The Conglomerate Discount

At the end of 2018 General Electric was worth about $88 billion, down more than 80% from a peak of over $500 billion in 2000. GE then broke itself into three independent companies: GE HealthCare, GE Vernova (energy equipment and services), and GE Aerospace. As of August 31, 2026 those three companies are worth roughly $620 billion combined. GE did not invent a slew of new products to accomplish this. It did not discover a new market. What it mainly did was remove a layer of management.

This is the conglomerate discount at work, and it is one of the oldest and best-documented results in corporate finance. It is also, judging by the behavior of some of the largest companies in the world, one of the most widely ignored. That combination — a well-understood value destroyer that sophisticated managements keep choosing anyway — is the puzzle this note is about.

The conglomerate discount is a statement about the value destroyed when a parent company owns two subsidiary businesses that have no operating synergies between them. Call them A and B. If A and B share nothing — no customers, no distribution, no manufacturing base, no technology — then the parent company sitting above them contributes no revenue and no cost savings. What the parent does contribute is expense: a corporate headquarters, a corporate staff, a chief executive and a chief financial officer and a general counsel and a board, all of whom are paid, and none of whom sell anything. Those are negative cashflows, and their discounted value is the conglomerate discount.

The direct costs are easy enough to see on an income statement. The indirect costs are larger and much harder to quantify. The time that A's management team spends preparing board decks, defending its budget, and attending corporate reviews is time not spent running A. The capital that A generates is not automatically A's to reinvest; it goes into a corporate pool and is allocated by people one level removed from either business. Recruiting suffers, because the operating executive who wants to run a company ends up running a division. Each of these is a real loss of value to the subsidiaries, and each belongs in the conglomerate discount.

Two implications follow immediately. The first is a rule for mergers: the way to create value by combining A and B is to make sure the operating synergies between them are worth more than the negative cashflows the parent company introduces. The second implication is a rule for breakups: shareholder value can be unlocked by spinning the subsidiaries out as independent companies and shutting the parent down. Nothing about the operating businesses needs to improve. Value is created purely by driving the parent's negative cashflows to zero.

That second implication has been the core thesis behind a large share of private equity activity for forty years. The most famous example remains KKR's purchase of RJR Nabisco — a food company and a tobacco company stapled together under a holding company with essentially no operating logic connecting them. GE is the more recent and more spectacular case. Splitting into three and eliminating the parent enabled what is widely regarded as the greatest corporate turnaround in history, and it was accomplished without a single new invention.

It is worth measuring that rather than asserting it, because GE spent most of the preceding period looking like a company that had earned its structure. At the end of 2000, GE's trailing five-year return was 33.9% a year against the S&P 500's 18.0% — nearly sixteen percentage points a year ahead. A shareholder looking at that record would have concluded the conglomerate form was vindicated. What followed was two decades of the opposite. Measured at every month-end from 2000 onward, GE's trailing five-year return trailed the index 71% of the time. The worst of those windows ended in July 2020, with GE compounding at negative 23.0% a year while the S&P 500 compounded at positive 11.4%. The widest gap, two months later, was 36.2 percentage points a year.

Then the structure changed, and so did the record. GE's trailing five-year return today is 39.3% a year against the S&P 500's 12.7% — twenty-seven points a year ahead, from thirty-six points behind. Measured from the first day GE HealthCare traded separately, GE Aerospace has returned 546% against the index's 111%, and GE Vernova, from its own first day, 588% against 51%. One of the three has not worked, and it is worth saying so: GE HealthCare has returned 18.5% since separation against the index's 109.1% over the same window. Separation removes a structural cost; it does not confer a good business, a good market, or good management.

So we have a well-established mechanism with a well-established remedy. Yet the structure persists at some of the most admired companies in the world. Let us look at three modern examples of the conglomerate discount in action, and then ask why the people running these companies keep choosing it.

Microsoft Enterprise and Xbox

Microsoft is primarily an enterprise software and services company. Inside Microsoft, however, is a consumer retail business: Xbox. The enterprise business sells to chief information officers on multi-year contracts; Xbox sells plastic boxes and $70 games to teenagers at Best Buy. The two go after entirely different customers through entirely different channels with entirely different economics. They share almost nothing. Even on the Redmond campus the two organizations are physically separated, with minimal interaction between them.

The financial mismatch is stark, and it has widened. In fiscal 2026 Microsoft earned roughly $155.2 billion of operating income on $331.8 billion of revenue, a margin of about 47%, with revenue up 18% on the year. Gaming revenue over the same twelve months was $21.8 billion, down 7% — 6.6% of the company, and shrinking, inside a company growing at eighteen. Gaming does not earn anything close to a 47% margin; a mid-teens operating margin is the generous read. On that basis Xbox produces something in the region of 2% of Microsoft's operating profit from 6.6% of its revenue.

That gap is not a curiosity. It is the entire explanation for a decade of strange behavior inside the gaming division. In 2023 Microsoft set an across-the-board 30% profit-margin target — internally called "accountability margins" — well above anything Xbox had ever achieved. The price increases, the studio closures, the cancelled projects, and the successive rounds of layoffs across Xbox all trace back to a business being held to a standard set by a cloud and software portfolio running at 40%-plus. The recurring public question, "why is Microsoft in the gaming business at all?," is the market asking the same thing in plainer language.

Now the part that makes Microsoft the interesting case rather than the obvious one.

Microsoft's five-year record has lately traced the same early sequence GE's did. At the end of January 2024, Microsoft's trailing five-year return ran 17.8 percentage points a year ahead of the index. By the end of 2025 that lead had narrowed to 3.4 points. In January 2026 it went negative for the first time, and by the end of June Microsoft trailed the index by 5.8 points a year. Set that beside GE, which was 15.9 points ahead at the end of 2000, 2.2 points ahead by 2002 and 5.0 points behind by 2006, and the two records trace the same early sequence at different speeds.

Exhibit 1

Microsoft is tracing the sequence GE traced twenty-four years earlier

Trailing five-year annualized return less the index’s, in percentage points, by month from each company’s own peak

The paragraph above gives three of Microsoft’s readings and three of GE’s; this draws every month of both. Microsoft runs from January 2024, General Electric from December 2000 — the comparison note 9 makes. Source: Tiingo adjusted close, retrieved 2 September 2026. The S&P 500 is tracked via SPY.

The difference is what happened next, and it is the point of the whole comparison. GE had nothing strong enough to arrest the slide, and it continued for another fourteen years. Microsoft did. In the quarter reported on July 29, 2026, Azure grew 43% and passed $100 billion of annual revenue for the first time, and the Intelligent Cloud segment grew 32%. In the same quarter, the segment containing Xbox shrank 4%, and Xbox content and services fell 10%. By the end of August Microsoft's five-year gap against the index had closed back to 0.8 percentage points.

Exhibit 2

The cost stopped being visible in the share price

Total return, month-end, rebased to 100 at 31 August 2021 · dividends reinvested

The same five years measured in price rather than in points ahead of the index. Microsoft ends at 175 against the index’s 182 — a total return of +75.1% against +81.8%. The two lines part company through 2024 and 2025 and finish within seven points of each other, which is the 0.8-point gap above expressed as a level. Source: Tiingo adjusted close, retrieved 2 September 2026. The S&P 500 is tracked via SPY.

So the enterprise business did not merely carry the gaming business through the quarter. It closed a five-year performance gap while the gaming business was contracting. That is what a conglomerate discount looks like inside a company good enough to absorb it: the cost does not disappear, it stops being visible in the share price. And a cost that has stopped being visible is not a cost that has stopped being paid — it is a cost that no longer generates the pressure that would otherwise force somebody to do something about it. Opacity, again, doing the work it always does.

That is the version of the discount that is hardest to do anything about. When the cost shows up in the share price, somebody eventually asks about it. When it doesn't, nobody has to.

Amazon Retail and Amazon Web Services

Amazon is composed of two businesses. Amazon Retail is the one virtually every consumer knows, because they shop there. Amazon Web Services (AWS) provides cloud infrastructure to enterprise customers. The target customers are orthogonal: retail versus enterprise.

One might object that this is a poor example, because here one business genuinely depends on the other. Amazon Retail runs on AWS for compute, storage, databases, and the logistics systems that make the whole front end work. In fact the causality runs backwards from the way people usually assume: AWS was originally built inside the retail business decades ago because the retail business needed it, and only later did Amazon realize it could sell the same services to outside corporate customers.

However the relationship between the two could be replicated in full by a data center contract — exactly the kind of contract AWS already has with hundreds of other corporate customers, including competing online retailers. There is no need for the two to sit under a common parent in order for one to buy services from the other. What we have, then, is a variation on the traditional conglomerate discount, in which a customer and a supplier happen to share a holding company. The operating link is real but it is contractual, not structural, and a contract does not require a corporate headquarters to enforce it. Both businesses are terrifically run. That is precisely the point: the deadweight loss from the extra layer of management above them is incurred regardless of how well they are run.

There is a second cost specific to this arrangement, which is that the transfer price between the two is set internally rather than by a market. Amazon Retail is AWS's largest captive customer and pays a price no outside party negotiated. That obscures the true profitability of both businesses from the very shareholders who are supposed to be evaluating them. The reported numbers are lopsided enough as it is: in 2025, AWS generated $45.6 billion of segment operating income against $29.6 billion for (Amazon Retail) North America and $4.75 billion for (Amazon Retail) International. AWS accounts for a little over half of the company's operating profit on well under a quarter of its revenue.

And again we can see the discount in the stock. Over the same five years to August 31, 2026, Amazon returned 49.7% in total, or 8.4% per year, against 12.7% per year for the index. Amazon owns the largest retailer in the world and the largest cloud infrastructure provider in the world, and the combination has compounded at two-thirds the rate of the market.

Exhibit 3

Amazon has trailed the index for almost the whole period

Total return, month-end, rebased to 100 at 31 August 2021 · dividends reinvested

Amazon ended the five years at 150 against the index’s 182 — a total return of +49.7% against +81.8%, the figures in the paragraph above. Source: Tiingo adjusted close, retrieved 2 September 2026. The S&P 500 is tracked via SPY.

Disney Parks and Disney+

Disney is a different and considerably more damaging case. Here one subsidiary does not merely fail to help another; it actively destroys the other's value, and it does so specifically because the two sit under the same parent.

Start with how Disney made money for most of a century. Disney Studios creates content — films, television, animation — that it wants distributed as widely as humanly possible. Wide distribution earns money directly, but its far more valuable function is to create and sustain global brands: Mickey Mouse, Cinderella, the Marvel Cinematic Universe, Star Wars, etc. Those brands are what pull families from Osaka and São Paulo and Manchester to Anaheim and Orlando and Paris. The theme parks monetize the brands, and they do it at enormous scale — Experiences (theme parks, cruise lines, etc.) is by a wide margin the biggest profit generator in the company. The parks then reinforce the brands, which makes people want more content, which the studios supply. It is a flywheel, and the flywheel's fuel is distribution breadth.

Disney+ breaks the flywheel. Disney+ takes content made by Disney Studios and, rather than distributing it as widely as possible, distributes it as narrowly as possible — to Disney+ subscribers. That is not a mistake; it is the correct strategy for a streaming service considered on its own. A subscription business is worth more when it is the only place to get the product, so Disney+ is structurally incentivized to withhold Disney content from theaters, from rival streamers, and from licensing partners. Its incentive is exclusivity. The studios' and the parks' incentive is ubiquity.

Under separate ownership, this conflict would be resolved by price: a streaming company would have to bid against theatrical and licensing windows for Disney content, and would win only when it was willing to pay more than the brand value it was destroying. Under common ownership, the conflict is resolved by fiat, and the fiat has consistently favored the smaller business. So the brand-building engine has been throttled to feed a subscription business that in fiscal 2025 produced $1.3 billion of operating income — against roughly $10 billion at Experiences, which is to say the parks and cruise lines earned close to three-fifths of Disney's total segment operating profit. Disney has been starving its largest profit engine to feed one of its smallest.

The market has noticed. Disney+ launched on November 12, 2019. In the seven years before that launch, Disney stock returned 224%, or 18.3% per year — 3.8 percentage points a year ahead of the S&P 500 over the same stretch. From the launch through August 31, 2026 it has returned negative 19.5%, or negative 3.1% per year, while the index compounded at 16.0%. Disney went from beating the market by roughly four points a year to trailing it by nineteen, and the hinge is the launch date. Over the last five years specifically the stock has returned negative 38.8%, or negative 9.4% per year, while the index compounded at 12.7%. Disney's market capitalization is about $191 billion today. It is worth pausing on that: a company holding the most valuable entertainment intellectual property ever assembled, plus the most profitable theme park business on earth, has destroyed nearly two-fifths of its shareholders' capital over five years in a market that rose more than 80%.

Exhibit 4

Ten years on, Disney is back where it started

Total return, month-end, rebased to 100 at 31 August 2016 · dividends reinvested

Over the ten years Disney returned +23.6% and the index +314.5%, dividends included in both. The window is ten years rather than five so that the stretch before the November 2019 launch is visible on the same axis as the stretch after it. Source: Tiingo adjusted close, retrieved 2 September 2026. The S&P 500 is tracked via SPY.

This is the conglomerate discount in its most pernicious form. It is not the overhead. It is one subsidiary systematically dismantling the economics of another, and the parent company being the mechanism that makes the dismantling possible.

Two caveats are worth stating plainly. First, three companies are three companies, not a controlled experiment. Each of these stocks has been buffeted by things that have nothing to do with corporate structure — Microsoft in particular absorbed the sharp repricing of AI-related capital spending that took hold across the market in the first half of 2026, and then recovered most of it in a single quarter. Second, none of these firms is badly run at the operating level; all three would appear on most people's list of the best-managed companies in the world. That is what makes the pattern interesting rather than trivial. The claim here is not that structure explains everything. It is that structure imposes a persistent, compounding drag that operating excellence has to overcome before shareholders see a dollar — and that over five years it has been large enough to swamp the operating excellence at two of these three, and large enough at the third to be visible in the record even though the operator eventually outran it. At Disney it overwhelmed the operator. At Microsoft the operator outran it. The structural arithmetic is the same in both; only the scoreboard differs.

Why It Persists

So we arrive at the real mystery. The conglomerate discount has been documented in the academic literature since the 1990s, it is taught in every introductory corporate finance course, the remedy is simple and legally routine, and the GE example is sitting right there with a $530 billion demonstration attached. Why do sophisticated boards and sophisticated managements keep choosing the structure anyway?

The answer is that the conglomerate discount is a cost to shareholders and a benefit to managers, and it is managers who make the decision. It is worth saying first that the discount is not a law of nature; a handful of multi-business companies have compounded capital at exceptional rates for decades, and the reasons they are exceptions are instructive.

Start with compensation. Executive pay scales far more tightly with the size of the enterprise than with its per-share value. Revenue, headcount, market capitalization, and span of control are the variables that set peer groups, and peer groups set pay. A chief executive who spins off a third of the company has, from the compensation committee's perspective, just demoted himself. The shareholder gets two focused companies; the executive gets a smaller job and a smaller pay package. Nothing in a standard compensation structure rewards him for that trade.

Second, and more defensibly, managers are undiversified in a way shareholders are not. A shareholder who wants exposure to both enterprise software and video games can simply buy two stocks, and can rebalance between them at will and at almost no cost. A chief executive cannot diversify his human capital that way; his career, reputation, and unvested equity are all concentrated in one company. Corporate diversification therefore delivers real risk reduction to the people running the firm while delivering none to the people who own it — shareholders were already diversified, and are now paying for the privilege of being diversified again, badly, inside a corporate wrapper. This is the oldest agency problem in the book and it has never stopped operating.

Third, there is the internal capital market argument, which is the honest intellectual case for conglomerates and which is largely wrong in practice. The theory holds that a parent company can allocate capital among its subsidiaries better than external markets can, because it has better information. Sometimes true. But internal capital allocation is also political, and the empirical record is that cash tends to flow toward the units with the strongest advocates rather than the strongest returns. Cross-subsidization lets a weak business survive on the earnings of a strong one for far longer than an external capital market would have permitted. Xbox held to a margin standard it cannot meet is one version of this pathology; Disney+ funded out of the brand equity the parks depend on is a more expensive version.

Fourth, and least discussed, conglomerate structures make performance illegible. When two businesses with different economics are reported together, or when transfer prices between them are set administratively, outside investors genuinely cannot tell how each is doing. Opacity is not usually the goal, but it is reliably the consequence, and it reduces the pressure on management that transparency would otherwise create.

Put those four together and the persistence stops being mysterious. The structure survives not because anyone believes it creates value but because the people empowered to dismantle it are the people it benefits.

What changes the equation is external pressure. GE did not break itself up in a burst of introspection; it did so after a decade of collapsing valuation, board turnover, and sustained activist attention. That is generally the pattern. Spin-offs are legally and mechanically straightforward — a properly structured spin-off is tax-free to shareholders — so the binding constraint is never feasibility. It is willingness.

And GE is not an isolated data point. The last four years have produced an unusually dense run of separations, and the pattern in the outcomes is hard to miss. GE Vernova, the energy business spun out of GE in April 2024, is worth about $239 billion today, having returned more than 580% since it began trading separately, and is now among the five most valuable industrial companies in the United States. Western Digital separated its flash memory business as Sandisk in February 2025; both stocks fell on their first day of independent trading, which looked at the time like a verdict, and in the eighteen months since Sandisk has returned 3,124% and Western Digital 823%. Kellogg split into Kellanova and WK Kellogg in October 2023, and within two years both halves had been acquired outright — Kellanova by Mars for $83.50 a share, about $35.9 billion, and WK Kellogg by Ferrero — which is to say the pieces were worth more to strategic buyers separately than the combination had ever been worth to the market. GSK's separation of its consumer health business as Haleon in July 2022 was the largest European listing in a decade, at roughly £30.5 billion on day one. 3M's spin-off of Solventum in April 2024 is the useful counterexample within the pattern: it languished for its entire first year, likely on forced institutional selling, before recovering to trade near all-time highs, a reminder that the value shows up on the market's schedule rather than management's.

Two of those outcomes deserve an asterisk. Sandisk and GE Vernova both landed in the path of an enormous, unforecastable demand shock — AI data centers need memory and they need power — and no honest reading attributes sixfold and thirtyfold moves to corporate structure. What structure did was determine who captured them. Inside Western Digital, a flash business booming while the hard drive business did something else nets out to a muddle on a consolidated income statement, and the shareholder who wanted flash exposure could not buy it. That is the under-appreciated cost of the conglomerate form: not only the value it destroys on average, but the optionality it denies shareholders when one of the pieces turns out to be sitting in front of a hurricane.

For investors the practical implication is that the screen worth running is not for cheap stocks but for structural mismatches: a company containing two businesses with materially different customers, different margins, and no shared cost base, trading at a multiple somewhere between the two. That gap is the discount, and it is recoverable, but only on a catalyst. The catalyst is almost always an activist, a proxy fight, a new chief executive with no attachment to the acquired business, or a valuation collapse severe enough to make the board's position untenable. Absent one of those, the discount does not close on its own; it compounds. Microsoft, Amazon, and Disney have each demonstrated, in their own way, exactly how long a great business can carry the cost of an unrelated one before anybody does anything about it.

For boards the implication is more uncomfortable, and it is the one worth ending on. Every argument for keeping an unrelated business — optionality, diversification, future synergies not yet visible, the strategic importance of being present in a growing category — is unfalsifiable, and every one of them is available at any time to justify any structure. The falsifiable question is narrower and much harder to dodge: if these two businesses did not already sit under the same roof, would this board vote to acquire one for the other, at today's price, and could it name the synergies that would justify the parent's cost? For a great many conglomerates the honest answer is no. The distance between that answer and the structure they continue to defend is the conglomerate discount, and eventually somebody else collects it.