Kodiak AI, and the Insurance Nobody Is Pricing

We initiate on Kodiak AI with a Speculative framework stance, and we make the note deliberately two part. Kodiak is the first company running fully driverless commercial trucks for paying customers, which is a real and rare distinction, but it is a sub scale, cash constrained, single customer dependent business carrying a going concern warning, and the equity is a binary option on scaling before the cash runs out and before a far larger competitor takes the category. The more durable idea in this note is not the equity at all. It is our house conviction that driverless freight is inevitable, and that the least watched way to express it is the commercial trucking insurance complex, a small and concentrated market whose core risk the technology is about to rewrite.

The bull case on Kodiak is concrete. It has crossed 40,000 cumulative paid driverless hours and delivered more than 20,000 loads, revenue rose 91% quarter over quarter and 595% year over year, and it reaches revenue earlier than most peers by starting off highway in the Permian Basin instead of fighting for the public highway first. The bear case is equally concrete and sits in Section 03. Cash of $151M against a half year burn of $64M is roughly fourteen months of runway, the auditors have flagged going concern, one customer is 65% of revenue, and Aurora is an order of magnitude larger by market value and already running driverless on public highways in Texas. This is a venture stage risk profile inside a public listing.

The insurance thesis, developed from Section 08, is where we would spend research time. Commercial auto liability is a $43B line that has been unprofitable for most of a decade, it is controlled by only five parent balance sheets, and carriers are exiting it because nuclear verdicts and social inflation have made the human driver the most expensive risk in freight. Driverless technology removes that risk and replaces it with a different one, product liability on the manufacturer. That is a regime change in a market almost no one is modeling, and it is the kind of concentrated, mispriced, second order setup that rewards early work.

Kodiak sells autonomy as a service. Under its Driver as a Service model, customers pay a per mile or per vehicle fee to run the Kodiak Driver, the company's self driving software and modular hardware, without buying the technology outright. The wedge that sets Kodiak apart is where it started. Instead of leading with public highway long haul, where regulation and edge cases are hardest, it went driverless first in the Permian Basin oilfields with Atlas Energy Solutions, on private and low speed roads, where trucks now run day and night. That choice is why Kodiak has driverless revenue today while much of the field is still testing with safety drivers.

The customer roster is stronger than the revenue suggests, which is the argument for patience. Beyond Atlas Energy, Kodiak names Maersk, IKEA, J.B. Hunt, Bridgestone and C.R. England, with Roush building the trucks. The near term catalyst is the move from the oilfield to public highway long haul, which the company has said is close. If that transition works, the addressable market widens from a niche industrial use case to the core of the freight economy. If it slips, the company is a single application business burning cash.

This is the section that governs the equity. Kodiak held $151.1M in cash and securities at 30 June 2026 against a half year operating burn of $63.6M, and the filing carries explicit substantial doubt about the ability to continue as a going concern for at least one year without additional financing.

The arithmetic is unforgiving. A $64M half year burn implies roughly $127M a year, and against $151M of liquidity that is on the order of fourteen months of runway before a raise. The company already returned to the market in May 2026 for $100M of common stock at $6.50 a share with attached warrants, which is well above the current quotation, so the next raise, if it comes at today's level, would be materially more dilutive against a 200.6M share count. The accumulated deficit stands at $814.7M. Revenue is real and accelerating but tiny, and 65% of it comes from a single customer, so the model is not yet self sustaining and will not be for years. We treat further dilution as the base case, not a risk, and we size the equity accordingly.

The ownership picture is the most surprising part of the story and, for a contrarian, the most interesting. This is a going concern micro capitalization that has fallen roughly two thirds, and institutions are buying it almost without exception.

Institutional ownership sits near 73% of shares. More telling than the level is the direction. Over the last twelve months thirty three institutions added a combined $108M of stock, against a single seller who trimmed less than $100k. That is close to unanimous accumulation into price weakness, and the buyers are not only index funds. Among the funds that file quarterly, ARK is the most aggressive buyer, up 190% quarter over quarter, and the equity hedge fund Alyeska and the macro fund Rokos Capital both hold multi million share positions. The largest single disclosed stake belongs to George Soros's fund, at 9.77M shares or 4.9%, but that figure carries a caveat we take seriously. It comes from a 13G amendment that sits just below the 5% threshold, and a drop below 5% can reflect selling or simply dilution from Kodiak's rising share count, so we do not read the Soros position as fresh buying. The accumulation signal rests on the breadth of quarterly filers and on ARK, not on Soros. Short interest is modest by contrast, 8.91M shares, or 6.87% of float, at 9.3 days to cover, up only modestly month over month. The weakness here is not a crowded short. The overhang is the drift lower since the listing and the going concern question, not an active short campaign, which is a very different setup from a squeeze candidate.

Selected institutional holders, most recent filings

We would not over read it. These are early, thematic positions in a speculative name, sized small relative to the funds that hold them, and they can reverse as fast as they were built. The offsetting structural point is supply: as a company taken public through a SPAC, Kodiak has large restricted holdings in the hands of the founder, the Ares sponsor and the venture backers, and the unwinding of those lock ups adds share supply over time. Still, the base rate is that thematic and macro money does not accumulate a going concern micro cap this uniformly unless it believes the option is worth owning. That belief, set against the financing risk in Section 03, is the whole debate in one name.

The board was built for a public listing, and it deliberately pairs autonomy operators with financial and mobility credibility, anchored by the SPAC sponsor, Ares Management.

Board and sponsor, selected

The network

The cap table and the board tie Kodiak into the center of the autonomy and mobility world. The founder came out of Google's self driving program and Otto, one director ran Cruise, another is the president of Lyft, and Ken Goldman is one of the most connected chief financial officers in technology. The venture backers include Lightspeed, Battery Ventures and CRV, with strategic capital from BMW i Ventures and Bridgestone and from Atlas Energy Solutions, which is both the anchor customer and an investor. This is a well networked company. It is also a reminder that autonomy is a small world where the same operators recur, and that two of the most relevant reference points on this board, Otto and Cruise, are cautionary as much as they are credentials.

Political exposure

We screened director federal contributions in FEC records. Kristin Sverchek is a documented Democratic donor, to Harris for President, the Harris Victory Fund, the Democratic National Committee and a California House Democrat. Mohamed Elshenawy shows no itemized federal giving. The founder Don Burnette and Ken Goldman are common enough names that we could not isolate them with confidence, and the Kodiak employer records in the database resolve to a different Southern California entity, not to this company, so we do not attribute them. No director we could isolate has given to a Trump aligned committee. As with most Bay Area technology boards, the documented giving leans Democratic.

Kodiak is a first mover on driverless revenue, but it is not the scaled leader on capital. Aurora is the larger and better funded public pure play, running driverless freight on public highways between Texas cities with truck maker partners in Paccar and Volvo. Kodiak competes by going where Aurora is not yet dominant, off highway and industrial, and by keeping a lighter capital model through Roush instead of owning manufacturing.

Autonomous trucking, selected developers

The competitive risk is not that Kodiak has bad technology. It is that autonomy is capital intensive and the prize is winner take most on each lane, and Kodiak is the least capitalized credible entrant with a going concern flag while its main rival has scale and OEM distribution. The off highway wedge buys time and revenue, but the long term equity value depends on winning public highway lanes against much deeper balance sheets.

This section is our stated conviction. It is not a neutral survey. We believe that autonomous technology in commercial trucking is not a question of whether, but of when and by whom, and that the direction is effectively settled even if the timeline is not.

The case rests on structure, not enthusiasm. Freight is short of drivers and the workforce is aging, and a human driver is legally capped near eleven hours of driving a day under hours of service rules, while a driverless truck can run close to twenty four hours, which roughly doubles the revenue a single asset produces. The driver is also the largest single line of operating cost and, as Section 08 shows, the largest single source of insured loss. Remove fatigue, distraction and the human error behind the great majority of severe crashes, and the economics and the safety case point the same way. When a technology lowers cost, raises asset utilization and reduces the worst losses at once, adoption is a matter of engineering and policy timing, not of demand.

We are equally clear about the bottlenecks, because they set the timeline and they are where the money is made or lost. Regulation is a state by state patchwork with no complete federal framework, safety validation across weather and edge cases is unfinished, the capital intensity is punishing, public acceptance is unproven, and, critically, the insurance and liability regime for driverless freight barely exists yet. That last bottleneck is usually treated as a footnote. We treat it as the thesis, because a technology that cannot be insured cannot scale, and the market that solves that problem is being built right now with almost no public attention.

Here is the part of the note we care most about. Commercial trucking insurance is a small, concentrated, chronically unprofitable market that almost no equity analyst covers as a theme, and it sits directly in the path of driverless freight. If autonomy scales, it does not just change the trucking companies. It rewrites the risk that the entire insurance line exists to price.

The market in figures

Very few markets actually write this risk

The important structural fact, and the one that supports your instinct, is that there are very few real underwriters of trucking risk. The dozens of brand names a fleet sees are subsidiaries of a handful of parents. Five balance sheets dominate: Travelers through Northland, the largest trucking wholesale insurer in the country; Old Republic through Great West Casualty, a trucking specialist since 1985; Berkshire Hathaway through National Indemnity, biBERK and Guard; Markel through State National, the largest fronting platform; and Chubb through ACE American. The top forty writers hold roughly 80% of the line. This is a thin, specialist market, and it has been getting thinner as carriers flee.

It is a hard market because the human driver is the loss

Commercial auto has been one of the worst lines in property and casualty for most of a decade. Nuclear verdicts, defined as jury awards above ten million dollars, keep rising, third party litigation funding has industrialized the plaintiff bar, and severe bodily injury claims from large truck crashes drive the severity. Nationwide exited primary commercial auto in 2023 over the combined ratio. Rates keep climbing even in years when crash frequency falls, because the cost per claim keeps setting records. The line is hard precisely because the insured event, a human making a mistake at speed in an eighty thousand pound vehicle, is so expensive when it goes wrong. That is the exact risk driverless technology is built to remove.

When the driver is software, the question of who pays for a crash moves from the trucking company's commercial auto policy to the manufacturer's product liability. That single shift is the whole opportunity, and it cuts in three directions at once.

Near term, premiums rise and the few carriers who will write it gain pricing power

In the transition, insurers face what one professional in the field called lots and lots of unknowns: no actuarial history for driverless fleets, unresolved questions of who is liable when a component fails, and new cyber exposure. Expensive, complex technology with no loss data is underwritten at a premium if it is underwritten at all. Aurora's own annual report warns that coverage may be insufficient or denied. In the near term this favors the small set of carriers with the appetite and balance sheet to price novel risk, on top of an already hard market. The bottleneck is real enough that a lack of insurance could delay deployment regardless of whether the technology is ready.

Long term, the severity that made the line unprofitable can collapse

If driverless trucks are genuinely safer, the nuclear verdict driven severity that has crushed commercial auto returns falls away. That is good for society and ambiguous for incumbents. The legacy commercial auto premium pool tied to human drivers can shrink, a slow headwind for the specialists most exposed to trucking, Old Republic through Great West and Travelers through Northland. At the same time a new pool is created, product liability, technology errors and omissions, cyber, and fleet level autonomous coverage, priced on data rather than on driver records.

The data moat decides who captures the new pool

Autonomous developers generate continuous safety and telematics data of a kind human fleets never produced. Whoever holds that data prices the new risk best. That advantages the incumbents built on data and unusual risk appetite, Progressive in commercial auto and telematics and Berkshire through National Indemnity, and it advantages the developers themselves, who may self insure through captives and keep the liability pool inside their own economics rather than handing it to a carrier. The open question, and it is the central one for this thesis, is whether the new pool accrues to public insurers, to the AV developers, or to a new class of specialist that does not exist yet.

This is the actionable output. If driverless freight is inevitable, the insurance value chain is the underfollowed way to hold the theme, and it splits into incumbents facing a regime change, data holders positioned to price the new risk, and a venture stage greenfield that is mostly still private.

Contra watch list, the trucking insurance complex

Our preferred framing is a barbell. On one end, the data rich incumbents that can price the new risk and survive the transition, Progressive and Berkshire. On the other, the private insurtech and program specialists building autonomous coverage from scratch, reached through venture exposure or through the fronting carriers that host them. In the middle, the pure trucking specialists, Old Republic and Travelers, are a paired trade rather than a simple long: a near term beneficiary of the hard market and the premium rise, and a long term holder of the exact book that driverless technology erodes. The single most valuable piece of work from here is to size how fast the legacy pool shrinks against how fast the new pool forms, because the answer determines whether this is a disruption to short or a greenfield to own.

Open workstreams, ordered by how much they move the thesis.

1. 01

   Size the pool shift. Model how fast the legacy commercial auto trucking premium shrinks against how fast the new autonomous liability pool forms. This answers whether to short the disruption or own the greenfield.

2. 02

   Map insurer trucking exposure. Quantify what share of premium and earnings each incumbent, Old Republic, Travelers, W. R. Berkley, RLI, actually derives from trucking, to find the most and least exposed.

3. 03

   Trace the captive question. Determine whether Kodiak, Aurora and peers intend to self insure, since that decides who captures the liability pool.

4. 04

   Scout the private insurtech. Build a list of autonomous insurance startups and MGAs and their backers, as the venture stage expression of the theme.

5. 05

   Kodiak long haul milestone. Track the public highway driverless launch and the next financing, the two events that decide the equity.

6. 06

   Nuclear verdict trajectory. Follow whether tort reform or continued social inflation changes the urgency of removing the human driver.

• Kodiak completes SPAC combination
• Old Republic on emerging trucking risks
• Kodiak Q2 2026 10-Q, going concern
• State of the commercial transportation insurance market
• KDK price, market cap, financials
• Kodiak board appointments, Goldman and Sverchek
• A handful of parents control trucking insurance
• KDK institutional ownership and flows
• Milliman 2024 commercial auto liability results
• Soros Fund Management 13G, KDK stake
• Nuclear verdicts and the motor carrier insurance crisis
• ARK adds to Kodiak position
• Fleets and insurers face the autonomous transition
• FEC individual contributions, director screen
• Kodiak and Paccar on autonomous liability
• Kodiak AI investor releases
• Aurora on autonomous risk and insurance cost