Phase Space AI

Valuation

Kodiak Gas Services [KGS]

Kodiak Gas Services [KGS] — Valuation

As of 2026-07-30. Spot $58.90. Two horizons, two instruments (per references/valuation.md).

---

COMPANY STATE — declared first

STATE B — cyclical / commodity-sensitive.

Test applied and evidence:

• Exogenous driver. Contract compression demand is a derived demand from Permian and Eagle Ford natural-gas volumes, which are themselves associated gas driven by the oil price. Kodiak's own 10-K risk factors lead with "A reduction in the demand for natural gas and oil". This is the definition of a State B driver.
• Margin variance across the window is material: filed operating margin 31.2% (FY2021), 31.4% (FY2022), 28.7% (FY2023), 21.5% (FY2024), 26.0% (FY2025), 27.0% (TTM). A 9.9pp range on a 6-year window is not a State A low-variance profile.
• A structural regime change is present: the CSI Compressco acquisition (2024-04-01) stepped revenue +43.7% in one quarter, which independently disqualifies State A.

Dating the cycle position — required for State B, and this is the load-bearing sentence:

Kodiak is at or very near a cycle peak on the variable that matters. Fleet utilisation is 98.0%, up from 96.9%, the highest in the disclosed series; revenue per revenue-generating horsepower is rising +3.7% year over year; and management's prepared-remark mentions of "pricing" went from 0–1 per call through 2024 to 9 in 2025Q4. Tight capacity plus rising price is the definition of a peak, not a mid-point. I am therefore NOT mean-reverting the margin downward by reflex — the brief is explicit about that error — but I am refusing to extrapolate the +3.7% price term, and I set terminal revenue growth without it.

Consequence of State B: normalise on both revenue and margin; report trough / mid-cycle / peak earnings power; do not combine a peak margin with a peak growth rate.

---

TERMINAL MARGIN — 27.0%

Basis: the company's own TTM filed operating margin, reconciled through the FY2025 10-K opex bridge. Not a peer median, not an industry percentile, not max(own, sector).

TTM through Q1 2026, EDGAR OperatingIncomeLoss (filed):

The opex bridge, from 10-K lines (FY2025)

The bridge closes to the dollar: 42.20 − 11.01 − 0.49 − 4.71 = 25.99% ✓

Hard constraint

m_EBIT,T (27.0%) ≤ m_gross,T (42.2%) ✓ — with 15.2pp of headroom, which is ample to fund an 11.0%-of-revenue SG&A base plus normalised disposal losses.

Why 27.0% and not something else

• It equals the trailing actual. The brief's warning is that a terminal margin below the trailing actual is almost certainly an error. This is not below it.
• Normalising the abnormal lines would justify more, not less. FY2025's loss on sale of assets was 4.71% of revenue against 2.55% in FY2024 and −0.09% in FY2023 (a small gain); the 3-year mean is 2.39%. Normalising disposal losses to 2.4% and impairment to 0.4% gives 42.2 − 11.0 − 0.4 − 2.4 = 28.4%. I did not take that, because the fleet-renewal programme that generates the disposals is ongoing and management has not said it ends.
• It sits inside the demonstrated range (21.5%–31.4% filed) and below the FY2021–22 pre-acquisition peak of ~31%.

steady_state_check.py — what it flagged, and a defect in its input

python3 steady_state_check.py --ticker KGS --terminal-margin 0.270 --exit-multiple 7.50

The tool passes KGS, and its headline "latest margin" is wrong. It reads AV's operatingIncome, which for FY2025 gives 31.7% against a filed 26.0% (see the defect register). Against the filed figure the terminal margin is 0.0pp from the trailing actual, not −4.7pp. The verdict is unchanged — the tool's substantive finding, that a 7.2x warranted multiple is consistent with Kodiak's low ROIC, is correct and is the most useful thing it says.

---

THE IMPLIED-PATH TEST — the Valuation Criteria

Verified inputs (every one re-derived; screen inputs cross-checked)

EV basis statement, as required: the headline EV of $7,921.3m is LEASE-EXCLUSIVE. Adding disclosed non-current operating and finance lease liabilities and an estimated current portion takes it to roughly $7,983m, +0.8%. The difference does not move any conclusion here (EV/EBIT 22.15x → 22.32x); it is stated because a lease-exclusive anchor previously invalidated a seven-name ladder in this project.

Current multiples

The instrument is a discounted cash flow — confirmed

Per the brief's CORZ/IREN warning: the instrument used here is ~/.claude/skills/investment-memo/assets/reverse_dcf.py, which discounts a five-year interim free cash flow stream plus a discounted terminal value at an explicit WACC and solves for the revenue CAGR that sets present value equal to today's EV. It is not an undiscounted required-revenue test. It was executed; the outputs below are its stdout, not a reconstruction.

--fcf-margin — sign and magnitude

Kodiak's free cash flow is POSITIVE, so the direction of the interim-FCF correction is the conservative one: omitting the flag would overstate required CAGR and could only suppress a true PASS, never manufacture a false one. It was supplied anyway.

• TTM operating cash flow $556.7m; TTM capex $354.5m; TTM FCF $202.2m = +15.3% of revenue.
• Marginal / forward basis check (the GOOGL lesson — trailing sign can mislead): Q1 2026 alone was OCF $71.2m against capex $110.9m, i.e. −$39.7m, −11.5% free cash flow, because growth capex stepped from $56.0m to $85.6m. Kodiak is capable of printing a negative FCF quarter. Over any four-quarter window in the record it has not.
• normalized_fcf.py returns 16.2% at 6% assumed growth (with the two offsetting input errors documented in the research file). Trailing measures 15.3%.
• Used: --fcf-margin 0.155. Sensitivity: at 0.05 the required CAGR rises ~1.3pp; at −0.05 it falls ~2.7pp and would flatter the name. The sign is positive and verified.

Parameters

The exit multiple is determined by the identity, and this is where KGS breaks

EV_T / EBIT_T = (1−t)(1 − g/ROIC) / (WACC − g)

with t = 28.1% (FY2025 effective rate 31,884/113,472), g = 2.5%, WACC = 8.5%, and ROIC = 6.66% (NOPAT $257.5m ÷ invested capital $3,867.6m):

(0.719)(1 − 0.025/0.0666) / (0.085 − 0.025) = (0.719)(0.6246) / 0.060 = 7.49x

steady_state_check.py independently returns 7.2x at its own 8.3% ROIC. Two derivations, same neighbourhood.

Kodiak trades at 22.15x EV/EBIT against a 7.5x steady-state-warranted multiple — a 2.95x compression. The cause is not the margin and not the growth rate; it is that ROIC is 6.66% against a WACC of 8.5%, so terminal reinvestment consumes 37.5% of NOPAT to buy 2.5% growth that is worth less than it costs.

RESULT

  EV implied by today's price          7,921m   (6.0x current revenue)
  Held fixed              terminal margin=27.0%, exit multiple=7.5x
  Discount rate / horizon 8.5% over 5y

  >>> THE MARKET REQUIRES: revenue CAGR of 28.2%

Sensitivity — over the exit multiple, never over scenario probabilities

The verdict is entirely the exit multiple, and the screen's is unsupportable. The screen used 24.2x, described as GROWTH_MATCHED from a comparator set of 223 names — which is a universe-wide median wearing a sector label, exactly the defect recorded in DATA_DEFECTS.md. At 24.2x the name passes by +13.8pp. At the identity-warranted 7.5x it fails by −23.3pp. A 37pp swing produced by one unexamined parameter.

The honest statement: KGS passes only if you assume its multiple never compresses. Holding today's 22.15x flat for five years, the required CAGR is 5.9% against 4.9% demonstrated — still −1.0pp, still a FAIL, but a narrow one. That is the most generous defensible framing and it is still not a pass.

Trough / mid-cycle / peak earnings power — required for State B

The price paid against mid-cycle earnings power is 25.7x EV/EBIT. For a business earning a 6.7% ROIC, that is the finding.

---

THE 12-MONTH TARGET

Multiple anchor — the name's OWN history, percentile stated

Daily EV/Sales series built from Alpaca split-adjusted closes and as-known annual revenue (FY revenue stepping in at its 10-K filing date, so the series is not forward-looking), current share count and net debt held constant so the series measures multiple movement.

Window: 2024-03-07 → 2026-07-29, n = 600 trading days. This is Kodiak's entire life as a filer with a 10-K on record — it IPO'd in June 2023.

History length caveat, declared: 2.4 years and 600 observations. That is short, and the whole window sits inside a single up-cycle in Permian compression. It is not long enough to have observed a downturn multiple. I am using it because valuation.md requires the name's own history and forbids substituting a peer median — but the percentile should be read as "where in this cycle", not "where in the range of possible regimes."

NTM revenue base

Estimate revision direction — 90-day, from the built-in history

Revisions are negative — the worst of the three names in this cluster. The last 7 days turned up (+2.6% on FY2026), which is a change of direction worth one quarter of observation, not a trend.

Target

At the 50th percentile of its own history, 5.95x EV/Sales, on consensus NTM revenue:

On the house NTM of $1,450m the same multiple gives $66.86, +13.5%.

The single largest swing factor is the consensus revenue path, and it is not evidenced. Consensus requires Q2'26 revenue of $383.0m against Q1'26 actual of $345.8m — a +10.8% sequential step in a business growing 4.9% year over year at 98% fleet utilisation. I report the consensus-based target as the method-compliant headline and the house figure as the number I would actually underwrite.

Leverage amplifies this: net debt is 34.0% of EV, so a 12% revenue miss becomes an 16% equity move.

No external professional target is on file for KGS in this project, so the sanity band is UNAVAILABLE rather than passed.

---

COMPARATOR SET — KGS only

Growth-matched, US-listed, contract compression and gas-handling infrastructure. Kept separate from the XYL and FSS sets.

Declared limitation: the comparator set has n = 2 direct peers, both of which share Kodiak's exact cycle position, leverage profile and ROIC problem. A median of two co-cyclical peers is not an independent anchor — it is the same observation three times. This is why the exit multiple in the implied-path test is taken from the ROIC identity and not from the comparator set, and why the 12-month multiple is taken from Kodiak's own history. The comparator set is recorded for completeness and is not load-bearing on any number in this memo.

The screen's exit_multiple_peer_n: 223 is the defect this replaces.

---

SUMMARY