The moment a company operates two or more businesses with different risk profiles, a single EV/EBITDA multiple becomes a blunt instrument that produces the wrong answer. Think about what a multiple actually represents: it is the market's shorthand for "how much uncertainty and growth potential lives in every dollar of earnings?" A defense contractor and a consumer app carry completely different answers to that question. Slapping one multiple across both is like using an average speed to describe a road trip where you drove 80 mph on the highway and sat still in traffic for an hour. The average is technically accurate and practically useless.

SpaceX illustrates this precisely. The rocket launch business is capital-intensive, government-contract-driven, and slow-growth. Starlink is a high-growth, recurring-revenue SaaS-adjacent business. xAI is pre-revenue and speculative. One multiple cannot honor all three realities at once.

SOTP treats each business unit as if it were a standalone public company, values it on the right comparable metric, then adds the pieces together. You are essentially building a portfolio valuation rather than a firm valuation. The four-step process is straightforward:

1. Segment the business into units with genuinely different economics.
2. Choose a valuation method per segment: revenue multiples for high-growth units, EBITDA multiples for mature cash generators, DCF for long-duration assets.
3. Apply comparable company benchmarks for each segment individually.
4. Sum the parts, then subtract net debt and central corporate costs.

The critical discipline is step one. You must segment on economic reality, not on org-chart lines. If Starlink's revenue props up SpaceX's balance sheet and allows the launch business to underprice contracts to win NASA deals, those two units are not independent. You have to model the subsidy explicitly.

Markets consistently value diversified companies at a discount to the sum of their individual parts, and that discount has a name: the conglomerate discount. Academic research (Berger and Ofek, 1995) put the average discount at 13 to 15 percent. More recent work finds it persists at 5 to 10 percent depending on the sector.

Here is why it exists. Imagine you hire a general contractor to build your house, and that same contractor also runs a landscaping business, a tile import company, and a commercial HVAC firm on the side. You cannot easily audit whether the margin on your project is subsidizing a money-losing part of their empire. You demand a discount for that opacity. Investors do the exact same thing. They apply a haircut to conglomerates because cross-subsidies, shared overhead, and tangled incentives make it hard to know whether management is allocating capital to its highest-value use.

For SpaceX specifically, the discount question is acute. If xAI is burning cash and Musk directs Starlink enterprise contracts toward xAI infrastructure at below-market rates, SpaceX public shareholders effectively fund a private AI company. The market will price that risk.

Practice this framework on a real case: the ge-welch-1981 case on BoardroomIQ puts you in the room.

The conglomerate discount is not inevitable. Some business combinations generate a conglomerate premium because the units are genuinely more valuable together. Amazon Web Services would not exist without Amazon's retail operation funding the infrastructure build-out. The discipline is distinguishing synergies that are real and quantifiable from synergies that are a story told to justify a bad acquisition.

In an interview, use this test: can you name a specific cost line or revenue stream that only exists because of the combination? If yes, you have a synergy. Quantify it and add it back. If the answer is vague ("they complement each other strategically"), apply the discount.

Build a paper SOTP model. Pick any conglomerate in the news, Alphabet, Berkshire Hathaway, or a fictional one your interviewer constructs. Segment it in five minutes, assign a benchmark multiple to each unit, and narrate your logic out loud. The narration is what interviewers grade.

Practice the discount conversation. Have a study partner ask you "why wouldn't you just use one multiple?" Force yourself to answer in two sentences without notes. If you cannot do it in two sentences, you do not own the concept yet.

Pressure-test your segment boundaries. For any company you segment, ask: does unit A's performance change materially if I remove unit B? If yes, they are entangled and you need to model the dependency, not ignore it.

The best way to practice conglomerate valuation is under realistic pressure, with a case that fights back. Open the ge-welch-1981 case on BoardroomIQ and work through the portfolio rationalization prompt until you can segment, discount, and defend your number in real time.