026: fewer than 4,000 jobs, under 5%. Robbins’ blog. 8-K: silicon, optics, security, AI. Reuters: raised the outlook the same day.
Headcount, 10-Ks: 84,900 FY2023, 90,400 FY2024, 86,200 FY2025. Splunk: nearly 8,000 at acquisition, last 10-K.
Rate model, now only in these notes. Year-end headcount FY2001–2025 sums to 1,652,141 person-years, mean 66,086. Multiply that sum by an assumed annual rate. Crude, constant-rate, certainly wrong. From inside: two rounds some years, none in others; 2006 and 2007 each added more than 11,000. Range, not a number. Body uses 2% and 4% as illustration (about 33,000 and 66,000). Full set: 2/3/4/5% = 33,041 / 49,561 / 66,082 / 82,602. 6,500 against 71,825 in 2011; 6,000 against 74,042 in 2014. Calendar 2024 stays as 5% and 7%.
2001: announced 8,500 staff and contract positions. 10-Q: initial estimate about 6,000 regular employees, about 5,400 actually terminated. Rest contractors and temps. I have used positions throughout.
Revenue: Cisco results, tabulated. FY2013 $48.6bn, FY2025 $56.7bn, about 1.3% a year.
EPE: Mark Patterson, CFO, FY2026 Q4 release, August 2026: “In fiscal 2026, Cisco achieved its highest productivity metrics in 30 years measured by revenue, non-GAAP operating margin, and earnings per employee.”
FY2024 10-K, MD&A: “We expect to reinvest substantially all of the cost savings from these restructuring plans in our key priority areas and key growth opportunities. As a result, the overall cost savings from these restructuring plans are not expected to be material for future periods.” Same language in the FY2024 annual report.
Share price: $80.06 on 27 March 2000 not exceeded until 10 December 2025, CNBC and Bloomberg. June 2026 high $130.37 is quoted price, not TSR. CPI-U March 2000 to mid-2026, factor ~1.94, price only. Dividends improve TSR; I have not calculated a 2000 purchase with reinvestment.
Scientific-Atlanta $6.9bn, NDS $5bn, later sold to Technicolor for $600m. Purchase minus sale, not lifetime loss. Intercloud switched off.
Lazonick and Hopkins, “The Pursuit of Shareholder Value: Cisco’s Transformation from Innovation to Financialization”, Institute for New Economic Thinking working paper 202. $152.3bn buybacks October 2001–October 2022, 95% of net income, plus $55.5bn dividends. Programme announced 14 September 2001, now about $172bn and 6.2 billion shares.
Graeber, Bullshit Jobs (2018). Also “From Managerial Feudalism to the Revolt of the Caring Classes” and this interview.
Councils: 2007, nine to three in May 2011. Chambers memo on lost credibility a month earlier. “More than 500 executives”: Business Insider, citing 2008 Fast Company. Analyst: Brian White, Ticonderoga Securities, quoted here. Robbins CEO from 26 July 2015; Chambers executive chairman until 2017.
VP count is not disclosed. I have not used a figure.
Spin-ins. Andiamo, February 2004: $750m for the 56% Cisco did not own (already 44%), 317 staff averaging $2.3m in stock, Bloomberg, source of the risk quote. Nuova: $70m in, 80% owned; $678m maximum success-based payout, $10m minimum. Insieme: already about 85%, remainder max $863m; $135m seed and core-switching point, Business Insider. $40m per principal and $2bn total also Business Insider, anonymous; treat as reported. Internal resentment: Network World. Body $2.3bn is published maxima, not cash paid. The four: Mario Mazzola, Luca Cafiero, Prem Jain, Soni Jiandani.
Mazzola defended the model: tens of billions of revenue against modest acquisition cost; earn-outs tied to revenue and margin; the four principals not in the first spin-in’s compensation. He may be right about the returns. The objection is who carries risk and who is on the list. Robbins ended the practice; the team left about ten months into his tenure.
Wayne Cascio, Clifford Young and James Morris, “Financial Consequences of Employment-Change Decisions in Major U.S. Corporations”, Academy of Management Journal 40(5), 1997: 5,479 S&P 500 employment changes, 1980–1994. ROA and industry-adjusted stock returns in this review. Fortune 100: Kenneth De Meuse, Paul Vanderheiden and Thomas Bergmann, “Announced Layoffs: Their Effect on Corporate Financial Performance”, Human Resource Management 33, 1994. Meta-analysis: Steel and House, “Short-term pain for long-term gain? A longitudinal meta-analysis of downsizing-financial performance relationships”, Frontiers in Behavioral Economics, July 2024. Cascio et al. found combined headcount cuts plus asset restructuring outperformed. Pure headcount reduction does not pay.
Splunk: bought growth, not a write-off. FY2026 Q4 security revenue +14%; CFO says they have turned the corner; no write-down. Earlier drag was licences to subscriptions, an accounting effect.
SEC Rule 10b-18, 1982: safe harbour from manipulation claims if a company buys its own shares within daily limits. Buybacks on this scale were rare in the United States until then. Restricting it is Lazonick’s central policy recommendation.
Disclosure: EU Collective Redundancies Directive and aml § 15-2 aggregate per establishment and per event. SEC disclosure is a third system. A multinational can distribute reductions so a global annual total is hard to reconstruct.
Germany: BetrVG § 102, heard before every dismissal; without consultation, void. Objections include social selection and another available job. Not a § 99 veto. Betriebsrat establishments: dismissal rates ~1.6pp lower, this study; pattern held through the pandemic. Observation, not a consent right.
Unfair dismissal: aml § 15-12. Invalidity possible (continuation of employment); damages whatever the court finds reasonable after economic loss, the parties’ circumstances and everything else. No statutory two-year cap. Case law often limits future loss; subsequent income generally deducted. I am not pricing anyone’s settlement.
Notice: § 15-3. After at least ten years: four months from 50, five from 55, six from 60. Seniority in some agreements, including LO–NHO. Distinguished from § 15-7: HR-2019-424-A (Skanska).
Suitable other work: aml § 15-7. From 1 January 2024 the third paragraph extends it to other Norwegian companies in the group; § 14-2 the same for preferential rights. Hiring in one part and downsizing in another can be connected, where the facts warrant it: HR-2018-880-A (Linjebygg). I am not applying that to 2025 vacancies. I do not presently have those advertisements.
Collective redundancies: § 15-2. Other terminations not based on the employee’s circumstances count if at least five are dismissed. A settlement does not, by itself, take that off the table. The essay is about the individual § 15-7 questions a settlement can leave untested.
arbeidsmiljøloven. I am not alleging that any law was broken in my case. A settlement instead of a dismissal is lawful. Not a compliance complaint. If that bothers you, change the law.