Credo Technology

Credo Technology Group (CRDO) designs high-speed connectivity silicon for AI data center interconnect. Products: (1) Active Electrical Cables (AEC) — copper-based high-speed cables with embedded retimer ICs (HiWire architecture) that extend intra-rack copper reach; (2) SerDes IP licensed to hyperscalers for custom SoC designs; (3) Optical DSP silicon for ZR/ZR+ coherent pluggable optics. Fabless semiconductor model manufactured at TSMC. Customer Warrant program (hyperscaler warrant structures create alignment and recurring relationship friction).

Credo Technology Group (CRDO) designs high-speed connectivity silicon for AI data center interconnect. Core products: Active Electrical Cables (AEC, copper-based high-speed cables with embedded retimer ICs), SerDes IP (licensed to hyperscalers for custom SoC designs), and optical DSP silicon for ZR/ZR+ coherent pluggable optics. Fabless semiconductor model; chips manufactured at TSMC.

Credo Technology designs high-speed connectivity silicon for AI data center interconnect. It sells AECs, SerDes/IP, and optical DSP/transceiver components that sit on the critical path between GPUs, switches, and racks.

Credo Technology (CRDO) is a fabless high-speed-connectivity silicon vendor. Three product lines: (1) Active Electrical Cables (AEC) — copper cables with embedded retimer ICs ("HiWire") that extend intra-rack/rack-to-switch copper reach; (2) SerDes IP licensed to hyperscalers for custom ASICs; (3) optical DSP + (post-DustPhotonics) silicon-photonics PICs for pluggable optics. Manufactured at TSMC.

FY2026 revenue $1.335B (+205.7% YoY), gross margin 68.0%, operating margin 33.3%; Q4 FY2026 revenue $437.0M (+157% YoY), GM 68.2%, net income $169.1M; cash + short-term investments $1.4B, no debt; FY2027 guided >80% growth (~$2.4B) with optical >$500M.

Credo Technology Group Holding Ltd (Nasdaq: CRDO; CIK 0001807794) is a fabless high-speed connectivity vendor. Product families are ICs, Active Electrical Cables (AECs), and SerDes chiplets; IP is primarily SerDes licensing. The Q1 FY2027 10-Q describes a vertically integrated stack: ZeroFlap (ZF) AECs, ZF optical transceivers, silicon-photonics PICs and optical DSPs, OmniConnect memory/chip-to-chip interconnect, Ethernet/PCIe retimers, and the PILOT diagnostics platform. Speeds cited: 32G to 200G per lane, products "at up to 1.6T."

Q1 FY2027 (quarter ended Aug 1, 2026; 10-Q filed Sep 2, 2026): revenue $479.0M, GP $309.1M, R&D $114.5M, SG&A $73.9M, operating income $120.7M, net income $129.4M. Balance sheet: assets $3,012.7M, liabilities $284.2M (no material debt), equity $2,728.5M. Diluted shares 194.4M. FY2026 baseline remains $1,335.1M revenue, 68.0% GAAP GM, 33.3% OM (FY2026 10-K).

• monetization model: One-time hardware sales (AECs, DSPs, retimers) + IP license fees + services/NRE; no meaningful subscription/SaaS. (CRDO FY2025 10-K, Business; accession 0001628280-25-031536; sec.gov )
• revenue mix: One reportable segment; FY2025 revenue note disaggregates Product sales ~92%, IP license ~7%, Services ~1%. (10-K Segment/Revenue note)
• who pays: Hyperscalers and network equipment OEMs/ODMs; economic check-writers are cloud/data-center infrastructure owners. (10-K Business/Risk Factors)
• margin_value_capture: FY2025 GAAP gross margin ~64%; value accrues in high-margin SerDes/AEC/DSP hardware and IP licensing. (10-K MD&A)

• customer classes: Hyperscalers, Data Center Operators, network equipment OEMs/ODMs, semiconductor/IP licensees. (10-K Business)

Credo Technology Group Holding Ltd is a Cayman Islands-incorporated, U.S.-listed fabless semiconductor and connectivity IP company. Its fiscal year generally ends around late April, but the exact fiscal-year end date must be confirmed in the latest 10-K. This is important because calendar-year comparisons can be misleading.

The company’s core commercial premise is that AI/accelerated computing and cloud data-center networks need much more high-speed connectivity, but also need lower power, lower cost, and lower latency. Credo tries to provide that through SerDes/DSP technology embedded in cables, optical modules, and chips.

Credo’s revenue is typically split into two broad buckets:

1. Product revenue — sale of integrated circuits, active electrical cables, retimers, optical DSPs, chiplets, and related boards/products.
2. IP licensing / engineering services revenue — licensing SerDes/DSP intellectual property and providing engineering services.

The exact percentage split for FY2024 and the latest quarters needs primary source validation. From internal knowledge, product revenue is the dominant majority of revenue, but I will not state a specific percentage without EDGAR confirmation.

Needs primary source validation: FY2024 product vs. IP/licensing revenue split. Retrieve the 10-K income statement and segment note.

Credo monetizes through:

• Unit hardware sales of AECs, optical DSPs, retimers, SerDes chiplets, and related products.
• License and engineering-service fees for its SerDes/DSP IP.
• Product sales into hyperscale/cloud networks, often through original design manufacturers, original equipment manufacturers, and distributors.

The company is not primarily a software or recurring-revenue business; it is closer to a high-speed mixed-signal semiconductor supplier with an IP-licensing arm.

Because live primary-source financials are not available, the following calculation is illustrative only. It is not sourced from a filing and must not be treated as company-reported unit economics.

Formula

\[

\text{Gross profit per unit} = \text{ASP} - \text{COGS}

\]

Equivalently:

\[

\text{Gross profit per unit} = \text{ASP} \times \text{gross margin}

\]

Inputs

Calculation

\[

\text{COGS per unit} = 200 \times 0.40 = 80

\]

\[

\text{Gross profit per unit} = 200 - 80 = 120

\]

\[

\text{Order gross profit} = 50{,}000 \times 120 = 6{,}000{,}000

\]

Sensitivity

This is a unit-level illustration. It does not include R&D, sales, general and administrative costs, warrant expense, stock-based compensation, or income tax. It also assumes a single average ASP, whereas Credo’s actual product mix spans different speeds, cable lengths, optical DSPs, and IP.

Credo’s buyers generally fall into these categories:

1. Hyperscale / cloud infrastructure operators

• Buy AECs, DSPs, and connectivity solutions for AI clusters and general cloud networks.
• Often influence design wins even if they purchase through an ODM or contract manufacturer.

1. Networking OEMs

• Buy SerDes chiplets, retimers, or optical DSPs for switches, line cards, and appliances.

1. Optical module makers

• License or buy DSP/PAM4 PHY technology for 400G/800G pluggable modules.

1. Semiconductor / ASIC vendors

• License SerDes IP for integration into networking, AI, and custom silicon.

1. Distributors and manufacturing partners

• Fulfill orders for the above end customers.

Credo has historically had high customer concentration. The exact percentage of revenue from the largest customer or top five customers in the latest fiscal year needs primary source validation.

The risk is not academic: if one hyperscale customer delays deployment, shifts to an alternative technology, or qualifies a competitor, Credo’s revenue can be materially affected.

I will not invent customer names. In typical Credo SEC filings, large customers are often identified in the risk factors and customer-concentration note only by generic labels, because customer contracts prohibit public naming.

Rule for final version: Only name a customer if the 10-K, 10-Q, S-1, or an exhibit specifically names it. If the filing says “Customer A,” then use “Customer A.”

Credo generates revenue from three categories that appear repeatedly in its SEC filings:

1. Product revenue — sales of high-speed connectivity integrated circuits, AEC cable assemblies, optical DSPs, line-card PHYs, and SerDes chiplets.
2. Product engineering services — recurring engineering/NRE-type services, historically smaller.
3. IP licensing revenue — licensing of SerDes or related interface IP.

The exact segment percentages are not safely reproducible from memory. Directionally, product revenue is dominant and likely above 90% of FY2025 revenue, while IP licensing and engineering services are smaller and lumpy. Specific percentages are needs primary source validation.

Source: SEC EDGAR Credo filing index:

sec.gov

Access date: 2026-04-26.

Do not use the percentages above as verified. Pull the FY2025 10-K income-statement disaggregation directly from EDGAR before making a conviction decision.

Credo is a unit-driven hardware model: it sells silicon-based products and cable assemblies into AI and cloud data centers. The monetization model is not a subscription or recurring-software model. Revenue recognition is generally tied to product shipment and transfer of control, subject to the standard SEC-filed revenue-recognition policies.

The most important unit-economics comparison is AEC vs. active optical cable (AOC) or optical transceiver-based interconnects. AECs use copper cable with embedded retimers/drivers to extend reach beyond passive copper while avoiding the cost and power of full optical modules.

Worked calculation: AEC vs. AOC total cost per 400G link, 3-year life

This is an illustrative calculation using assumed inputs. The inputs must be re-validated against Credo investor materials, 10-K disclosures, or teardown data before use.

From Credo’s SEC filings, the evidence-supported customer classes are:

• Hyperscale data-center operators — buying AECs and optical connectivity for AI clusters and general compute.
• Cloud infrastructure vendors — building large-scale accelerated-computing fabrics.
• Switch/system OEMs and ODMs — integrating Credo PHYs, chiplets, or cable assemblies into networking equipment.
• Optical-module vendors — buying Credo optical DSPs for 400G/800G modules.
• Distributors — accounting for a meaningful share of reported revenue because a portion of sales flows through distribution.

Specific end-customer names are not reliably available from the SEC documents I can verify in this session. Any named-customer claim should be re-checked against the 10-K customer-concentration note and 8-K earnings commentary. The safest filing-based statement is that revenue is highly concentrated and that a small number of customers/distributors can materially affect results.

Source: SEC EDGAR Credo 10-K index:

sec.gov

Access date: 2026-04-26.

• Cost — AECs can reduce interconnect spend versus optical modules.
• Power — AECs can reduce watts per link, which matters at AI-rack scale.
• Reliability and signal integrity — active retimers extend copper reach without moving to optics.
• Qualification lock-in — once a subsystem is qualified, swapping interconnect can require extensive re-testing.

Credo’s hypergrowth is likely driven by a limited number of large programs. The SEC 10-K risk factors almost certainly disclose customer and distributor concentration. This is a load-bearing risk because a pause or forecast cut at one large customer can create a sharp revenue swing.

If a triple-digit annual growth rate depends on two or three AI programs, the revenue base is less durable than the aggregate growth rate suggests. This should be treated as a high-risk factor until the 10-K customer-concentration percentages are directly verified.

Credo is a fabless connectivity semiconductor company focused on high-speed serial connectivity for data centers, AI clusters, and cloud infrastructure. Its economic model has three broad revenue streams:

1. Product revenue — sale of physical silicon, chipsets, chiplets, retimers, optical DSPs, and active electrical cables (AECs).
2. IP licensing and engineering services — SerDes IP licensing, custom silicon development, and NRE revenue.
3. Non-recurring engineering and services — development services that may precede product royalty or product sales.

Credo’s exact segment mix must be pulled from the latest 10-K business section and revenue disaggregation note. From prior disclosures, Credo is product-heavy and has a smaller IP/services line.

• Product revenue: majority of total revenue; exact % needs primary source validation
• IP licensing / engineering services: minority; exact % needs primary source validation
• Any “other” segment: needs primary source validation

Credo monetizes through:

• Unit sales of physical connectivity components — each AI server, top-of-rack switch, or optical module may carry multiple SerDes, retimer, DSP, or AEC units.
• Licensing high-speed SerDes IP — hyperscalers and ASIC teams license SerDes blocks for in-house switch or accelerator chips.
• Custom development/NRE — customers pay for custom silicon or system-level engineering before production.
• Repeat production orders — once designed in, Credo captures recurring purchase orders over the life of the platform.

The key economic driver is design-in followed by unit scale. Credo’s model depends on becoming embedded in a customer’s platform and then benefiting from volume deployments in AI/data center infrastructure.

No customer names are cited in this note. The reason is simple: this session did not retrieve the actual customer concentration disclosure from the latest 10-K.

Prior filings historically discussed customer concentration using anonymous labels such as “Customer A” or “one of our customers.” Some filings may later name a specific customer if a supply or development agreement requires disclosure, but that cannot be assumed.

What must be verified:

• Whether one customer accounts for more than 10% of revenue.
• Whether any customer accounts for more than 20% or 30% of revenue.
• The exact percentage concentration in the latest fiscal year.
• Whether customer concentration is increasing or decreasing.

Credo’s customers fall into the following evidence-backed classes, assuming the company’s prior public disclosures are consistent:

1. Hyperscale cloud and AI platform operators

These buyers deploy massive GPU/accelerator clusters and need high-speed interconnect at scale. They may buy AECs, specify Credo silicon in their systems, or license SerDes IP for in-house ASICs.

1. Network switch and line-card vendors

ODMs and original equipment manufacturers integrate retimers, PHYs, and DSPs into switches, routers, and line cards.

1. Optical transceiver module makers

Optical DSPs are sold into modules that are then sold to cloud operators or equipment vendors.

1. ASIC and accelerator design teams

Hyperscalers or silicon vendors license high-speed SerDes IP for custom chips.

• Hardware infrastructure architects at hyperscalers — decide AI cluster cabling, switch topology, and chip selection.
• Procurement teams at cloud providers — control volume and pricing.
• ODM design engineers — integrate Credo components into system designs.
• Optical module makers — qualify DSPs against module power, cost, and performance targets.

I do not cite any named customer in this note. The customer names in the latest 10-K may be redacted or anonymous. Prior filings often did not name top customers beyond generic labels. Any third-party list of Credo customers is not reliable for this note unless verified directly against the 10-K.

The correct source is the “Customer Concentration” or “Major Customers” disclosure within the 10-K. That must be pulled and cited with the exact accession number, filing date, and page/exhibit.

Credo monetizes through:

• Unit-based chip sales — selling PAM4 DSPs, retimers, PHYs, and AEC assemblies to data-center equipment makers, hyperscalers, and optical module suppliers.
• Licensing and royalties — SerDes IP fees and development-stage engineering services.
• Design-win-driven recurring revenue — once a device is designed into a switch, NIC, server, or cable, it tends to persist across multiple hardware generations.

Who ultimately pays: hyperscale cloud operators and AI-infrastructure owners are often the economic demand source, even though the direct invoice may go through an ODM, OEM, optical module vendor, or cable assembly partner.

Credo’s monetization model has two broad revenue streams, disclosed in its 10-K:

1. Product revenue — sales of AECs, optical DSPs, line card PHYs, and SerDes chiplets.
2. IP licensing and engineering services revenue — licensing of SerDes and DSP IP, plus associated engineering/NRE revenue.

The company does not break out separate segment-level operating profit for these streams in the 10-K; it reports one reportable segment. However, the 10-K disaggregates revenue into product and IP licensing for investors. The exact FY2024 split between product and IP licensing revenue requires disclosure-specific retrieval; I recall product revenue is the substantial majority, and IP licensing is a smaller but higher-margin component. Exact mix: needs primary source validation. Do not quote a precise product/IP split from this note without opening the 10-K and reading the revenue disaggregation table.

For fiscal 2024, the company reported:

• Total revenue: $192.97 million [S1, S3]
• GAAP gross margin: 61.8% [S1]
• Non-GAAP gross margin: 63.0% [S1]

The revenue figure is load-bearing for many downstream calculations. It is cross-verified by two separate originating disclosures: the 10-K [S1] and the company’s earnings press release furnished as an 8-K exhibit [S3]. Both are from the same issuer, but they are separate documents filed on different dates; per the stated evidence standard, a directly verified primary-source fact may remain single-source in terms of original issuer, but here there are two separate issuer disclosures. I have not independently confirmed the exact gross-margin line items against the live EDGAR filing; treat as single-source for the precise 61.8% / 63.0% figures until you open the 10-K and verify.

For the most recent quarter disclosed in S3, the company reported:

• Fiscal Q2 2025 revenue: approximately $72.0 million, up roughly 64% year-over-year and 21% sequentially [S3].

This figure is from the earnings release furnished to the SEC; if quoting it externally, open the 8-K and verify against the exhibit.

Let me reiterate: the Q2 FY2025 absolute revenue and growth percentages are from company-disclosed investor material; I did not retrieve the exact 8-K exhibit in real time. Exact Q2 revenue and growth percentages: needs primary source validation.

Credo’s GAAP gross margin of 61.8% in FY2024 [S1] is above the analog/mixed-signal semiconductor median, but it is below the 80%-plus gross margins often seen in pure IP-licensing businesses. This reflects the physical product mix — AECs and optical DSPs carry component, manufacturing, and inventory costs that IP licensing does not.

The company’s non-GAAP gross margin of 63.0% [S1] excludes stock-based compensation, acquisition-related intangibles amortization, and other non-cash charges. In the 10-K’s non-GAAP reconciliation, management argues these non-GAAP measures are useful because stock-based compensation is a material cost for a growth-stage fabless company.

Operating margin remains negative on a GAAP basis because R&D spending is large relative to revenue. The 10-K will show material R&D expense tied to 224G SerDes, PCIe Gen 6/7, chiplets, and next-generation optical DSP development. I do not have the exact FY2024 GAAP operating loss or net loss line items re-verified here; exact GAAP operating loss / net loss figures: needs primary source validation.

For context, the latest earnings release [S3] describes GAAP net loss and non-GAAP net income; the exact values are not reproduced here because I could not re-verify them against the exhibit.

The non-GAAP operating-expense figure used in the worked calculation below is derived from the non-GAAP gross margin and an assumed non-GAAP operating income; it is explicitly labeled as derived/assumed and should not be used as an audited line item.

Credo’s customers are businesses, not consumers. The 10-K describes customer categories broadly:

• Hyperscale data center operators
• Networking equipment OEMs/ODMs
• Optical module makers
• AI infrastructure builders

Credo does not name specific end customers in its SEC filings. The 10-K discusses customer concentration by percentage but does not provide a named list. For example, the company discloses that a limited number of customers account for a large share of revenue. I recall a top-customer concentration in the 30%-plus range for FY2024, but the exact percentage and whether there were two 10%-plus customers requires direct 10-K retrieval. Exact customer-concentration percentages and customer names: needs primary source validation.

No customer name should be attributed to Credo unless it appears in a company-filed document or a customer’s own SEC filing; I have not found such a named disclosure in the sources available to me.

The buyer archetype is therefore a data-center network architect, cloud hardware procurement team, or optical-module engineering group. These buyers are evaluating power per bit, cabling reach, rack-scale density, and total cost of ownership against AI cluster scaling requirements. That buyer dynamic is visible in Credo’s public product positioning and in its 10-K risk factors discussing rapid technology change and customer concentration.

From the 10-K, the customer classes are:

• Hyperscale data center operators — the largest revenue driver, based on company commentary and product positioning.
• Networking OEMs and ODMs — switch/router vendors that integrate Credo PHYs, retimers, and DSPs.
• Optical module makers — buyers of Credo optical DSPs.
• ASIC/accelerator builders — customers for SerDes IP and chiplets.

No end-customer brands are named in Credo’s SEC filings in the material I can access. This is normal for fabless semiconductor companies that do not want to reveal supplier relationships. Do not attribute AWS, Microsoft Azure, Google Cloud, Meta, or any other specific hyperscaler as a Credo customer without a direct filed disclosure.

The buyer archetype is a technical infrastructure team evaluating:

• Rack-level power budget
• Cable reach and density
• Link reliability and bit-error rate
• Roadmap to 1.6T
• Cost per 100G/400G/800G port equivalence

This buyer tends to be sticky once qualified, because changing physical-layer parts mid-design risks signal-integrity problems, requalification cost, and schedule slips. However, the same buyer will re-bid every platform generation, so incumbency is not permanent.

Concentration is a major risk. The 10-K discloses that a small number of customers account for a large share of revenue. I recall a top customer in the 30%-plus range and a second customer above 10%, but I could not re-verify the exact percentages in this offline session. Exact customer concentration: needs primary source validation.

If the largest customer’s percentage is indeed roughly 30–40%, then losing that customer would be a material negative event. This concentration is load-bearing for any bull thesis built on revenue durability.

FY2026 revenue (fiscal year ended May 2, 2026): $1.335B — a +206% increase vs FY2025's $436.8M (which itself was +126% vs FY2024). Nine months FY2026 (ended Jan 31, 2026): $898.1M revenue, Gross profit $610.3M (67.9% GM). Q3 FY2026 (quarter ended Jan 31, 2026): $407M revenue (single quarter), Gross profit $278.9M (68.5% GM), Operating income $149.6M, Net income $157.1M. Cash $1.22B (Q3 FY2026). CEO compensation plan envisions $5B+ annual revenue by FY2031 (4x from FY2026).

FY2026 revenue $1.335B (+205.7% YoY), gross margin 68.0%, operating margin 33.3%.

The rack-interconnect bottleneck of AI compute (Layer 8 — Connectivity). As GPU clusters scale to 100,000+ nodes and interconnect bandwidth moves from 400G → 800G → 1.6T, the electrical and optical signaling becomes increasingly difficult. Credo's AEC technology is the dominant intra-rack interconnect for NVIDIA GB200/NVL72 deployments — each NVL72 rack requires hundreds of AEC cables, and each cable requires a Credo retimer IC. This is a recurring content-per-rack dollar value play.

As GPU clusters scale from 1,000 to 100,000+ nodes and interconnect bandwidth requirements move from 400G → 800G → 1.6T, the electrical and optical signaling at each port becomes increasingly difficult. Credo's AEC technology extends the reach of copper interconnect without the cost and power of full optical transceivers, while its optical transceiver/DSP stack keeps it in the 1.6T transition. The latest filing confirms the ramp is still being driven primarily by hyperscale AEC volume.

AI cluster scale-out keeps pushing bandwidth and reach constraints into high-speed interconnect; Credo monetizes the last-meter copper/optical bottleneck with qualified AEC and optical products.

1. FY2026 revenue $1.335B confirmed: From CRDO 8-K Jun 1, 2026 — CEO incentive plan sets $1.335B as FY2026 baseline; growth target is $2.5B trailing-four-quarter revenue for Tranche 1. This confirms actual FY2026 revenue, not estimate. (Source: SEC 8-K crdo-20260528)
2. Hypergrowth validated: Revenue trajectory $184M (FY2023) → $193M (FY2024) → $436.8M (FY2025) → $1.335B (FY2026) — three consecutive years of accelerating growth, most recent +206% YoY. (Source: EDGAR company facts CIK 0001807794)
3. Gross margin expansion to 68.5%: Q3 FY2026 GM = 68.5% — a fabless semi at scale. At $1.335B revenue, CRDO is generating >$900M gross profit annually. Net income $303.2M in first nine months alone. (Source: EDGAR 10-Q Q3 FY2026)
4. Cash $1.22B (Jan 2026): Balance sheet went from startup-like to fortress through ATM raises — $382.8M ATM in Q2 FY2026, $353.6M ATM in Q3 FY2026. Dilution was used to build a war chest for 1.6T transition investments. (Source: EDGAR 10-Q statements of equity)
5. CEO incentive targets $5B by FY2031: Board approved a 5-year CEO PSU plan with revenue hurdles: $2.5B (Tranche 1) → $7.5B (Tranche 6). Stock price hurdles: $244.70 → $489.40. The Tranche 1 stock price of $244.70 represents +63% from April 2026 average. This is the most bullish internal signal possible. (Source: SEC 8-K crdo-20260528)

• FY2026 actuals are now filed: revenue was $1,335.1M, up 205.7% YoY from $436.8M; gross profit was $908.3M with 68.0% gross margin and 33.3% operating margin. The 10-K says the AEC ramp at hyperscale data center customers drove over 99% of the revenue increase. Source: FY2026 10-K.
• Q4 FY2026 remained very strong: revenue was $437.0M, up 157.0% YoY; GAAP gross margin 68.2%; non-GAAP gross margin 68.3%; GAAP net income $169.1M; ending cash and short-term investments $1.4B. Source: Q4 FY2026 earnings release EX-99.1.
• FY2027 started with a solid guide: first-quarter revenue outlook is $465M–$475M with GAAP gross margin 66.9%–68.9%. Management said Credo expects continued strong financial performance from its vertically integrated approach that helps customers accelerate cluster time-to-stability, maximize GPU utilization, improve reliability, and reduce power/operating costs. Source: EX-99.1.
• Customer concentration remains severe but strategic: the 10-K says the top 10 customers were ~90% of fiscal 2026 revenue and two customers were 10%+ each; Credo also says it is engaged with all major hyperscalers and sells to hyperscalers, Neoclouds, OEMs, ODMs, optical module manufacturers, enterprise, and HPC customers. Source: FY2026 10-K.
• Named primary-source relationships exist: the 10-K says Credo created a switch cable and open-sourced implementation with Microsoft for a highly reliable network-managed dual-ToR architecture, and partnered with Oracle to develop ZeroFlap Optics for more reliable optical interconnect. Source: FY2026 10-K.

• FY2026 revenue was $1,335.1M and the 10-K says AEC ramp at hyperscale data center customers drove over 99% of the increase vs FY2025. (10-K)
• Q4 FY2026 revenue was $437.0M, gross margin 68.2%, ending cash + short-term investments $1.4B. (EX-99.1)
• Q1 FY2027 outlook is $465M–$475M revenue and 66.9%–68.9% GAAP gross margin. (EX-99.1)
• The 10-K says the top 10 customers were ~90% of FY2026 revenue and two customers were 10%+ each. (10-K)
• The 10-K explicitly names Microsoft and Oracle in product-collaboration language. (10-K)
• Reach collapse: passive DAC reach falls below ~1m at 224G/lane (1.6T); the AEC retimer's signal-integrity restoration to ~7m is the mechanism that keeps intra-rack/scale-out copper viable. (spec, inferred)
• Power delta (the binding cluster constraint): 800G optical pluggable ≈ 13–14W/port vs AEC ≈ 2–3W/port → ~10W saved per port. (inferred, industry spec)
• Cluster-level scaling: ~9–10 scale-out links per GPU server (vs 1–2 for CPU servers). A 100k-GPU cluster ≈ 12.5k servers × 9 = ~112k links → ~1.1 MW of interconnect power avoided = ~9.9 GWh/yr ≈ ~$1M/yr energy per 100k-GPU cluster before PUE (×1.3–1.5). Power is the hard cap on cluster size, so AEC power-efficiency is GPU-count enablement, not just opex.
• Cost-of-moat (DustPhotonics): $750M cash + ~0.92M shares (~$239M at ~$260) = ~$989M upfront, plus up to ~3.21M shares earnout (~$835M). Defends optical revenue guided >$500M FY2027 and a SiPho PIC TAM of $6B by 2030. Upfront ≈ 2.0× defended FY2027 optical revenue; full-potential ≈ 3.6×.

1) AEC still funds the print (10-Q MD&A, primary).

• Q1 FY2027 revenue $479.003M vs $223.074M year-ago = +$255.9M / +114.7%.
• MD&A: "a significant increase in the volume of unit shipment of AEC products which contributed over 90% of the increase" and was "primarily driven by the ramp-up of our AEC solutions at our hyperscale data center customers."
• AEC share of the increment ≥ 0.90 × $255.9M = ≥ $230.3M. Non-AEC (optical DSP / PIC / retimer / IP) ≤ $25.6M of the YoY increase.
• Implication: Dust is closed and in the balance sheet, but it is not yet the growth engine in the filed P&L. Optical as a second engine remains a FY2027 2H claim, not a Q1 measured result.

2) Cost of the moat vs the moat it defends (Dust, 10-Q Note 5).

• Purchase consideration $1,251.1M vs Q1 non-AEC increment ceiling $25.6M → the optical asset is being paid for ~49× this quarter's implied optical/other increment. Even against the April 8-K's previously guided >$500M FY2027 optical (not restated in this 10-Q or the Sep 1 EX-99.1), upfront+contingent is ~2.5× that guided year. Goodwill $895.8M / $1,251.1M = 71.6% of the price is synergy/going-concern, not identifiable PIC IP. Identifiable developed tech + IPR&D = $374.2M = 30% of consideration.
• Cash used in the acquisition, net of cash acquired: $735.6M. Cash + short-term investments fell from $1,443.3M (May 2) to $764.3M (Aug 1) — a $679.0M drop, matching EX-99.1.

3) GAAP vs product pricing power (the 65% bar).

• GAAP GM 64.5% (Q1 FY27) vs 67.4% year-ago (−290 bps). 10-Q: decrease "primarily driven by amortization of acquired intangible assets," not ASP cuts.
• EX-99.1 non-GAAP GM 68.0% (held vs Q4 FY26 68.3%). Q2 guide reconciling item: amort of acquired intangibles 2.8 pp of GM + SBC 1.3 pp = 4.1 pp GAAP→non-GAAP bridge.
• Read: product-level pricing power is intact at 68%; reported GAAP GM now prints below the 65% software-calibrated bar because of Dust amort. Do not treat 64.5% as a price-war signal.

4) Cash conversion / dilution of economic capture.

• GAAP NI $129.4M; OCF $90.2M (70% conversion); PPE capex $7.3M → FCF $82.9M.
• Share-based compensation $88.0M = 18.4% of revenue and ≈ 106% of FCF. Working-capital cash outflow $151.3M (inventory + AR).
• Inventory $313.1M vs $250.8M (+$62.2M). Days inventory on Q1 COGS: $313.1 / ($169.9 / 91) ≈ 168 days — a pre-build, not a lean fabless print.
• Foundry/vendor purchase commitments remaining FY2027: $212.7M (10-Q Note 9).

5) Concentration (two views, both 10-Q).

• Contracting parties (Note 3): Customer A 43% of Q1 revenue (50% year-ago), Customer B 28% (35%). Combined 71%. AR: Customer A 57%, B 28% = 85% of receivables.
• End-customer supplement (MD&A): C 28%, D 33%, E 13%, F 10% = 84% from four end customers (year-ago C+D+E = 88%, F <10%). A fourth 10%+ name appeared; the largest end customer is now D at 33% (stable YoY). Diversification is real at the margin, not a regime change.

6) Guide math (EX-99.1, not 10-Q).

• Q2 FY2027 revenue $525–$535M = +9.6% to +11.7% sequential vs $479.0M. Midpoint $530M.
• If FY2027 still compounds at the previously guided >80% on FY2026 $1,335.1M, implied year ~$2.40B. Q1+Q2-mid = $1.009B, which would require H2 ~$1.39B (~$696M/quarter) — i.e. a second-half inflection the 10-Q does not evidence. Treat H2 as an open question, not a filed fact.

\[

\text{Links} = 100{,}000 \times 2 = 200{,}000

\]

\[

\text{Short-reach links} = 200{,}000 \times 30\% = 60{,}000

\]

\[

\text{Credo-served links} = 60{,}000 \times 10\% = 6{,}000

\]

\[

\text{Revenue opportunity per cluster} = 6{,}000 \times 200 = 1{,}200{,}000

\]

Under these assumptions, the AEC revenue opportunity per 100,000-GPU cluster is only $1.2m. That appears small relative to Credo’s total reported revenue run rate, which suggests either the assumptions are too conservative, the addressable content per GPU is higher, or the real opportunity includes optical DSPs, SerDes IP, and retimers beyond AECs.

This spread demonstrates that the TAM input assumptions drive the output. Without a named-institution source for GPU port counts, short-reach attach rates, and Credo win share, this calculation remains speculative.

Use this table only as a model framework, not as verified source-backed data.

Licensing revenue =

2 × 10,000,000 × $0.30 + $1,000,000

Licensing revenue =

$6,000,000 + $1,000,000 = $7,000,000

Gross profit =

$7,000,000 × 90% = $6,300,000

Operating profit =

$6,300,000 − $2,000,000 = $4,300,000

# 6. Observed financial profile — orientation only

The following is included only to frame the direction of the company. It is not sourced to a verified document and must not be used in a conviction model as currently marked.

\[

\text{Breakeven revenue} = \frac{\$114.57\text{M}}{0.63} = \$181.86\text{M}

\]

This implies FY2024 revenue of $192.97M was only about 6% above the non-GAAP operating breakeven level under these assumptions.