Teardown

Daily digest · 2026-08-14

Scan #033: Four positions that broke, and four mechanics coming for what's left

Four incumbents whose defenses either failed on schedule — Allstate losing #1 auto to Progressive, FirstEnergy failing to keep the lights on after bribing the legislature, Norfolk Southern accepting a $85B rescue merger — or actually held (Home Depot's Pro-plus-SRS pivot), paired with four emerging companies attacking the physical mechanics of pick, power, plant and policy: Sereact's VLA-driven picking, Valar's HTGR microreactor, Arrakis's industrial-AI deployment layer, and Indigo's AI-underwriting bet in medical malpractice.

Eight companies today, split four and four: the entrenched half is the useful half. Three of the four incumbents on this list — Allstate, FirstEnergy, Norfolk Southern — have already run out of quiet ways to defend their positions and are now doing the loud ones: shedding customers, getting rebuked by regulators, or selling the company. The fourth, Home Depot, spent $18.25B to bolt distribution onto a footprint the internet was supposed to have killed by now, and it looks like it worked. The four emerging companies are each priced against a specific mechanic these incumbents built businesses around: picking by hand, generating with gas, deploying software with services, and pricing malpractice risk with actuarial regressions from the 1990s. Every one has an answer condition that lands inside 24 months.

Allstate — Insurance · Incumbent, at risk. Allstate is the second-largest US personal-lines writer, a Sears 1931 spin-out with ~$60B of P&C premium, and as of May 2026 it is no longer #1 in US personal auto — S&P Global’s trailing-12-month direct-premiums estimate put Progressive ahead of State Farm and Allstate for the first time since World War II. The specific rot: Allstate’s NAIC private-passenger-auto complaint index has been running above 2.0 (roughly twice what a carrier its size should produce) with the concentration in claims-denial and rate-renewal disputes, and the “Transformative Growth” plan launched under CEO Tom Wilson was supposed to close the direct-channel gap with GEICO/Progressive years before the ranking flipped. The retreat from California and Florida homeowners in 2023-2024 protected the balance sheet at the cost of the household bundle the agent model was built to sell.

FirstEnergy — Energy · Incumbent, at risk. An Ohio/Pennsylvania/New Jersey investor-owned utility holding company whose story is now inseparable from the House Bill 6 bribery scandal: FirstEnergy paid a $230M deferred-prosecution-agreement penalty in July 2021 for funneling ~$60M to former Ohio House Speaker Larry Householder in exchange for a $1B nuclear-plant subsidy, and PUCO tacked on a $250.7M penalty order in November 2025. The most damaging finding in 2026 isn’t the fine — it’s the July 2026 PUCO ruling rejecting FirstEnergy’s request to loosen the SAIDI/SAIFI thresholds counted as unacceptable reliability, followed a week later by ~10,000 Cleveland-area customers losing power in a 100°F heat wave. Lakewood is exploring legal action; Brookfield Super-Core’s 49.9% stake in FirstEnergy Transmission is the piece an activist would carve off first if the parent’s cost of capital keeps climbing.

The Home Depot — Retail · Incumbent, well-positioned. $164.7B FY25 revenue home-improvement retailer whose Pro segment now approaches half of sales despite a frozen housing market — the number that made the SRS Distribution acquisition (announced March 2024, ~$18.25B enterprise value on ~$1.1B of EBITDA, roughly 16.6x) look less like a distribution roll-up and more like a rebuild of the Complex Pro supply chain the DIY footprint was never designed to serve. The Blueprint Takeoff AI tool rolled out early 2026 lets a contractor upload plans and get automated material lists in minutes, and it plugs directly into Pro Xtra’s tiered credit and delivery. Four consecutive quarters of double-digit online growth through Q1 FY26 and a Q2 print due August 18, 2026 make the case that the pro-plus-distribution flywheel is the defense that actually works — the specific structural asset (the truck, the yard, the account manager, the aggregated demand for roofing/pool/landscape trades) is not one Amazon Business or Lowe’s can copy in a cycle.

Norfolk Southern — Logistics · Incumbent, at risk. NSC’s answer to the question “can an Eastern Class I railroad compete alone against a UP-BNSF West and CSX East plus a re-shored trucking network?” was decided on July 29, 2025 when Union Pacific announced an $85B stock-and-cash acquisition — the STB accepted the application May 28, 2026 (in abeyance pending supplemental filings by July 27, 2026), with closing guided to mid-2027. The structural weakness the deal admits: NSC’s FY25 operating ratio was 64.2%, roughly five points wider than UP’s 58.7% and worse than CSX’s, and Alan Shaw’s PSR reversal — replaced when he was fired for cause in September 2024 alongside CLO Nabanita Nag — did not close the gap. East Palestine liability has now cleared $1.7B in disclosed spend on top of $310M in federal and $600M in class-action settlements, with ~150 residents already trying to exit the class. The verdict is not “the merger fails” — it is “even after Ancora seated three directors in May 2024 and Mark George was promoted to fix the operation, the answer to the strategic question was to be acquired.”

Sereact — Supply chain · Emerging. A Stuttgart-founded (2021) physical-AI startup for warehouse picking whose Series B closed at $110M in April 2026 led by Headline, extended to $116M when Zalando joined in July 2026 — total funding ~$140M. The product is Cortex, a vision-language-action model that runs on third-party arms (UR/Franka/Fanuc/KUKA-class) rather than proprietary hardware, sold as a subscription. Named customers: BMW, PepsiCo, Zalando. Bricks & Bytes reporting and Latka’s unverified ~$6.4M 2025 revenue on 68 heads are the only public traction anchors. The open question is falsifiable: does Cortex actually generalize across SKUs and robot embodiments without customer-specific fine-tuning — landing new sites on subscription economics — or does every deployment slide into an integration engagement that caps unit economics and lets Physical Intelligence, Skild, or the Covariant stack now inside Amazon Robotics out-generalize the vertical foundation-model bet.

Valar Atomics — Energy · Emerging. El Segundo-based fission startup that closed a $1B Series B led by Sequoia (Shaun Maguire) on August 3, 2026 at a reported ~$6B valuation, on top of a $19M seed (Riot/AlleyCorp, February 2025), a $130M Series A (November 2025), and a $340M+$110M debt round at ~$2B (March 2026). The product is Ward 250 — a helium-cooled HTGR with TRISO fuel, starting at 100 kWt with a roadmap to ~5 MWe modules; achieved zero-power criticality on June 18, 2026 at Utah’s San Rafael Energy Lab under DOE’s Reactor Pilot Program, the second such criticality and the first outside a US national lab. The company is also pursuing a Philippines siting path at UP Diliman with PNRI (Meralco evaluating), and is suing the NRC over the “utilization facility” definition alongside Last Energy and state AGs. There is no signed hyperscaler LOI — the Nvidia relationship is a 30 MW feasibility study only, while Microsoft, Meta, Google, and Amazon have already inked with rivals. The answer condition lands when a Ward 250 sells a commercial kilowatt-hour, on either regulatory track.

Arrakis — Industrial AI · Emerging. London/Paris industrial-agent deployment platform founded in January 2026 by ex-Accel investor Rafael Quintanilla with three ex-Palantir Forward-Deployed Engineers (Haroun Beltaifa, Romain Fouilland, Mikhail Galkov). Emerged from stealth July 22, 2026 with $38M raised in ~3 months: a $7.5M seed (Accel, March 2026) and a $30M Series A led by Blossom Capital at a reported $140M post-money (July 2026), with individual checks from Olivier Pomel (Datadog), Olivier Godement (OpenAI), and Junaid Hussain. The pitch is model-agnostic industrial agents built with FDE labor, priced with a performance-fee component; the only concrete customer proof point publicly quoted is a 90% reduction in a procurement cycle at an unnamed NYSE-listed enterprise. The open question is whether that layer stays durable — or whether Palantir AIP (already at Airbus and BP), Databricks Agent Bricks, Cohere North, and Cognite (now Schneider-owned) absorb the industrial agent surface into their own SDKs within 24 months. The Palantir tribute-act critique is not adjacent skepticism — it is the base case if the wrapper thesis fails.

Indigo — Insurance · Emerging. Miami-based AI-driven medical-professional-liability insurer, founded 2023 by Jared Kaplan (ex-OppFi, Cadre, Insureon) and Matt Kim (Rubicon Founders), writing through Indigo Risk Retention Group (Charleston, SC; A- AM Best). $50M Series B closed January 29, 2026 (Rubicon Founders lead; Town Hall Ventures new; Optum Ventures and Oak HC/FT continuing) on top of the 2023 launch capital. Traction: ~1,000 providers, north of $10M premium, 20% of underwriting decisions handled AI-only by year-end 2025. Competitors are The Doctors Company ($1B+ direct written premium, physician-owned), MedPro (Berkshire), and ProAssurance (public) — a market where the incumbents underwrite by specialty code, board actions and manual chart review, and where the AI wedge is claims-tail data plus NPI-level severity modeling. The open question is binary and dated: does the loss experience actually hold up as 2023-2024 policies clear the two-to-four-year MPL claim tail, or does the AI-underwriting edge evaporate the way telematics-led Lemonade’s did in personal auto when real severity showed up in 2022-2023? The 2027-2028 loss triangles will decide it.

Full pages linked above. Yesterday’s scan: Scan #032.

Full deep dives