Evid Invest
EvidInvest
← Back to Blog

Credo's AI data-center connectivity thesis: what the filings confirm

·EvidInvest Team
AI infrastructureData center connectivityOptical interconnectsSEC filingsAetherCRDO

Credo's AI data-center connectivity thesis: what the filings confirm

Draft for EvidInvest review. This is financial research content, not investment advice. Built from the parent Aether/SEC evidence pack, SEC EDGAR filings and filed 8-K earnings exhibits, SEC companyfacts, and Yahoo Finance one-year price data. Live Aether financial/transcript search was quota/server-limited in the parent run, so the evidence hierarchy here treats SEC filings and SEC-filed exhibits as the primary source. The research did not confirm a SpaceX or Starlink customer, supplier, partner, revenue, or counterparty relationship for Credo; any SpaceX graph overlap remains unverified graph-proxy inference unless a separate source is added.

Thesis

Credo is not an AI accelerator story. It is a high-speed connectivity bottleneck story.

The filing-backed thesis is that AI data centers need faster, lower-power links between switches, servers, accelerators, memory, racks, and optical/electrical network layers. Credo's own filings place it directly in that stack: active electrical cables, optical transceivers, SerDes, DSPs, retimers, Ethernet, PCIe, and emerging 1.6T connectivity.

Based on Yahoo Finance chart data saved in the parent research pack, CRDO moved from $62.65 on 2025-06-02 to $226.10 on 2026-06-01, a measured one-year gain of roughly +260.9%. That move changes the research question. The question is not whether the market has noticed AI connectivity. It has. The better question is what Credo's own filings confirm after that rerating, and where the evidence stops.

The SEC-first evidence stack points to five conclusions:

  1. Credo has filing-confirmed AI/data-center connectivity exposure. Its 2026 Q3 Form 10-Q says its high-speed copper and optical interconnect products deliver power and performance at up to 1.6T to meet data infrastructure demands of AI.
  2. The product layer is connectivity, not compute. Credo describes AECs, optical transceivers, OmniConnect memory solutions, retimers, DSPs, SerDes, Ethernet, PCIe, and chiplets.
  3. Demand has scaled quickly. SEC-filed FY26 quarterly releases show revenue up more than 150% year over year in each of Q1 through Q4 FY26.
  4. The customer signal is powerful but concentrated. The filings disclose a Microsoft HiWire Switch AEC partnership, Amazon warrant/customer evidence, and very high customer concentration, but some large customer identities remain anonymized.
  5. SpaceX is not confirmed. The 2025 10-K mentions Starlink only as a broader low-earth-orbit connectivity market example, not as a Credo customer, supplier, partner, or revenue source.

That is the EvidInvest/Aether value: separating confirmed SEC facts from inferred supply-chain relationships and claims that still need a source.

What changed

The key change is that Credo's filings now describe the company as a supplier to the data infrastructure demands created by AI.

In the 2026 Q3 Form 10-Q, Credo says its mission is to transform connectivity at scale, and that its high-speed copper and optical interconnect products deliver industry-leading power and performance at up to 1.6T to meet the ever-expanding data infrastructure demands of AI. The same filing lists ZeroFlap Active Electrical Cables, ZeroFlap optical transceivers, OmniConnect memory solutions, and retimers and DSPs for optical and copper Ethernet and PCIe.

That language matters because it keeps the thesis specific. Credo is not claiming to sell GPUs. The filing-supported exposure is the connectivity fabric around AI infrastructure.

The 2025 Form 10-K provides the broader product map. Credo says its connectivity solutions are optimized for optical and electrical Ethernet and PCIe applications, including 100G, 200G, 400G, 800G, emerging 1.6T Ethernet, PCIe5, and upcoming PCIe6 markets. It says products are based on SerDes and DSP technologies and include integrated circuits, active electrical cables, and SerDes chiplets.

Evidence label: confirmed SEC fact for Credo's AI data-infrastructure language, product categories, 1.6T/Ethernet/PCIe framing, and copper/optical interconnect exposure. The conclusion that Credo benefits from AI data-center buildout is an inferred relationship supported by those filings.

Financial growth: the FY26 revenue ramp is the cleanest signal

Credo's recent reported numbers show why the stock rerated.

SEC-filed 8-K earnings exhibits show:

  • Q1 FY2026 revenue: $223.1M, up 274% year over year and 31% quarter over quarter.
  • Q2 FY2026 revenue: $268.0M, up 272.1% year over year and 20.2% quarter over quarter.
  • Q3 FY2026 revenue: $407.0M, up 201.5% year over year and 51.9% quarter over quarter.
  • Q4 FY2026 revenue: $437.0M, up 157.0% year over year and 7.4% quarter over quarter.

The SEC companyfacts data in the parent pack shows FY2025 revenue of $436.775M and nine-month FY2026 revenue of $898.113M through January 31, 2026. The Q4 FY26 8-K then adds another $437.0M quarter.

The margin profile also matters. The Q4 FY26 8-K reported 68.2% GAAP gross margin, 68.3% non-GAAP gross margin, $169.1M GAAP net income, and $226.7M non-GAAP net income. The Q3 FY26 8-K reported a similar 68.5% GAAP gross margin and 68.6% non-GAAP gross margin.

That is the upside evidence: rapid revenue growth with high gross margins in a product layer that filings tie to AI data infrastructure.

Evidence label: confirmed SEC fact for the revenue, growth-rate, margin, and net-income figures because they come from SEC-filed exhibits and SEC companyfacts. This draft does not include a valuation model, price target, or investment recommendation.

Customer and end-market signals: strong hyperscale evidence, but anonymized concentration

The strongest named customer/partner evidence is Microsoft.

Credo's 2025 10-K and 2026 Q3 10-Q say the company partners with Microsoft on its HiWire Switch AEC and open-source implementation to help realize Microsoft's dual-Top-of-Rack architecture and improve data-center deployment and reliability.

That is a confirmed SEC fact. The careful wording is important: Credo partners with Microsoft on HiWire Switch AEC. The retrieved evidence does not support a broader claim like "Microsoft is X% of Credo revenue" unless another source is added.

The filings also show Amazon-related customer evidence. The 2025 10-K discloses a warrant issued to Amazon.com NV Investment Holdings LLC. The 2026 Q3 10-Q says Amazon exercised all customer warrant shares, resulting in a net issuance of 3.8M shares, and that the warrant was no longer outstanding. The parent pack notes that earlier filing evidence described up to 4.1M shares at $10.74 per share.

That supports an Amazon/customer relationship, but it still requires discipline. The evidence does not identify which anonymized revenue customer is Amazon, and it does not support an unsourced AWS revenue percentage.

The concentration risk is unusually important:

  • The 2025 10-K says Credo's top ten customers accounted for approximately 90% of fiscal 2025 revenue.
  • One customer accounted for 67% of fiscal 2025 revenue.
  • The 2026 Q3 10-Q concentration table shows Customer A at 48% of Q3 FY26 revenue and 53% of nine-month FY26 revenue.
  • The same table shows Customer B at 39% of Q3 FY26 revenue and 31% of nine-month FY26 revenue.
  • Customer A, B, and C represented 57% / 22% / 11% of accounts receivable at January 31, 2026.

This is both the bull case and the risk case. Credo appears tied to very large data-center/hyperscale connectivity programs. But if a small number of customers change timing, design choices, qualification schedules, inventory levels, or procurement plans, Credo's reported revenue can move sharply.

Evidence label: confirmed SEC fact for the Microsoft partnership wording, Amazon warrant exercise, customer-concentration percentages, and hyperscale/data-center served-market language. Customer identities behind Customer A/B/C remain needs source unless separately verified.

Supplier-chain and dependency signals: Credo is fabless, concentrated, and scaling working capital

Credo's filings show a classic fabless semiconductor dependency stack.

The 2025 10-K says Credo exclusively used TSMC for semiconductor wafer production in fiscal 2025. It also uses Amkor and ASE for packaging, and KYEC and Sigurd for testing. The same filing flags global trade, tariff, and supply-chain constraints.

That supplier map matters for an AI infrastructure thesis. If demand accelerates faster than supply, Credo still depends on external foundry, packaging, and testing partners. If trade restrictions, Taiwan risk, packaging capacity, or test capacity become constraints, the revenue ramp can be affected even when end demand remains strong.

Working capital confirms the physical scale-up. SEC companyfacts in the parent research show inventory rising from $90.029M at 2025-05-03 to $207.958M at 2026-01-31. Accounts receivable rose from $162.144M to $243.213M over the same period. The Q3 FY26 10-Q cash-flow table shows inventories used $127.879M of operating cash and accounts receivable used $81.069M over the nine-month period.

Those numbers are not automatically negative. Inventory and receivables can rise because a company is scaling into demand. But they are exactly the operating signals Aether should track because they define whether revenue growth is converting cleanly into cash and whether customers are absorbing product as expected.

Evidence label: confirmed SEC fact for the TSMC/Amkor/ASE/KYEC/Sigurd supplier relationships, inventory/receivables figures, and cash-flow impacts. The interpretation that working-capital build is linked to scaling AI-connectivity demand is an inferred relationship.

Optical roadmap: Hyperlume expands the interconnect story

Credo also used M&A to extend its optical interconnect roadmap.

The 2026 Q3 10-Q says Credo acquired Hyperlume on September 29, 2025. Hyperlume is described as a developer of miniature LED, or microLED-based, optical interconnect technology for chip-to-chip communication. Total consideration was $92.0M. The filing says the acquisition was intended to expand Credo's end-to-end system-level connectivity solutions.

This is useful because it connects Credo's current AEC/optical transceiver/DSP/retimer story to a broader chip-to-chip and system-level optical connectivity roadmap.

Evidence label: confirmed SEC fact for the Hyperlume acquisition, technology description, consideration, and stated strategic purpose. Any specific revenue contribution, customer adoption timeline, or product-win impact remains needs source.

SpaceX graph overlap: label it as unverified

The task topic includes SpaceX graph overlap. The evidence boundary is straightforward.

The parent research found no SEC filing evidence that Credo is a customer, supplier, partner, revenue source, or contract counterparty of SpaceX or Starlink. The 2025 10-K mentions Starlink only as one example of disruptive low-earth-orbit satellite IP solutions in the broader connectivity market.

That gives us the correct labels:

  • Confirmed SEC fact: Credo sells high-speed copper and optical interconnect products for AI/data infrastructure, Ethernet, PCIe, and hyperscale/data-center markets.
  • Inferred relationship: Credo is thematically adjacent to broader high-speed connectivity demand, including networks and systems where AI, cloud, and low-earth-orbit connectivity all increase bandwidth requirements.
  • Needs source: any direct Credo-SpaceX supplier claim, Starlink sales claim, SpaceX partnership claim, or direct SpaceX purchasing benefit claim.

The draft should not convert graph proximity into a customer claim.

Risks and caveats

The filing-supported AI connectivity thesis is real, but the caveats are just as important:

  • Valuation/rerating risk: the measured one-year move was roughly +260.9%. A lot of optimism may already be embedded in the share price.
  • Customer concentration: the top ten customers were approximately 90% of fiscal 2025 revenue, and one customer was 67%.
  • Anonymous customer boundary: Customer A/B/C concentration is disclosed, but identities remain unconfirmed in the SEC evidence used here.
  • Program timing risk: large hyperscale programs can create sharp revenue ramps, but timing, qualification, design wins, and inventory digestion can change quarterly results.
  • Supply-chain risk: Credo relies on TSMC for wafers and on outsourced packaging/testing partners including Amkor, ASE, KYEC, and Sigurd.
  • Working-capital risk: inventory and receivables rose materially during the ramp, requiring monitoring.
  • SpaceX evidence boundary: no SpaceX or Starlink relationship was confirmed.
  • Forecast boundary: this draft does not include a target price, rating, or investment advice.

The strongest framing is not "CRDO is the next AI winner" or "CRDO is too expensive." The stronger framing is narrower: Credo has filing-confirmed AI data-center connectivity exposure and exceptional FY26 growth, but the stock has already rerated sharply and the filing evidence also shows customer concentration and supply-chain dependencies.

A simple upside/downside evidence frame

This draft does not include a fresh valuation model. Without that model, the cleanest publishing frame is an evidence boundary rather than a target price.

Upside evidence supported by filings:

  • Credo's filings explicitly connect its copper and optical interconnect products to AI data-infrastructure demands.
  • The product set maps directly to AI networking connectivity: AECs, optical transceivers, SerDes, DSPs, retimers, Ethernet, PCIe, and 1.6T links.
  • SEC-filed FY26 quarterly results show revenue growth above 150% year over year in every quarter from Q1 through Q4 FY26.
  • Q4 FY26 gross margin was above 68% on both GAAP and non-GAAP bases.
  • Microsoft HiWire Switch AEC is a named filing-confirmed partnership.
  • Amazon warrant/customer evidence supports major hyperscale relevance, while preserving identity/economics boundaries.
  • Hyperlume adds microLED-based optical chip-to-chip interconnect technology to Credo's system-level roadmap.

Downside evidence supported by filings:

  • Revenue is highly concentrated in a small number of customers.
  • Customer identities and exact economics for Customer A/B/C are not fully confirmed by the retrieved SEC evidence.
  • The stock's measured one-year move already embeds major AI-connectivity enthusiasm.
  • Inventory and accounts receivable rose materially during the demand ramp.
  • Wafer, packaging, and test dependencies create execution and geopolitical risk.
  • SpaceX/Starlink is not confirmed and should not be used as a catalyst without another source.

Evidence label: evidence frame, not investment advice and not a forecast. Any explicit upside/downside percentage, target price, or direct SpaceX catalyst needs separate sourcing and modeling before publication.

How Aether changes the workflow

A normal market search can summarize the "AI connectivity" narrative. Aether's SEC-first workflow makes the narrative auditable:

  • 10-K: product categories, served markets, customer concentration, Amazon warrant/customer evidence, Microsoft HiWire AEC partnership, manufacturing dependencies, and risk factors.
  • 10-Q: current customer concentration, working-capital movement, Amazon warrant exercise, Hyperlume acquisition, and updated product language around AI data infrastructure.
  • 8-K / earnings exhibits: quarterly revenue growth, margins, net income, and management-reported FY26 operating momentum.
  • SEC companyfacts: revenue, inventory, receivables, and comparable financial facts.
  • Market data: one-year stock performance, which changes the valuation question.

The parent research also records an important workflow caveat: live Aether MCP financial search was quota-limited and transcript search was temporarily unreachable in that run. The fallback evidence used here is SEC EDGAR/companyfacts plus Yahoo Finance chart data. That makes the evidence boundary more important, not less.

The useful output is not a hype label. It is a classification system:

  • Confirmed SEC fact: Credo's AI/data-infrastructure product language, Microsoft HiWire AEC partnership, Amazon warrant exercise, revenue/margin numbers, customer concentration, Hyperlume acquisition, and TSMC/OSAT manufacturing relationships.
  • Inferred relationship: Credo as a beneficiary of AI data-center capex through high-speed connectivity and hyperscale networking demand.
  • Needs source: SpaceX/Starlink relationship, exact identity of Customer A/B/C, AWS/Amazon revenue percentage, Nvidia/Meta/OpenAI customer claims, precise valuation upside/downside, and Hyperlume revenue timing.

Bottom line

Credo's AI story is strongest when it is kept specific.

The filings support a real AI data-center connectivity thesis: Credo sells the copper, optical, SerDes, DSP, retimer, Ethernet, PCIe, and AEC layers that help AI infrastructure move data at high speed and lower power. FY26 revenue growth and gross margins show that demand has translated into reported financial momentum.

But the same evidence defines the boundaries. CRDO has already rerated sharply, customer concentration is material, supply-chain dependencies are real, working capital is scaling, and the Aether/SEC evidence does not confirm a SpaceX or Starlink relationship.

My current evidence-grounded classification:

CRDO is a filing-confirmed AI data-center connectivity beneficiary, best understood as a high-speed copper/optical interconnect and networking silicon story rather than an AI accelerator story. The key research question is whether Credo can convert concentrated hyperscale connectivity demand into durable, diversified revenue after a very large stock rerating.

That is the gap EvidInvest and Aether are built to close: not headlines, but evidence boundaries.

Source notes

Fair Value Weekly

Get DCF breakdowns, fair value updates, and portfolio ideas for serious investors. No spam, no paywalled teasers.