Sum-of-the-Parts Valuation: When to Break a Business Apart

A sum-of-the-parts valuation is really an admission. It's what you reach for when a single multiple can no longer tell the truth about a business — when one number, applied uniformly across a company running three or four fundamentally different economic engines, systematically overpays for the weak segment and underpays for the strong one. Blend them together and the multiple lands somewhere in the middle, satisfying nobody and describing nothing.

That's not a reason to reach for SOTP every time a company has more than one product line. Most diversified businesses are coherent enough that a single owner-earnings multiple does the job fine. SOTP earns its keep in three specific situations: true conglomerates holding unrelated businesses under one ticker, holding companies whose value is mostly the sum of stakes in other entities, and single companies where one division is mature and cash-generative while another is early-stage and still burning capital. Miss all three conditions and SOTP just adds complexity without adding accuracy.

I'll walk through the mechanics using Alphabet (GOOGL) — a company that hits the third condition about as cleanly as any large-cap in the market, with a highly profitable advertising and platform business, a fast-growing but historically thinner-margin cloud segment, and a portfolio of early-stage bets that lose money by design. We'll build the pieces, sum them, and then check the answer against what the market is actually paying today, which turns out to be the more interesting half of the exercise.

When a Blended Multiple Actually Lies to You

Start with the diagnostic question, not the math: would an acquirer buying only one segment of this company pay the same price-to-earnings ratio a generalist investor pays for the whole thing? If the answer is obviously no — because one piece is a near-monopoly cash machine and another is a venture-stage experiment — a blended multiple is quietly transferring value from the strong segment to the weak one in your model, and you won't see it happen unless you break the pieces apart first.

Sum-of-the-Parts Enterprise Value = Σ (Segment Earnings or Revenue × Segment-Appropriate Multiple) − Capitalized Corporate Overhead

Convert that enterprise value to an equity value the same way you would for a single business: add cash and marketable securities, subtract total debt and any minority interests, then divide by shares outstanding. See our guide to enterprise value if that bridge isn't second nature yet. The aggregation step is arithmetic. Nobody gets this wrong. The multiple you assign each segment is where the entire argument for or against SOTP actually lives, and it's the part analysts tend to gloss over fastest.

Alphabet, Segment by Segment

Alphabet reports three segments: Google Services, Google Cloud, and Other Bets. Here's what each looked like for fiscal 2025, and the assumption attached to each — treat these multiples as a starting hypothesis to argue with, not a verdict.

Google Services — the ad and platform machine

Google Services — Search, YouTube ads, Android, Play, and hardware — generated $342.7 billion in revenue for fiscal 2025, up 12% from $304.9 billion in fiscal 2024, per Alphabet's fourth-quarter and full-year 2025 results. Segment operating income rose by $18.1 billion year over year. This is the part of the company that behaves like a mature, dominant platform business: enormous margins, modest incremental capital needs relative to the cash it throws off. A blended multiple appropriate for a diversified conglomerate would badly undervalue this piece standalone. A reasonable starting multiple here sits in the 7–9× revenue range — roughly where the market has historically priced mature, highly profitable digital-advertising franchises — which puts a standalone Services business at somewhere between $2.4 and $3.1 trillion. Call it $2.7 trillion at the midpoint.

Google Cloud — priced for growth, not for the segment it used to be

Google Cloud posted quarterly revenue of $12.3 billion, $13.6 billion, $15.2 billion, and $17.7 billion across 2025's four quarters — roughly $58.8 billion for the full year, with the fourth quarter alone running at close to a $70 billion annualized pace. Segment operating income improved by $7.8 billion year over year as the business crossed into durable profitability. Growth-stage infrastructure businesses trade richer than mature platforms on a revenue basis, even at lower margins, because the market is pricing the growth curve rather than today's earnings. A 6× revenue multiple — toward the lower end of where hyperscaler-adjacent infrastructure businesses have traded — puts Cloud at roughly $350 billion standalone.

Other Bets — mostly Waymo, and mostly a guess

Other Bets is where this exercise gets honest about its own limits. The segment's operating loss widened by $3.1 billion from 2024 to 2025, driven largely by compensation expense tied to Waymo's rising valuation — which is itself the best evidence of what the segment is actually worth. In February 2026, Waymo raised $16 billion in new funding at a $126 billion post-money valuation, more than double its $45 billion valuation from October 2024. Alphabet led that round and remains the controlling owner, though its exact stake isn't precisely disclosed; outside investors are estimated to hold somewhere in the 15–20% range, which would put Alphabet's share of Waymo alone at roughly $100 billion. I'm less confident in this number than in the other two segments — a single private funding round is a thin reed to hang a hundred billion dollars on, and the rest of Other Bets (Verily, Wing, and the smaller bets) is close enough to value-neutral standalone that it's treated as a rounding error here.

Add it up: roughly $2.7 trillion for Services, $350 billion for Cloud, and $100 billion for Alphabet's stake in Waymo gets you to about $3.15 trillion, before two adjustments that matter. First, subtract a capitalized value for unallocated corporate costs — what Alphabet's segment disclosures call "Alphabet-level activities" — which reduce the sum of segment operating income below the consolidated figure and get skipped in almost every back-of-envelope SOTP that circulates online. Second, net out the balance sheet: $126.8 billion in cash and marketable securities at year-end 2025, against debt that grew meaningfully after a $37.3 billion note issuance during the year to help fund the AI infrastructure buildout. Net those two adjustments against each other and they roughly wash. So the SOTP estimate lands somewhere around $3.1 to $3.2 trillion.

Here's the part that makes this worth doing right now rather than as an abstract exercise. Alphabet's market capitalization crossed $4 trillion this year and stood at roughly $4.3 trillion in mid-July, on the back of a rally that has more than doubled the share price and added over $2 trillion in market value over twelve months — moving Alphabet from the fifth-most-valuable company in the world to the second, as Bloomberg detailed in its July 1 report on the stock's run. That's not a discount to the sum of the parts. That's a premium of $1 trillion or more, depending on which multiples you trust. A year or two ago, the standard SOTP argument for Alphabet was that the market was undervaluing Cloud and Other Bets by burying them inside a single ad-business multiple. That argument has essentially inverted.

The Discount, the Premium, and Why the Market Might Be Right Either Way

A conglomerate discount is the market's way of pricing in costs that SOTP math tends to ignore: capital gets allocated across segments by a single management team that may not be equally skilled at running an ad platform and an infrastructure business, a breakup would trigger real tax leakage on unrealized gains, and investors who want pure-play exposure to Cloud can't get it without also buying the ad business and the robotaxi bets. None of that shows up in a segment-multiple sum. All of it shows up in the price an actual buyer will pay for the whole company.

A conglomerate premium runs the opposite direction: cross-subsidization can be a feature rather than a bug when a mature cash generator is funding a genuinely valuable option that couldn't get financed on its own economics. Waymo, funded off Alphabet's balance sheet rather than through an independent IPO, likely wouldn't have raised $16 billion at a $126 billion valuation as a standalone pre-scale business — the backing of a company generating well over $100 billion a year in net income is part of what made that valuation credible. So is management's demonstrated willingness to keep funding a decade-plus bet before it produced a dollar of segment profit. That's capital-allocation discipline the market is entitled to price.

Whether Alphabet's current price is right depends on which side of this you believe, and the SOTP math alone doesn't answer it. Price is what the market pays today; intrinsic value is what the underlying cash flows across all three segments are actually worth, and reasonable analysts can build defensible models landing anywhere from $3 trillion to $4.5 trillion depending on the Cloud growth trajectory and the multiple they're willing to pay for the Services core. The honest position is that the current price sits inside a wide range of plausible SOTP outcomes rather than clearly outside it in either direction.

Where the Math Breaks

SOTP fails in predictable ways. Watch for these.

• Cherry-picked peer multiples. It's easy to find a peer trading at 12× revenue if you want the sum to look big, and just as easy to find one at 4× if you want it small. Name the peer set and the multiple range explicitly, and show your work when you pick a point inside it.
• Ignored corporate overhead. Segment operating income sums to more than consolidated operating income at almost every diversified company, because unallocated corporate costs sit outside the segments. Skip that deduction and the whole exercise is overstated before the multiples even matter.
• Cross-holdings and minority interests treated as free. If a segment's reported earnings include income from a stake the parent doesn't fully own, valuing 100% of that segment's earnings at a full multiple double-counts value that belongs to someone else's balance sheet.
• Treating a private funding round as a precise public-market price. Waymo's $126 billion valuation is real information, but it's one transaction, priced by investors who may have reasons — strategic access, a desire to be in business with Alphabet, enthusiasm about autonomous driving specifically — that a pure financial buyer wouldn't share. Use it as a data point, not a valuation.

None of this makes SOTP useless. It makes it a compass, not a GPS coordinate — a way to check whether a blended multiple is quietly hiding a large mispricing in one segment, not a machine for producing a number you should trust to three significant figures. Keep that distinction clear from the price quoted on a screen, which reflects what someone was willing to pay this morning, not necessarily what the underlying businesses are worth. Widen your discount rate or multiple assumptions by a margin of safety before treating any SOTP output as precise.

Key Takeaways

• SOTP earns its keep for true conglomerates, holding companies, and single companies where one division is mature and another is early-stage — not for every multi-product business.
• The aggregation math is trivial. The segment multiples you choose are where the entire argument lives, and they deserve to be named and stress-tested, not assumed.
• Alphabet's fiscal 2025 segments — roughly $2.7 trillion for Services, $350 billion for Cloud, and $100 billion for Alphabet's Waymo stake — sum to something in the $3.1–3.2 trillion range, against a market cap near $4.3 trillion in mid-2026. The market is currently paying a premium to this sum, not a discount, which flips the argument SOTP models made on this stock a few years ago.
• Always net a capitalized corporate-overhead deduction and the actual cash-versus-debt position before comparing an SOTP estimate to the market price — skipping either step is the most common way these models overstate their own precision.