Startup Geographies Are Being Redrawn
The default mental map of "where startups happen" -- SF, NYC, maybe Boston -- is breaking. Capital, founders, and policy attention are clustering around specific theses in places that didn't exist on the map five years ago, and a parallel layer of access-constrained foreign geographies is functioning as a live testbed for the kind of hardware and infrastructure that can't iterate inside the US regulatory perimeter. The Mythos / a16z 5.6 noise is downstream of this: tech became political infrastructure, and the fight is now over who gets to build it.
The cracks in the duopoly
SF and NYC aren't dying -- SF is still where foundation-model capital sits and NYC still owns fintech, media, and consumer. What's changing is the marginal dollar. The thesis-specific hubs are pulling the next layer of company formation because they offer something the Bay can't: cheap floor space next to a runway, a port, a fab, a range, or a Tier-1 customer. Add export controls, defense spending, reshoring incentives, and an aggressive shift of federal procurement toward dual-use, and the map fractures along thesis lines, not lifestyle lines.
The pattern: hard things need real estate, energy, and proximity to a customer with a budget. The new hubs are the places where all three exist.
The new US map
El Segundo ("the Gundo") -- hard tech and defense. The densest cluster of new American defense and aerospace companies. Anchors: Anduril, Hadrian, Apex (now $2.3B post a $200M Series D in Sept 2025). Emerging: Wardstone (YC, space-based interceptors for missile defense), CX2 ($31M Series A, June 2025, electromagnetic warfare), Picogrid (25K sqft expansion + $1.1M Army contract for distributed C2), Space Kinetic (Whirlwind OS for on-orbit ops), Icarus (stratospheric comms), ABL Space Systems (small-sat launch). The cluster is now self-reinforcing: ex-Anduril, ex-SpaceX, ex-Hadrian engineers spin out within walking distance of their old employers.
Austin -- defense, AI, and capital migration. Austin VC funding hit an all-time high in 2025 (Crunchbase). Less about a single thesis and more about a tax/regulatory environment plus Palantir, Oracle, and Tesla gravity. Emerging on the defense-tech side: a different shape than El Segundo -- intelligence, cyber, and cognition rather than airframes. Companies like Anduril's Austin software org, Saronic (autonomous surface vessels, $600M Series C 2025), Saildrone's Austin presence, and a long tail of AI-for-defense and cyber primes-of-tomorrow.
West Texas / Permian / Starbase -- energy + compute + launch. The most underrated map move of 2025. The Permian has the cheapest stranded power in North America, and that has become a wedge for AI compute. Fermi America is building a ~1GW AI campus with Texas Tech. FO Permian is building 150MW of off-grid behind-the-meter power for hyperscalers. Sangha Renewables is monetizing renewables via bitcoin mining (Ector County, 20MW project energized Dec 2025 with TotalEnergies). Ambrosia Energy (ex-SpaceX founders) is building solar+battery plants targeted at AI hyperscalers. Starbase itself was incorporated as a city in 2025, which gives SpaceX zoning and permitting control of the launch corridor.
Arlington / Northern Virginia -- Beltway defense software. The intelligence-community-adjacent equivalent of El Segundo. Rune Technologies ($24M, battlefield logistics software), Quartermaster ($43M Series A May 2026, First Round / Quiet, AI-driven contested logistics), Tercero (edge AI for electromagnetic spectrum operations), SwarmInt (tactical autonomy), AgentForge / Hypha (agentic AI for DoW and IC), Nooks ($25M Series A, Classified-Infrastructure-as-a-Service -- turn-key SCIFs for software companies). Starburst's National Landing Launchpad and the National Innovation Quarter (launched Feb 2026) are now formalizing what was already happening.
Detroit -- robotics, reshoring, and physical AI. The robotics map's quiet winner. Atomic Industries ($25M Series A, AI-driven injection mold automation), Inbolt (vision-enabled adaptive robot programming, opened US HQ in Detroit), Intermode (mobile robotics infrastructure for OEMs), Borg Robotics (humanoid for assembly), Motmot (water-main inspection robots), plus Newlab Detroit as the soft-landing space. Michigan put up roughly $900M in state-level robotics and mobility incentives over 2024-2025, which is showing up as company formation, not just relocation.
Pittsburgh -- the CMU robotics export engine. Robotics Factory at Hazelwood Green / TechForge is producing a continuous pipeline. HEBI Robotics (modular actuators, NASA / Army contracts), FuturHand (dexterous manipulation), Whelix / Mito Robotics (cell culture automation), roboLoop (e-waste robotic dismantling). The Pittsburgh Robotics Network's 2025 Discovery Day had ~32 startup exhibitors -- a useful proxy for the depth of the bench.
Columbus / Central Ohio -- manufacturing capital. Intel's $20B Ohio One fab project plus Anduril's Arsenal-1 factory near Rickenbacker (production targeted July 2026, ~5M sqft autonomous-weapons factory). Columbus is positioning as the "make it" half of the new defense economy where El Segundo is the "design it" half.
Miami -- fintech, crypto, and Latam bridge. Less buzzy than 2021-2022 but the substrate is still there. Florida startups raised $5.83B across 575 deals in 2025 (eMerge Americas 2025 Annual Insights); fintech led with $1.22B statewide. Miami is now the natural US base for Latam-facing fintech and crypto infra, plus a growing pocket of defense-tech (SOUTHCOM is here) and healthtech.
Bentonville -- supply chain and CPG. Arkade (Spring Capital's tech "greenhouse" at Ledger, downtown Bentonville) is actively recruiting retail-supply-chain startups to physically locate near Walmart, J.B. Hunt, and Tyson HQs. If your customer is the largest retailer in the world, "be in Bentonville" is now a strategic decision, not a punishment.
Foreign testbeds: where the iteration is faster than the regulation
Some of the most important hardware iteration in 2025 isn't happening in the US. It's happening in places where the cost of failure is low and the cost of not iterating is existential.
Ukraine. Brave1, the state-run defense innovation cluster (founded 2023), launched the Test in Ukraine program in July 2025: foreign defense companies can test their hardware under real combat conditions. 45 international companies applied in the first window; Diehl was the first foreign UGV trial. Brave1 ran formal test cycles in 2025 for FPV bombers (14 systems, 10-18 inch frames), fiber-optic drones (20+ km range, 15+ manufacturers), and kamikaze drones (40+ km range). Ukraine is now producing on the order of millions of drones annually with a feedback loop measured in days. Anything that needs to learn to fly in a contested EW environment learns there or doesn't learn at all.
Israel. Still the densest defense-tech ecosystem per capita. Regulus (autonomous counter-UAS), plus a long tail of secretive startups raising large wartime rounds (one recent $60M raise, reported by Calcalist). Israel is where electronic warfare, sensors, and autonomy stacks get hardened before they show up in US programs of record.
Estonia. Not a defense story -- a digital state story. e-Residency, X-Road, and the fact that a startup can incorporate, file taxes, sign contracts, and operate entirely in software make Estonia the cleanest sandbox in the world for testing what a fully-digital state-corporate interface looks like. Skype, Wise, and Bolt all came out of this substrate, and the next wave (digital identity, AI-government tooling, sovereign cloud) is taking the same shape.
The pattern across all three: a state apparatus that has decided it would rather host the iteration than block it, because the alternative is irrelevance.
Section 4 -- Tech is now political infrastructure, and citizens are picking sides
The Mythos / a16z 5.6 noise is being read as VC drama. It isn't. It's the surface of a deeper shift: the tools of governance -- communication, persuasion, coordination, defense, identity -- are now software, and ordinary citizens (not just enterprises) are the end users. That makes startup geography a political question whether founders want it to be or not.
A few things are happening at once:
- a16z's "Little Tech Agenda" (July 2024) explicitly positioned startups as a political constituency, not an apolitical industry. They followed it with active 2024 election spending and PAC infrastructure. The Vanderbilt Policy Accelerator published A Little Tech Policy Agenda (Feb 2026) translating that into legislative form.
- The political response runs both directions. State AGs in California, New York, and Texas are now writing AI policy that explicitly treats which state you're incorporated in as a policy lever -- not a neutral choice.
- Citizens, not just founders, now have access to the production stack. What used to require a TV station, a PAC, and a media buyer now requires a phone and three subscriptions. Spencer Pratt's one-person political-content operation after the Palisades fire (tens of millions of views) is the consumer-side proof. The RNC's all-AI Biden ad, the DeSantis Trump-Fauci images, and the FCC-fined AI Biden robocalls are the political-operative-side proof.
The investable consequence: citizen-facing political infrastructure is its own category now. Identity / provenance (post-deepfake), small-donor and small-operator coordination tools, AI-native canvassing and ad ops, civic-grade messaging, and the security layer that defends all of it.
Investable pattern
Three layers across both the US hubs and the foreign testbeds:
- Picks and shovels per hub. Energy in the Permian, fabs and clean rooms in Columbus, ranges and test ranges in El Segundo and West Texas, SCIFs and classified compute in Arlington, autonomy data infrastructure in Pittsburgh and Detroit. These are the boring, capital-intensive enablers that don't appear in the press release.
- Founder factories tied to a hub's thesis. Anduril, SpaceX, Palantir, Hadrian, and Boston Dynamics are all now producing alumni clusters that re-form into new companies inside the same geography. The next 50 defense-tech companies look more like 1971-era Fairchild → Intel than 2015-era SF.
- The bridges. US companies that can run hardware iteration through Ukraine or Israel and ship into US programs of record. Estonia-incorporated software businesses with US distribution. Founder-residency arbitrage products. The bridges are where most of the asymmetric returns probably live.
What we're watching
- Which El Segundo company is the first to cross $5B without a defense exit
- Whether Anduril Arsenal-1 ships on its July 2026 production target -- and what that does to Columbus's gravitational pull on the next wave of factory-scale defense companies
- The Permian's first "AI-only" 1GW campus going live
- How many Test-in-Ukraine cohorts US companies join in 2026, and which ones leverage it into Pentagon contracts
- The 2026 election cycle as the first one where citizen-operated AI content tooling is the dominant ad-spend substrate