CRDO — Valuation

Two outputs, two horizons. Reporting only one is a defect. Spot $192.28 (2026-07-28 close, Alpaca SIP). As of 2026-07-29.

---

1. Balance-sheet corrections applied before any valuation

Both bases are carried through the sensitivity tables below so the effect of the correction is visible rather than assumed.

---

2. Terminal value dominates → the reverse DCF is the primary long-horizon output

With FY2026 EBIT of $445.0m and an EV of $36.1bn, essentially all of the value sits beyond the visible forecast period. Terminal value is far above the 60%-of-EV threshold, so per references/valuation.md the reverse DCF is mandatory as the primary output and a forward DCF would be supporting evidence at best. No forward DCF is presented, because it would only restate the terminal assumption with false precision.

---

3. The implied-path test

assets/reverse_dcf.py, solving for revenue CAGR. Horizon 5 years, WACC 10.0%, exit multiple applied to terminal EBIT.

3.1 The required parameter, and everything held fixed

Held fixed in the base case, named explicitly: terminal EBIT margin 25.0%; exit multiple 27.0x EBIT; WACC 10.0%; horizon 5 years; diluted shares 192.0m; net cash $673.3m; base revenue $1,335.116m (FY2026 as filed).

3.2 Why the terminal margin was moved off the screen's 16.4%

The screen set terminal margin as max(own, industry median) then capped it at the industry 75th percentile of 16.4%, down from Credo's own demonstrated 33.3%. That is a 17pp haircut to a margin the company is already earning, and it is the single largest driver of the answer: it alone moves required CAGR from 37.1% to 58.0%.

The same double-counting objection the framework raised against NTRA's exit multiple applies here to the margin: a distant-year haircut stacked on a figure that has already been conservatively set, applied silently. Credo is a fabless designer with 68% gross margin, ~0.2% cash tax (Cayman domicile) and no fab capex; capping it at a semiconductor-industry p75 that includes capital-intensive manufacturers is not conservatism, it is a category error.

Base case uses 25.0% — below Credo's demonstrated 33.3%, to reflect (a) genuine ASP erosion in semiconductors, which Credo's own risk factors state explicitly, (b) the acquisition-driven opex step-up already visible in Q1 FY2027 guidance, and (c) mix dilution from optical transceivers, which carry lower margins than retimer silicon. Both 16.4% and 33.3% are shown so the reader can pick their own.

3.3 The margin — demonstrated − required

This is the number the strategy ranks on, and which "demonstrated" is used changes the sign.

Against the screen's trailing measure the margin is +48.3pp. Against the rate the business is compounding at today it is −11.4pp. Both are arithmetically correct. The trailing measure describes a period that includes a +205.7% year off a $184m base and a +4.8% year; the run-rate measure describes what management guided to on 2026-02-09 and has now printed twice. I regard the run-rate measure as the decision-relevant one, and the gap between the two is the entire investment question on this name.

Note the required CAGR for the first year is not 45.2% — it is whatever the company does. FY2027 at $2.08bn (base path, §5) is +55.5%, comfortably above the required average. The strain is in years 3–5, where a $1.34bn business must still be adding ~$2bn of revenue a year.

3.4 Exit multiple, and the implied compression — stated as a number

The screen's anchoring is growth-matched with n=44, which satisfies the "growth brackets the subject at the exit year" requirement — the defect that produced every prior high-growth rejection (a 1.0–7.5% anchor set used on 15–39% growers) is not present here. The base multiple of 27.0x is not below every stated anchor: on a sales basis, CRDO's own AI-era EV/sales percentiles run 17.2x (p10) to 46.4x (p90), and 27.0x EBIT at a 25% terminal margin corresponds to 6.8x sales — well below its own history, which is the conservative direction.

3.5 Sensitivity over the exit multiple — the highest-variance parameter

Corrected balance sheet, terminal margin 25.0%, WACC 10.0%, 5 years.

And over terminal margin, at 27x:

The plausible range of required CAGR is 34–58%. Every point in that range is below the trailing 93.5% and above the 33.8% current run-rate. There is no parameterisation that makes the answer obvious either way — which is the honest result, and is why the sensitivity is run on the two parameters that can change it rather than on scenario probabilities.

---

4. The management-revealed cross-check: the CEO's PSU hurdles

This is the strongest external check available on this name, and it is a filed document rather than an estimate.

10-K Note 16: on 2026-05-28 the board approved 100% performance-based PSUs for the CEO, six equal tranches of 239,500 shares, over a five-year performance period ending 2031-06-30, each requiring both a revenue goal and a stock-price goal:

Read this carefully, because it cuts both ways and both cuts matter.

Against the price: the base implied path requires $8.62bn of FY2031 revenue. That is 15% above the maximum-payout hurdle in the CEO's own five-year incentive plan — the tranche the board designed as the "progressively challenging" ceiling. On the screen's uncorrected parameters the requirement is $12.5bn, 66% above the maximum hurdle. A compensation committee that thought $8.6bn was the central case would not have put $7.5bn at the top of the ladder.

For the price: the board also attached a $489.40 stock-price hurdle to that top tranche — 2.55x spot, a 20.5% annual return over five years — and the ladder starts at $244.70, which is 27% above spot. Management and board are underwriting a materially higher share price than today's. And at tranche 5 ($6.5bn revenue), the current price implies a 26.9x exit EBIT multiple on a 33.3% margin, which is not an absurd number for a business still growing ~30%.

The reconciliation: the price is roughly consistent with tranches 5–6 being achieved, on a generous terminal margin, with a full multiple compression to 27x. It is not consistent with tranches 1–4. The market is pricing the top of management's own incentive ladder as the base case.

Independent scale check

Marvell's trailing twelve-month revenue is $8.39bn. The base implied path requires Credo to reach $8.62bn by FY2031 — i.e. to become larger than Marvell is today, from $1.34bn, in five years, selling primarily interconnect. The 2026-04-13 8-K sizes the entire silicon-photonics PIC market at $6bn by 2030 (LightCounting + Credo estimates). AECs and DSPs are separate markets and no total-TAM figure is disclosed by the company, so this is a scale sanity check rather than a TAM constraint — but it is the right order-of-magnitude question to ask before accepting $8.6bn.

---

5. The 12-month target

Built per references/valuation.md: near-term estimates plus dated product events, on CRDO's own multiple history with the percentile stated. Not a peer median and not a DCF output.

5.1 Near-term revenue base

Q1 FY2027 is guided to $465–475m (midpoint $470m, +7.6% sequential). Consensus is unavailable (Alpha Vantage quota exhausted), so the paths below are built off company guidance and the observed sequential series.

The base path is management's own "mid-single-digit sequential" language taken at face value with a mild decay. The bull path requires the ">$500m optical revenue in fiscal 2027" target to land and AEC to hold — optical alone at $500m against $2.25bn is consistent with it.

5.2 The multiple anchor — CRDO's own history, percentile stated

EV / TTM sales computed daily from 2022-06-06 (first date with four reported quarters) to 2026-07-28, n=1,039 trading days, using reported shares outstanding and reported net cash on a 35-day filing lag — the same basis throughout the series, so it is internally comparable.

Regime-change treatment, declared. The full 2022–2026 history spans two distinct pricing regimes: pre-AI median 11.9x, AI-era median 28.5x. Blending them is not meaningful. The AI-era sub-window (2024-01-01 → 2026-07-28, n=644 trading days) is used as the anchor set. That is CRDO's own history, not a peer median, so this is not a case for declaring the multiple UNIDENTIFIED — but the restriction is stated explicitly and the full-history percentile is reported alongside so nothing is concealed.

Calendar-year medians, for the shape: 2022 12.6x, 2023 11.7x, 2024 20.8x, 2025 38.7x, 2026 YTD 29.0x (range 13.9x–40.9x). The de-rating is already well advanced — 25.6x is the 14th percentile of the trailing twelve months and the stock is 36.4% below its 52-week high of $302.52.

5.3 The targets

Valuation date July 2027. TTM revenue is Q2 FY2027 → Q1 FY2028. Net cash held at $673.3m (today's corrected figure, no credit taken for twelve months of free cash flow — deliberately conservative, and it costs roughly $3–4 per share of target). Shares 192.0m diluted.

12-month target: $238 (+23.6% to spot). Bear $123 (−36.3%). Bull $376 (+95.8%).

Why p25 and not the median. The AI-era median of 28.5x was set in a period when CRDO printed 150–275% YoY growth. At July 2027 the base path shows roughly 50% TTM YoY growth decelerating toward 30%. A multiple one quartile below the median of the higher-growth regime is the coherent choice, and 20.4x is a stated anchor from CRDO's own distribution rather than an arbitrary haircut — the NTRA defect (base below every named anchor, silently) is avoided. The base target also embeds ~20% multiple compression from today's 25.6x alongside 65% TTM revenue growth, which is a coherent pair rather than a double-count in either direction.

Sanity band. No external professional target was available for CRDO in this corpus, so the required comparison against an outside target cannot be made. Recorded as a gap.

Note on shape. The distribution is extremely wide — −36% to +96% on defensible parameters — because a 91%-vol name with three customers is genuinely that uncertain. A point target on this name conveys less than the range does, and the range is the output.

---

6. Peer Spread Criteria — MEASURED

ALAB — PCIe/CXL connectivity into the same hyperscaler AI buildout — is the closest comparable and trades at roughly 1.7x CRDO's multiple on 25% less revenue, while growing sequentially at roughly twice CRDO's rate. That relationship is internally coherent and it means CRDO is not the expensive name in its own peer set; it is priced between the pure-play hyper-grower and the scaled diversified incumbent, which is roughly where its growth profile now sits.

Limitations: ALAB's MarketableSecuritiesCurrent tag is absent from companyfacts, so its net cash is understated and its EV/sales is an upper bound (with ~$1bn of investments it would be ~43.5x). Peer own-history percentiles were not computed — point-in-time only. Prices are 2026-07-28 closes.

---

7. Valuation Criteria verdict

PASS WITH ARGUMENT.

The price requires 45.2% (range 34–58%) against a demonstrated 93.5% trailing three-year CAGR — that is a PASS on the letter of the test. But the letter of the test is satisfied by a measure the business is no longer running at, and against the current run-rate of 33.8% the margin is negative by 11.4pp. So the criterion does not resolve to a clean PASS and it would be dishonest to record one.

The specific, evidenced, dated argument required for PASS WITH ARGUMENT:

1. A named product cycle with a quantified target and a date — ">$500 million in optical revenue in fiscal 2027" (8-K, 2026-04-13), spanning ZeroFlap Optical Transceivers, Optical DSPs and DustPhotonics silicon photonics. Against $1,335m of FY2026 revenue this is a step-change in a second product line, and it is testable within four quarters.
2. $897m of capital already deployed to build it — Hyperlume, CoMira and DustPhotonics, closed, not contemplated.
3. The management-revealed forecast — a five-year PSU ladder running to $7.5bn of revenue and a $489.40 share price, filed 2026-05-28.
4. A margin structure that has already proven out — 68.0% gross margin and 33.3% GAAP operating margin at $1.34bn of revenue, which is what makes even the 25% terminal margin assumption a haircut rather than a hope.

What is explicitly NOT the argument: that 93.5% is repeatable. It is not, on the company's own guidance.

The residual question the argument does not close is whether Credo can add roughly $1.5–2.0bn of revenue per year in FY2029–FY2031, which is what the back half of the required path demands, from three customers with no contracted backlog. Nothing in the file evidences that, and it is the reason this is PASS WITH ARGUMENT and not PASS.

---

8. Downside Criteria — MEASURED, with a named and realised cause

Named cause: a demand pause or inventory correction at one of the three end customers that constitute 84% of revenue.

This is not a modelled scenario. It has already happened to this company. In FY2023 end-customer Z was 55% of revenue (~$101m). In FY2024 it was 26% (~$50m). Quarterly revenue fell 40.9% sequentially in Q4 FY2023, full-year FY2024 grew 4.8%, and the stock drew down 62% peak-to-trough within calendar 2023.

The mechanism, restated for today's book. Customer D is ~$441m, Customer B ~$427m, Customer E ~$254m of FY2026 revenue. There is $31.9m of contracted backlog against $1,335m of annual revenue and orders are cancellable on short notice without penalty. A 50% cut at one of the top two — the exact magnitude already observed — removes ~$215m, ~16% of revenue, and would take reported growth to roughly flat.

The leading indicators to watch, in order: (1) days of inventory, already 164 and up 16 days sequentially while revenue grew 7.4%; (2) a fourth consecutive quarter of sub-10% sequential growth; (3) any change in the composition of the customer-concentration table in the FY2027 10-Q filings.

Quantified bear case. Bear revenue path (§5.1) at 13.9x — the 2026 year-to-date low and the AI-era minimum multiple, not an invented floor — gives $123, −36.3%. This is not the worst case: it holds the multiple at a level the stock has actually traded at during a growth quarter. A genuine correction would compress the multiple and the revenue simultaneously. At 10x on $1,641m TTM the target is $89, −54%, which is close to the 52-week low of $87.81 and is the more honest tail.

Probability assigned: 30% that at least one of the three end customers cuts orders materially within eight quarters. Basis: it has occurred once in the four years of disclosure available (FY2024), hyperscaler capex digestion cycles have historically run on a two-to-three-year cadence, and Credo's own risk factors state the customers may "cancel purchase orders on relatively short notice… without penalty." This is a judgement, it is logged for Brier scoring, and it constrains nothing.

Going concern: not argued. $673.3m of net cash after DustPhotonics, no debt, $464.3m of FY2026 operating cash flow and 68% gross margin. Even the FY2024 correction left Credo cash-generative. The realistic permanent-loss case here is a permanent de-rating, not an impairment of the enterprise — and that distinction matters, because a business that survives its air pocket and re-accelerates (as this one did, from $164m TTM to $1,335m in nine quarters) is a very different holding from one that does not.

Volatility, for sizing only, not as the risk: 252-day realised 90.7%; within-year peak-to-trough drawdowns of −39% (2022), −62% (2023), −33% (2024), −61% (2025), −46% (2026 YTD). Inverse-volatility sizing will size this name down hard and that is the intended interim control.