Teardown

Daily digest · 2026-07-29

Scan #018: The layer you don't own

Four challengers selling layers on top of someone else's platform and four incumbents defined by who they answer to — FedEx, Nordstrom, CCC, Comfort Systems, Keychain, Emerald AI, Tractable, Triple Whale.

The thread running through today’s eight: who owns the layer where the money actually settles. All four emerging companies sell software that lives on top of somebody else’s system — Tractable’s estimates land inside CCC’s platform (the two spent seven years in litigation over exactly that; both are covered today, on purpose), Triple Whale referees between Shopify and Meta who both want the whistle, Emerald AI orchestrates workloads on NVIDIA hardware NVIDIA may orchestrate itself, and Keychain is trying to convert free retailer logos into paid plant software. On the incumbent side, the same question in reverse: FedEx built the layer and is watching Amazon sell a cheaper one, Comfort Systems owns the one layer nobody can bundle — craft labor — while CCC owns the claims rails outright, and Nordstrom paid $6.25B to stop renting its strategy from the public market.

Keychain — Supply chain · Emerging. AI manufacturing platform matching CPG brands and retailers to contract manufacturers, founded 2023 by Handy/Angi’s Oisin Hanrahan and Umang Dua, ~$78M raised. The demand side came free — 8 of the top 10 US retailers pay nothing — while revenue sits on mid-tier manufacturers paying for leads: one Trustpilot account describes $5,000 for a three-month trial, roughly ten dead leads, and a forced three-year contract. The open question is whether the lead-gen wedge converts into system-of-record plant software (KeychainOS) before supply-side churn erodes the only side that pays.

Emerald AI — Energy · Emerging. Varun Sivaram’s Washington DC startup ($68M in 16 months from Radical, NVIDIA, Eaton, GE Vernova, even In-Q-Tel) sells curtailment as currency: a data center gives back ~25% of its power a few hundred hours a year and jumps the interconnection queue. The uncomfortable finding: through July 2026 there is no disclosed paying customer, pricing, or revenue — traction is demos and pilots — and the question is whether owners of scarce GPU fleets, where idle silicon burns more in depreciation than it saves in power, accept live utility dispatch once NVIDIA folds flexibility into its own reference designs.

Tractable — Insurance · Emerging. London computer-vision pioneer for photo-based damage estimates, ~$185M raised, $1B-valued in 2021, used by The Hartford and Admiral. What other coverage misses: CCC holds ~85% of US auto estimatics per Tractable’s own March 2024 antitrust filing — a case that began with CCC accessing Tractable’s product under a fake identity (“JA Appraisal”) and ended in a sealed January 2025 settlement. No new capital since July 2023, headcount down ~40% from peak, and every system of record it depends on now bundles its own photo-AI.

Triple Whale — Ecommerce · Emerging. Columbus/Israel attribution platform born from Apple’s ATT crackdown, ~$55M raised with Shopify as investor, ~30,000 brands, Moby AI agents since 2025. Its referee position is being squeezed from both benches: Shopify keeps deepening free native analytics inside the admin, and Meta’s Andromeda closes the optimization loop on its own signals — while the company hasn’t raised in 41 months and sub-$100 pixel clones commoditize the scoreboard from below.

FedEx — Logistics · Incumbent, at risk. The $94.7B express pioneer completed its founding-era unwind in thirteen months: Fred Smith died June 2025, FedEx Freight spun off June 1, 2026 with a $4.1B dividend back to the parent. The verdict turns on a number FedEx doesn’t publish in its own releases: by ShipMatrix’s March 2026 count Amazon moved 6.7B US packages in 2025 to FedEx’s 3.6B — and Amazon began selling that network to outside shippers in May 2026, aiming directly at the ground business where FedEx’s profit actually lives. $6B of DRIVE/Network 2.0 cuts read as margin defense, not moat.

Nordstrom — Retail · Incumbent, well positioned. The 125-year-old Seattle retailer went private May 2025 at $6.25B — the family plus Mexico’s Liverpool at 49.9% — at roughly half what the family offered in 2018. First private year: record ~$15.9B revenue, Rack compounding at 20+ stores annually, and a buyout financed with ≤$450M of new borrowings while luxury rival Saks Global — built the levered way — hit Chapter 11 in January 2026. The dying-category tailwind is real: being the last well-capitalized full-service player standing is a position, and the family just stopped paying the public market to second-guess it.

CCC Intelligent Solutions — Insurance · Incumbent, well positioned. The 46-year-old network between 27 of the top 30 US auto insurers, 30,000 repair shops and the parts ecosystem crossed $1B revenue in 2025 with 99% gross dollar retention — and its stock halved anyway, which is why Elliott took a stake and a sale process is reportedly live in 2026. The detail that pairs with today’s Tractable page: CCC spent seven years litigating against the AI challenger, settled sealed, and now books ~$120M of AI revenue itself — the rare incumbent monetizing the technology that was supposed to kill it.

Comfort Systems USA — Construction · Incumbent, well positioned. The 1997 Houston HVAC roll-up became the AI buildout’s indispensable trade: Q2 2026 revenue $3.27B (+50%), net income up 92%, backlog $14.1B. The number that cuts both ways: hyperscalers were 58% of Q2 revenue, up from 33% two years ago — the deepest customer concentration of any company we’ve covered that isn’t Symbotic. The position is unassailable for years — craft labor and sold-out modular capacity can’t be bundled or cloned — but the 42x multiple, not the company, is the fragile thing, and the stock slipping on a record beat says the market knows it.

Full pages linked above. Yesterday’s scan: Scan #017.

Full deep dives