Credo Beat Wall Street. Its Optical Pivot Raised the Stakes
Credo reported record quarterly revenue of $479 million, yet its shares fell after hours despite results exceeding Wall Street estimates.
That conflict defines the fiscal 2027 first-quarter earnings call held on September 1, 2026. Revenue rose 114.7% year over year, while non-GAAP net income increased 140% to $236.3 million. The company also projected another sequential revenue increase for the current quarter.
The call was not simply a celebration of another earnings beat. Management asked investors to accept a more demanding story. The business must expand from active electrical cables into optics without weakening margins or losing momentum in copper.
Broadcom, Marvell, and other connectivity suppliers already compete for spending inside AI data centers. Investors therefore need evidence that the optical portfolio represents durable expansion, not a temporary addition to a concentrated hyperscaler business.
Credo's Quarter Beat Estimates, but Expectations Were Higher
The quarter confirmed exceptional operating momentum, while the market response showed that historical growth no longer sets the standard.
The company reported revenue of $479 million for the quarter ended August 1. That result increased 9.6% sequentially and surpassed the upper end of management's previous guidance.
Its quarterly filing also reported GAAP net income of $129.4 million. GAAP diluted earnings reached $0.67 per share, while non-GAAP diluted earnings reached $1.20.
Non-GAAP gross margin held at 68%. Non-GAAP operating income reached $230.6 million, producing a 48.2% operating margin. Non-GAAP net margin was 49.3%.
These margins matter because connectivity hardware companies often sacrifice profitability while financing new products and production capacity. This company is expanding research spending while still converting nearly half of adjusted revenue into adjusted net income.
Management expects second-quarter revenue between $525 million and $535 million. At the midpoint, that outlook implies another sequential increase of roughly 11%.
The company also maintained its expectation for more than 85% revenue growth across fiscal 2027. It expects non-GAAP net margin near 50%, even as non-GAAP operating expenses rise approximately 55%.
Those figures would normally support a straightforward positive reading. However, shares fell in after-hours trading after already declining during the regular session.
The after-hours decline captured a familiar earnings-season problem. A company can exceed published estimates while still missing expectations embedded in its valuation and recent momentum.
Seven consecutive quarters of triple-digit year-over-year revenue growth have changed the comparison. Investors are no longer asking whether AI infrastructure spending benefits the company. They are asking how long its growth can remain far above the semiconductor industry.
The current outlook also places more weight on the second half. Management expects optical products to help create a pronounced revenue inflection later in fiscal 2027.
That timing introduces execution risk. The company must secure components, complete customer qualifications, and expand production before the expected revenue becomes reported revenue.
The quarter therefore changed the debate. The central question is no longer whether demand exists. It is whether a rapidly expanding optical portfolio can meet expectations that already assume strong execution.
Why the Credo Earnings Call Focused on Optics
Management positioned optics as the next major growth engine, not as a replacement for the cable business that created its current scale.
Active electrical cables, or AECs, remain the company's largest business. An AEC contains signal-processing electronics that extend reliable copper connections beyond the practical reach of passive cables.
These cables commonly connect accelerators to the first network switch across short distances. Management described the relevant segment as connections measuring roughly one to seven meters.
Copper has advantages within that range. It can consume less power than optical transceivers and remove lasers from connections where customers prioritize reliability.
The company now has relationships with five hyperscalers, according to CEO Bill Brennan. It is also expanding its engagement with neocloud operators, which provide specialized computing capacity for AI workloads.
However, larger clusters require several types of connection. Copper works across short reaches, while optical links become necessary as distance and bandwidth requirements increase.
Management's thesis is that no single medium, architecture, or protocol will serve every connection. Future clusters will mix copper and optics across scale-out networks, scale-up systems, storage, memory, and front-end infrastructure.
Scale-out connects more servers or accelerators through a broader network. Scale-up connects processors more tightly so they can operate on a workload as one larger computing system.
That distinction explains the optical expansion. The company wants to supply connectivity across distances measured from millimeters to kilometers, instead of defending one cable category.
Its optical portfolio now includes digital signal processors, silicon photonics chips, and complete ZeroFlap optical transceivers. An optical DSP converts and corrects high-speed signals inside a transceiver.
A silicon photonics photonic integrated circuit, or PIC, places optical functions on a semiconductor substrate. This integration can improve density and simplify high-volume optical systems.
ZeroFlap optics adds a system-level product. Management describes it as a highly reliable optical platform designed to reduce link failures that interrupt expensive AI computing workloads.
The company expects optical revenue to exceed $600 million in fiscal 2027. ZeroFlap optics, silicon photonics PICs, and optical DSPs should each contribute more than $100 million, according to management.
That forecast exceeds the optical target disclosed when the company announced its DustPhotonics deal in April. The earlier optical portfolio plan projected more than $500 million in fiscal 2027 optical revenue.
The raised figure indicates stronger confidence, broader product coverage, or both. It also increases the amount of revenue exposed to manufacturing schedules and customer deployment decisions.
Management said optical DSP revenue reached a record during the quarter. Existing deployments included 50-gigabit and 100-gigabit-per-lane products, while engagement continued around 200-gigabit-per-lane designs.
The company expects 800-gigabit ports to remain relevant during the transition toward 1.6-terabit connections. That overlap creates revenue opportunities across multiple generations, but it also complicates product planning.
Brennan told analysts that the business is adding optics from an established operating base. His message was that AEC growth continues while optics grows faster.
That is the favorable version of the story. The harder test is whether the two businesses reinforce each other without creating supply, integration, or margin problems.
Copper Versus Optics Is the Wrong Contest
The strategic reversal is that the company built its position with copper, then used that position to argue against choosing one physical medium.
Investors often frame data-center connectivity as copper versus optics. Faster port speeds and denser AI systems appear to favor optical links, particularly across longer reaches.
That framing can make the existing AEC franchise look vulnerable. If 1.6-terabit systems require more optics, a company associated with active copper cables might face slower growth.
Management rejected that binary view during the full call transcript. It placed copper and laser-based optics inside one expanding market for pluggable transceivers.
The argument depends on reach. AECs handle short connections where power and reliability can favor copper. Optical transceivers address longer links and configurations requiring greater density.
At 800 gigabits, management said copper replaced some optical connections between accelerators and their first switches. It expects AECs to remain relevant as ports transition toward 1.6 terabits.
The company plans AEC products reaching up to 6.5 meters at that speed. Initial fiscal 2027 contribution should arrive during the second half, with a larger ramp expected in fiscal 2028.
Meanwhile, its optical products target links where copper becomes less practical. This creates a portfolio based on distance, density, serviceability, and power rather than one medium.
The shift also changes the competitive map. Broadcom and Marvell supply networking silicon and optical DSP products, while module makers integrate components into finished transceivers.
Credo is moving in the opposite direction from a narrow component strategy. It wants to sell individual chips where appropriate and complete systems where customers value tighter integration.
That approach can increase the amount of company technology inside each AI cluster. It also requires competence across silicon, firmware, photonics, testing, manufacturing, and system qualification.
The DustPhotonics acquisition brought silicon photonics technology in-house. Completed on May 28, the DustPhotonics acquisition added products spanning 800 gigabits, 1.6 terabits, and planned 3.2-terabit systems.
Vertical integration can shorten development cycles and reduce dependence on external suppliers. It can also improve coordination between a PIC, its DSP, and the finished optical engine.
However, integration does not automatically guarantee better economics. Optical production introduces yields, packaging requirements, thermal constraints, and component availability beyond traditional chip design.
Management is also pursuing near-packaged optics, or NPO. This architecture places optical engines closer to a processor or switch, reducing the electrical distance traveled at high speeds.
Co-packaged optics moves that integration even closer by placing optical components in the same package as networking silicon. Both approaches address density and power constraints, but deployment models remain unsettled.
The company described itself as form-factor agnostic. It plans to follow customer requirements across pluggable, near-packaged, and eventually co-packaged designs.
That flexibility is sensible because hyperscalers do not share one architecture. It also means research spending must support several possible deployment paths before the market selects winners.
The primary opponent is therefore not another single semiconductor company. It is the gap between a broad portfolio promise and the operational reality of scaling several technologies together.
The Optical Ramp Carries Three Material Risks
Strong demand does not remove execution risk, especially when revenue depends on a few customers and a back-loaded production plan.
The first risk is customer concentration. Four end customers each represented at least 10% of quarterly revenue, according to the call.
The transcript appears to contain an error around the largest customer's percentage, but management clearly identified the other shares as 28%, 13%, and 10%. The inaccurate transcript figure should not be treated as verified.
The concentration remains substantial even without that disputed number. A delayed deployment by one large hyperscaler can affect quarterly revenue, inventory, and purchasing commitments.
The company's fiscal 2026 annual filing said its ten largest customers generated approximately 90% of annual revenue. Two customers individually represented at least 10%.
Management says the customer base is diversifying across hyperscalers, neoclouds, and other buyers. It also says product consumption is broadening inside major accounts.
Those developments help, but they do not eliminate platform concentration. Several nominal customers can still depend on similar accelerator architectures, deployment schedules, or capital-spending cycles.
The second risk is the optical production ramp. The fiscal 2027 forecast assumes a meaningful second-half increase from products that require qualification and high-volume manufacturing.
An analyst asked management directly about supply constraints, foundry costs, and the challenge of scaling ZeroFlap optics. Brennan said the company began preparing its supply chain 18 to 24 months earlier.
Inventory reached $313.1 million at quarter end, increasing $62.2 million sequentially. Management described that working-capital commitment as part of its preparation for higher volumes.
Building inventory can protect a ramp from component shortages. It can also create exposure when customer timing changes or a product transition moves faster than expected.
The third risk is the difference between GAAP and adjusted profitability. GAAP operating expenses were $188.4 million, compared with non-GAAP operating expenses of $95.2 million.
The reconciliation excludes share-based compensation, acquisition expenses, intangible amortization, and related tax adjustments. These exclusions are disclosed, but investors must decide which costs remain economically relevant.
GAAP gross margin declined to 64.5% from 68.2% in the previous quarter. Non-GAAP gross margin remained within management's expected range at 68%.
Some pressure was expected after the acquisition and product-mix changes. Still, optical systems can carry different economics from high-margin semiconductor components.
Management forecasts fiscal 2027 non-GAAP gross margin broadly consistent with fiscal 2026. Maintaining that level while complete optical products scale would support its vertical-integration thesis.
A material decline would weaken that thesis. It could indicate pricing pressure, production inefficiency, acquisition costs, or a less favorable product mix.
Cash flow offers another signal. Operating cash flow fell sequentially to $90.2 million, primarily because of working-capital changes. Free cash flow was $82.9 million after $7.3 million in capital spending.
The business ended the quarter with $764.3 million in cash and short-term investments. Its balance sheet provides room to finance inventory and research, but execution still determines the return on those investments.
Investors should therefore separate proven performance from forecasts. The reported AEC business, record DSP revenue, and current margins are observable. The full optical ramp and sustained fiscal 2027 profitability remain management expectations.
Inference Expands the Opportunity Beyond Networking
The most ambitious call takeaway was not another cable product, but a plan to connect processors with much larger pools of memory.
AI inference runs trained models to generate outputs. Its economics depend on accelerator utilization, memory capacity, memory bandwidth, and the cost of serving each request.
Large models can require more memory than one accelerator package provides. High-bandwidth memory offers speed, but packaging capacity and physical proximity limit expansion.
The company introduced OmniConnect as an answer to this memory fan-out problem. The proposed architecture combines licensed serializer-deserializer technology with external gearbox chips.
A serializer-deserializer, or SerDes, converts parallel data into a high-speed serial signal and then reverses that process. It moves information efficiently between chips or boards.
The first OmniConnect gearbox carries the name Weaver. It is designed to translate between an accelerator's embedded SerDes and lower-power DDR memory.
Management expects the first version to support LPDDR5, followed by LPDDR6. The gearbox approach could let an accelerator platform change memory generations without redesigning the main processor.
The initial announced customer is Positron, an AI hardware company developing inference systems. Management said Positron's first product targets up to two terabytes of memory.
These claims remain early. The call did not provide production revenue, benchmark validation, or broad customer adoption for OmniConnect.
Management expects initial revenue during fiscal 2028. It believes the technology can represent thousands of dollars in content per GPU, but that forecast has not been independently verified.
The opportunity is strategically important because it moves the company closer to the processor and memory subsystem. That market differs from its established role in network links.
It also creates a longer-reach growth path. Network connectivity benefits from expanding clusters, while memory connectivity addresses performance inside each inference server.
The company has contributed OmniConnect to the Open Compute Project. An open specification can encourage multiple vendors to build compatible products and reduce customer concerns about a proprietary interface.
However, an open contribution does not guarantee adoption. Accelerator developers must allocate chip area and power to the relevant SerDes, while memory and system designers must support the architecture.
Competition also comes from established memory technologies and custom hyperscaler designs. High-bandwidth memory will continue improving, while chipmakers are exploring several approaches to memory pooling and scale-up connectivity.
The immediate value of OmniConnect is therefore optionality. It offers a route into inference infrastructure without contributing to the current quarter's revenue.
That distinction matters. Investors should not use an early fiscal 2028 product concept to justify the execution of the fiscal 2027 optical plan.
The more credible near-term argument remains the combined copper and optical portfolio. OmniConnect becomes material only after design wins turn into production deployments.
Three Signals Will Decide Whether the Credo Strategy Holds
The next two quarters must convert product breadth into visible revenue without sacrificing the economics that made the growth story unusual.
The first signal is second-quarter revenue. Management guided to a range of $525 million to $535 million, with non-GAAP gross margin between 67% and 69%.
Revenue near or above the upper end would show continued demand before the expected second-half optical acceleration. A result below the range would challenge the timing behind the annual outlook.
Gross margin deserves equal attention. Stable adjusted margin would suggest that new products and acquisition integration are not overwhelming operating leverage.
A meaningful margin decline would raise questions about optical manufacturing, pricing, or product mix. GAAP results should be evaluated alongside management's adjusted measures.
The second signal is measurable optical contribution. Management expects more than $600 million during fiscal 2027, including over $100 million from each of three product groups.
Future calls should clarify how much revenue comes from optical DSPs, silicon photonics PICs, and ZeroFlap systems. Broad descriptions of customer engagement will become less useful as the fiscal year advances.
Customer qualifications are another leading indicator. Announced design wins matter most when management provides a production schedule and identifies whether the deployment affects fiscal 2027 or fiscal 2028.
The company already reported two next-generation optical design wins following the DustPhotonics integration. Their larger ramps are expected in fiscal 2028, with possible contribution late in fiscal 2027.
The third signal is concentration. Management expects three or four customers to remain above 10% of revenue during coming quarters.
The strongest outcome would combine more large customers with broader product purchases inside each account. That would reduce dependence on one platform without weakening relationships with existing buyers.
A weaker outcome would show revenue growth concentrated in one deployment or one optical product. Such growth can still look impressive, but its durability becomes harder to judge.
AEC performance must also remain visible. Management expects copper revenue to grow more slowly than optics because AECs start from a much larger base.
Slower growth is not the same as contraction. Continued AEC expansion would support the argument that optics adds a growth engine instead of replacing an aging franchise.
The September call established a demanding benchmark. Record results now matter less than proof that several product ramps can move together.
For readers tracking this story across transcripts, filings, and product announcements, a structured knowledge blending workflow can keep forecasts separate from reported results. Record each target with its date, then update it when the company publishes actual revenue.
Watch the next quarter for three answers: Did revenue reach guidance, did optical contribution become measurable, and did customer concentration improve? Those signals will show whether the selloff reflected excessive expectations or an early warning about execution.
