Kodiak AI

Phase A — Understand the business

Lens 1 · Company Overview

Kodiak builds and operates the Kodiak Driver — "a single-platform virtual driver that combines advanced AI software with modular hardware" — for long-haul trucking, industrial trucking, and U.S. defense. Three revenue streams: (i) DaaS (license the Driver onto customer-owned trucks, per-vehicle or per-mile); (ii) freight delivery on Kodiak-owned autonomous trucks (the legacy "PDP"/pilot model it intends to retire); (iii) ground-autonomy solutions to the U.S. military.

The business model thesis is asset-light recurring revenue: by putting the Driver on customer-owned/-operated fleets, Kodiak avoids owning the trucks and books per-truck or per-mile license fees under 3–4-year master service agreements ("MSAs") that also bundle the Kodiak OnTime integration suite (fleet/transport/yard management hooks) and a 24/7 operations center. It launched DaaS in December 2024 with its anchor customer Atlas.

• Founded: April 2018 by Don Burnette. HQ: Mountain View, CA; ops in Lancaster & Odessa, TX. Headcount: 342 as of 31 Dec 2025.
• Key customers: Atlas Energy Solutions (industrial/oilfield, the anchor); U.S. Army / DoW (defense, episodic); and newer freight/logistics names — Roehl Transport, Martin Brower, Loadsmith.
• Contract structure: per-truck or per-mile license; the one ≥1-year contract on the books is entirely the Atlas DaaS deal, with $28.8M of remaining performance obligations recognized ratably through Dec 2029 (up from $21.8M at YE2025).
• Concentration: the structural fact of this company — Atlas was 46% of FY25 revenue and 82% of Q1'26 revenue, and is the sole customer with driverless trucks deployed.

Read: This is a pre-commercial AV company with one paying driverless customer, a credible-but-narrow industrial wedge (oilfield private roads), and a much harder, much larger prize (public-highway long-haul) it has not yet unlocked.

Lens 2 · Supply Chain

Kodiak is the integrator/software node; it owns the Driver software and assembles a third-party hardware stack onto OEM trucks. The chain:

Upstream inputs → Kodiak → end customer:

• Compute: NVIDIA — in Mar 2026 Kodiak announced it is scaling the Driver on NVIDIA DRIVE Hyperion / Thor-class automotive compute. NVIDIA is the single most important upstream dependency (on-device inference).
• Sensors: LiDAR / radar / camera suite. The 10-K flags dependence on "a limited number of suppliers" for key sensing/compute components and exposure to tariffs/lead-times.
• Trucks / OEMs & upfitters: Kodiak is "vehicle-agnostic" and integrates onto Class 8 OEM trucks via third-party upfitters (the Driver is "modular and vehicle-agnostic hardware").
• Kodiak → customers who own the trucks: Atlas owns the trucks, owns the depots, and "dictates the product roadmap" in the oilfield deployment. This is the asset-light model's flip side — the customer holds the assets and the leverage.

Chokepoints / single-source dependencies:

• NVIDIA compute — concentration risk shared across the entire AV/robotics sector; no real substitute at the perception-stack performance tier today.
• Atlas as the demand chokepoint — not a supplier, but the binding constraint: one customer is ~all the driverless revenue.
• Data / safety-case validation — the binding internal bottleneck for the highway expansion (see Lens 3); proprietary on-road miles + simulation are the input that gates the "long-haul safety case."

Lens 3 · Competitive Advantages (moats)

What's genuinely defensible:

• First-mover in driverless-at-scale, no-observer commercial ops — Kodiak says the Permian Basin is "the first market in the U.S. where customer-owned fully autonomous Class 8 trucks (no safety observer onboard) are operating commercially at scale". ~23,500 cumulative paid driverless hours and 15,600 loads as of Q1'26 is real operating data — the moat candidate is the fleet/safety data flywheel (the robotics layer flags "fleet-scale teleop data" as a low-substitutability chokepoint ).
• Switching costs via Kodiak OnTime — once a customer rewires fleet/yard/TMS workflows around the Driver, ripping it out is costly. Management explicitly leans on this for retention.
• Founder/team pedigree — Burnette (Google self-driving → Otto → Uber) and an AV-veteran bench; genuine technical credibility in a field where talent is the scarce input (Lens 9).

Where the moat is thin / bargaining power weak:

• Bargaining power is inverted in the anchor deal. Atlas owns the trucks/depots and sets the roadmap; Kodiak is capped at the 100-truck initial order with no obligation for Atlas to buy beyond it. That is a customer with power over the supplier, not the reverse.
• The hard moat (the highway safety case) is not yet built. The industrial wedge runs largely on private roads; the defensible, high-TAM business is public-highway long-haul, which Kodiak has not launched (targeted 2H'26) — exactly the same race Aurora has a multi-quarter head start in.
• No durable cost/scale advantage yet — at 28 deployed trucks vs. Aurora targeting ~200 by year-end, Kodiak has neither the data scale nor the unit-cost curve that would make the flywheel self-reinforcing.

Lens 4 · Segments

segments.csv is empty (header only) — the company reports as a single operating segment (CODM uses consolidated net loss). The meaningful breakdown is by revenue type, all ``:

Geography: 100% United States.

Read: The headline "revenue down 75% in FY25" is real but mix-driven, not demand-driven — a one-time defense contract ($12.3M) lapsed while the strategic line (DaaS) is tiny but growing fast. The honest framing: revenue is de minimis and the company is being valued entirely on the forward DaaS ramp, not the trailing P&L.

---

Phase B — Measure performance

Lens 5 · Earnings Result (latest print: Q1 2026, period ended 31 Mar 2026, filed 8 May 2026)

Critical interpretation — the print is two stories fighting each other:

1. The "earnings beat / swing to profit" headline is an accounting illusion. The +$26.5M "net income" is driven by a +$64.7M non-cash gain on the fair-value remeasurement of common-stock warrants (the stock fell, so the warrant liability shrank). Ignore it. The real number is the $37.9M GAAP operating loss and $35.0M free-cash-flow burn — both roughly doubling YoY.
2. The operating story is a company spending aggressively into the 2H'26 long-haul launch — opex nearly doubled, R&D headcount +$5.1M, ops infra +$2.0M "to support DaaS operations and development related to our over-the-road launch".

Balance-sheet flags: Cash+securities fell $120.7M → $90.2M in one quarter (operating burn ~$29.5M + capex). Going-concern doubt is explicit and persists even after the $100M PIPE — management states the cash plus PIPE "will not be sufficient to fund operations for at least one year," funding the plan only "into the second quarter of 2027". Accumulated deficit $827.2M.

Market reaction: The stock fell ~37% on/after the print — but the trigger was not the operating loss; it was the simultaneously announced $100M PIPE at $6.50 vs. a $9.10 close (a steep discount) plus reports of an Atlas 100-truck delay. The market reacted to dilution + slippage, not the (illusory) EPS beat — a healthy sign that holders see through the warrant noise.

Lens 6 · Earnings Calls (sentiment trend)

transcripts/ is empty — no call transcripts on the shelf, so this lens is `` and necessarily thinner.

Kodiak has only been public for ~two quarters (Q3'25 stub + Q4'25 + Q1'26), so there is no multi-year tone arc yet. From management commentary in press and the MD&A:

• Recurring themes: "operations as the key to autonomous trucking success" (Burnette's framing — execution/ops discipline over demos); "capital-efficient asset-light DaaS"; "strong margins at scale"; the 2H'26 long-haul driverless launch as the gating catalyst.
• What they emphasize now vs. before: a clear pivot in language from "freight delivery" (Kodiak-owned trucks) toward "transition customers to DaaS" and diversification (logging, international, defense logistics) — an effort to de-risk the Atlas concentration narrative.
• Tone shift: more defensive post-Q1 — the PIPE and the Atlas-delay disclosure forced an acknowledgment of slippage and a heavier capital-need message. The "into Q2 2027" runway language is notably specific and sober for a growth-stage AV name.

Lens 7 · Comps

AV-trucking peers trade on TAM/optionality, not fundamentals — multiples are meaningless at near-zero revenue, so the honest table is operational + a market-cap/scale ranking, not a P/E grid.

The only comp that matters: KDK at ~$1.05B vs. AUR at ~$13–14B. The market is paying ~13x more for Aurora — and Aurora has ~7x the deployed trucks, a launched public-highway product, and a vastly stronger balance sheet. Kodiak is "cheaper" only because it is earlier, smaller, and capital-fragile.

Lens 8 · Stock-Price Catalysts

KDK has only traded since 25 Sep 2025, so the 5-year-pattern instruction collapses into a ~9-month tape. The moves that mattered:

Pattern: KDK trades as a high-beta story stock — it spikes on validation (Soros, partnerships, NVIDIA, the long-haul-launch narrative) and craters on dilution and milestone slippage. The market reacts to capital structure and execution dates, not to the (tiny, noisy) reported numbers. That makes the 2H'26 long-haul driverless launch the single binary catalyst for the next leg.

---

Phase C — Judge people & books

Lens 9 · Management

• CEO / Founder: Don Burnette. Genuine AV pioneer: Software Tech Lead on Google's self-driving project (Waymo precursor), 2010–2016; co-founded Otto (first self-driving-truck startup, acquired by Uber Aug 2016); Tech Lead at Uber 2016–2018; founded Kodiak April 2018. MS Robotics, Carnegie Mellon; physics/math/EE degrees, U. Florida. Track record: built and sold one AV company (Otto/Uber), then built Kodiak from zero to the first no-observer driverless commercial Class-8 deployment — a credible operator, not a promoter.
• CFO: Surajit Datta (Principal Financial & Accounting Officer; signed the Q1'26 10-Q). Limited public-company-finance history at Kodiak specifically; the 10-Q itself flags "management has limited experience operating a public company".
• Skin in the game / ownership: Founder retains a meaningful legacy stake (exact % n/a — not cleanly sourced; the 10-K Item 12 beneficial-ownership table on the shelf was not parsed to a number — verify on refresh). Notable outside holder: Soros Fund Management 5.7%; Ares Management affiliate (the SPAC sponsor) retains exposure and added $5M in the May'26 PIPE. Sponsor earnout: 6,250,000 shares subject to vesting.
• Capital-allocation history: This is a pre-revenue burner — "capital allocation" is really financing and runway management, and the record here is mixed-to-poor by necessity: a cash-light de-SPAC (only ~$26.2M net from the trust after redemptions), funded instead by $145M Series A preferred + $43.9M second-lien + $29.7M debt + $23.7M SAFEs in FY25, then a dilutive $100M PIPE at a 29% discount within 8 months of listing. ROE/ROIC are deeply negative and not meaningful.
• Red flags (governance/comp): (i) a company-officer departure in FY25 triggered $3.0M of SBC modification expense; (ii) heavy warrant/preferred overhang with anti-dilution reset mechanics that transfer value from common holders; (iii) classic SPAC sponsor dynamics (Ares earnout, sponsor influence). Nothing fraudulent — but the capital structure is shareholder-unfriendly by construction.
• Archetype: Founder-led technical visionary, appropriate for the R&D/validation stage — but paired with a thin public-company finance function facing a balance sheet that demands world-class capital-markets execution. That is the tension.

Lens 10 · Forensic Red Flags

Acting as a forensic analyst. The accounting itself is clean but noisy — the risk is not fraud, it is balance-sheet fragility and SPAC-remeasurement obscuring the real burn.

• Going concern (the headline): Both auditors (Deloitte for FY25/24, EY for FY23) issued explicit going-concern paragraphs, and management reaffirms substantial doubt in the Q1'26 10-Q even after the $100M PIPE — funded only "into Q2 2027". This is the single most important forensic fact.
• Earnings ≠ cash, in both directions: FY25 "net loss" of $585.5M was ~85% non-cash SPAC/SAFE remeasurement (a $210.7M loss on equity issuance, $190.1M SAFE mark, $35.0M warrant mark, $24.4M second-lien mark). Q1'26 "net income" of $26.5M is the reverse — a $64.7M non-cash warrant gain. The only honest profit metric is operating loss / FCF burn (FY25 FCF $(116.5)M; Q1'26 FCF $(35.0)M). Both reported bottom lines are analytically useless.
• Revenue recognition: low risk — revenue is tiny ($3.8M FY25), recognized at a point in time for freight/defense and ratably for DaaS; RPO $28.8M disclosed and bounded; no unbilled-receivable or contract-asset games (none material). Receivables are negligible ($0.9M).
• SBC: rising fast ($5.6M FY24 → $19.1M FY25, incl. the $3.0M departure modification) — flatters nothing here because Kodiak is GAAP-loss-making anyway, but watch the non-GAAP "loss from operations" reconciliation, which strips SBC ($93.5M non-GAAP vs. $112.6M GAAP FY25).
• Capital-structure complexity (the real hazard): Series A preferred $223.2M, common-stock-warrant liability $93.7M (was $158.3M), second-lien loans $11.3M, plus PIPE warrants with anti-dilution resets to $6.00. Dilution risk is structural and ongoing.
• Debt/leases: modest (~$31M total debt + small operating leases through 2031); not the binding risk vs. the equity overhang.

Regulatory findings (required sub-section):

• SEC Litigation Releases / AAERs: None. "total_sec_findings: 0" — no LR or AAER names Kodiak AI across 2021-06-30 → 2026-06-30.
• 10-K Item 3 (Legal Proceedings): the company's own disclosure — "We are not currently a party to any such claims, lawsuits or proceedings... likely to have a material adverse effect".
• Non-SEC enforcement (FTC/DOJ/FDA/NHTSA): web search surfaced no material enforcement action, consent decree, fine, or penalty against Kodiak. The relevant forward regulatory exposure is NHTSA / state-by-state AV rules (TREAD Act reporting duties, first-responder protocols, insurance standards) — a compliance-complexity and accident-liability risk, not a current enforcement matter.
• Conclusion: No material regulatory or legal findings — verified via SEC EDGAR EFTS (LR, AAER), web search, and 10-K Item 3 as of 2026-06-30.

---

Phase D — Project & stress-test

Lens 11 · Forward Projection

Kodiak is GAAP-loss-making with de-minimis revenue, so an EPS line is not the right primary metric — the analytically honest projection is revenue ramp + cash runway to the next value-inflection (the 2H'26 long-haul launch), with EPS modeled only directionally. All ``, inputs labeled.

Base inputs (anchored to actuals): FY25 revenue $3.8M; Q1'26 $1.83M (annualizing ~$7M+ but mix is shifting to DaaS). Atlas fleet 28 trucks → 100-truck initial order (delayed). Opex run-rate ~$40M/qtr and rising. Liquidity $90.2M + $100M PIPE ≈ $190M gross, runway "into Q2 2027".

Runway / capital path (the real model): at ~$35M/qtr FCF burn, $190M gross funds ~5–6 quarters → consistent with management's "into Q2 2027". A further raise is required within ~12 months regardless of operational success — and if the highway launch slips again, it raises into a weaker tape (more dilution). This is the dominant valuation driver, above any revenue point estimate.

Brier forecast — NOT logged (per --watchlist rule: skip forecast.ts create in the breadth loop). The natural binary to log on a future committed pass: "KDK launches driverless (no-observer) long-haul highway operations by 2026-12-31" — base-case probability ~45%, given the explicit 2H'26 guide but a demonstrated pattern of milestone slippage (the Atlas delay).

Lens 12 · Bull vs Bear

Bull case. Kodiak is the only pure-play AV-trucking name besides Aurora that has real, no-safety-observer driverless trucks earning revenue today (28 Atlas trucks, 23.5k paid hours, +120% QoQ hours). The DaaS model is genuinely asset-light and high-margin at scale — Kodiak develops the brain, the customer owns the metal. A founder with an Otto/Uber/Waymo pedigree, an NVIDIA-Hyperion compute path, Soros and Ares on the cap table, and a 2H'26 long-haul launch that — if it lands — re-rates the equity toward Aurora's ~$13B comp. At a ~$1.05B cap, the upside is asymmetric if the highway works: this is a call option on U.S. driverless freight with a live, paying industrial deployment as the strike's intrinsic value. Sell-side consensus PT ~$12.60 (~2.4x) reflects that optionality.

Bear case (2–3 permanent-impairment risks). (1) Capital-structure death spiral: going-concern doubt persists post-PIPE; the company must raise again by ~mid-2027, each raise is dilutive (the last was −29% to market and reset warrants to $6), and a thin float + warrant overhang amplifies every down-move — equity holders can be ground to dust even if the technology succeeds. (2) Single-customer cliff: 82% of revenue is one customer whose 100-truck order is capped and already delayed; if Atlas defers/walks, revenue is structurally near-zero. (3) Aurora runs away with it: the leader has ~7x the trucks, a launched highway product, and a fortress balance sheet — in a capital-intensive, data-flywheel race, the #2 with the weakest balance sheet is the most likely to be starved out.

Pre-mortem (18 months out, thesis broke): It's December 2027. The long-haul launch slipped twice; Atlas capped at ~60 trucks and didn't re-order; Kodiak burned through the PIPE and raised again in Q2'27 at ~$3 with fresh warrants; the float doubled; Aurora signed the Tier-1 carriers Kodiak was courting. The stock is sub-$2 and the story is "right technology, wrong balance sheet."

Are multiples too high? On any trailing metric, absurdly (≈276x revenue) — but that's true of the whole sector and is the wrong lens. The right question is option-value vs. dilution-and-survival risk. At ~$1.05B for a going-concern with one customer, the market is already pricing meaningful execution; it is not a cheap option.

Contrarian view (what the market refuses to see): The bull consensus treats the $100M PIPE as "problem solved." It is not — management itself says it doesn't reach a year of runway. The market is also under-weighting that Kodiak's current revenue runs on private oilfield roads, the easiest possible ODD; the public-highway business everyone is paying for is the unproven part. Conversely, the bears under-weight that no-observer commercial driverless at scale is genuinely hard and Kodiak has actually done it — that operating data is not nothing, and a single Tier-1 long-haul win in 2H'26 would invalidate the "distant also-ran" framing overnight.

Lens 13 · Devil's Advocate (short-seller)

Dismantling the bull case.

• What structurally breaks the model: revenue is 82% one customer on a capped, delayed order. There is no diversified book to fall back on — "diversification" (logging, international, defense) is announcements, not revenue.
• Why the moat is weaker than bulls think: the deployed business is on private roads (the trivial ODD); the hard, defensible highway business is unlaunched; and the customer (Atlas owns the trucks and sets the roadmap) holds the bargaining power. The "data flywheel" has ~23k hours — Aurora and Waymo have orders of magnitude more relevant highway data.
• The most dangerous competitor bulls underestimate: not Aurora (everyone sees Aurora) — it's the OEM-backed players (Torc/Daimler) and the AI-first Waabi, who can bundle the Driver into truck sales or out-engineer on a leaner model, eroding the standalone-software value proposition.
• Worst capital-allocation / incentive issues: a cash-light de-SPAC (heavy redemptions) papered over with $240M+ of preferred/second-lien/SAFEs and then a 29%-discount PIPE with warrant resets to $6 — a structure that systematically transfers value from common to insiders/financiers. The SPAC sponsor (Ares) earnout adds to the overhang.
• What must hold for today's price: that the highway launch happens ~on time, Atlas re-orders beyond 100, and the next raise is small and non-punitive. All three are required; the first two already wobbled in Q1.
• If growth disappoints 20–30%: at this revenue scale a 20–30% miss is rounding error on the P&L but fatal to the narrative — the multiple is 100% story, so a missed highway date is a multiple-compression event, not an earnings event.
• The single permanent-impairment scenario, and plausibility: a forced, deeply dilutive down-round in 2027 into a slipped-launch tape — highly plausible given management's own "into Q2 2027" runway admission and the demonstrated milestone slippage. This is the base-rate outcome for cash-fragile de-SPAC AV names, not a tail.

Lens 14 · Management Questions (ordered by information value)

1. Atlas is 82% of revenue and the 100-truck order is delayed — what exactly slipped, what is the revised truck-deployment schedule, and what has Atlas committed to (if anything) beyond the initial 100?
2. Your own 10-Q says cash + the $100M PIPE funds you only "into Q2 2027" — what is the specific plan and expected size/structure of the next raise, and what triggers it?
3. The 2H'26 driverless long-haul highway launch is the entire thesis — what are the precise remaining gating items on the safety case, and what is your confidence interval on the date?
4. On public highways you face Aurora's multi-quarter lead and far larger data scale — what is your concrete, customer-provable advantage over Aurora, in one sentence a fleet CEO would act on?
5. What is the actual per-truck or per-mile DaaS economics today (price, gross margin, payback for the customer) — not the "strong margins at scale" abstraction?
6. Atlas owns the trucks, depots, and roadmap — how do you prevent the customer's bargaining power from compressing your take-rate as you scale?
7. Which Tier-1 long-haul carrier conversations are furthest along, and what would convert a pilot to a multi-year MSA?
8. The warrant/preferred anti-dilution resets to $6.00 keep transferring value from common holders — how do you protect common shareholders in the next financing?
9. You diluted at a 29% discount in May — why was an equity PIPE the right instrument versus debt or a strategic/OEM partnership, and would you do it again?
10. What is the path off going-concern — at what revenue/burn point do the auditors remove the qualification?
11. Defense/Army revenue collapsed from 89% to ~26% of mix — is defense a real growth vector or a legacy line, and is the episodic revenue worth the engineering distraction?
12. How dependent is the roadmap on NVIDIA DRIVE Hyperion/Thor, and what is your second source if supply, cost, or roadmap timing slips?
13. The CFO/finance function is new to public-company scale facing a complex cap table — what is the plan to build out capital-markets and FP&A capability?
14. What does the unit-economics model look like at 1,000 trucks — the bridge from today's $35M/qtr burn to operating cash breakeven?
15. International and logging were just announced — are these revenue in 2026 or optionality, and do they dilute focus from the highway launch?

---