RESEARCH MEMO · AI Infrastructure
AI Connectivity Becomes the Next Compute Bottleneck: CIEN Leads on Risk-Reward, CRDO on Growth
AI Infrastructure | Networking & Optical Comparative Research | 22 September 2026
Companies covered: Ciena (NYSE: CIEN) · Credo Technology (Nasdaq: CRDO) · Astera Labs (Nasdaq: ALAB) · Coherent (NYSE: COHR) · Arista Networks (NYSE: ANET)
AI connectivity is moving from supporting equipment to a core constraint on cluster utilization. CIEN offers the more balanced risk-reward profile, CRDO the greatest growth torque, COHR diversified optical picks-and-shovels exposure, ALAB excellent business quality with less room for expectation error, and ANET the mature quality benchmark. Research priority: CIEN ≈ CRDO > COHR > ALAB; this is not a valuation-blind buy ranking.
A durable growth and margin reset could support structural re-rating
Copper and optical exposure can raise content, but concentration must decline
A high-margin network platform proves system-layer value, with wider recognition
Constructive industry trend | CIEN and CRDO prioritized | Valuation discipline still required
01 · RESEARCH BRIEF
The one-minute brief
AI connectivity is moving from standardized 800GbE into a 1.6T product cycle, while 1.6TbE standards work remains underway. Market references to “3.2T” may describe aggregate chip bandwidth, a future module road map or system capacity; they do not all mean volume deployment of 3.2TbE ports. As GPU counts rise, data movement, power, reliability and network architecture increasingly determine whether expensive compute is fully utilized. The five companies sit at different layers: ALAB is closest to in-rack connectivity, CRDO spans copper and optical links, COHR supplies lasers, components and modules, CIEN provides coherent transport between data centers, and ANET operates at the mature Ethernet systems layer. The investment question is not which company looks most like an optical vendor. It is which company can turn rising connectivity content into revenue, margins and free cash flow. Industry direction, company execution and security price remain separate questions.
Audio transcript
Our conclusion is that AI connectivity is becoming the next compute bottleneck, but the five companies sit at different layers. CIEN offers the more balanced risk-reward profile, subject to sustained growth, margin and cash-flow delivery. CRDO has the greatest growth torque, but customer concentration must decline. COHR is the broader optical picks-and-shovels exposure, ALAB combines excellent business quality with higher expectations, and ANET remains the mature systems-layer quality benchmark. Our research priority is CIEN roughly equal to CRDO, ahead of COHR and ALAB. This is not a valuation-blind buy ranking.
Known facts and open questions
- Confirmed
- Latest company results show firm AI networking or optical demand, with product road maps moving toward 1.6T.
- Not directly comparable
- Fiscal periods, product scopes and non-GAAP definitions differ; revenue growth is not a uniform quality ranking.
- Still to prove
- 1.6T standardization and adoption, the different meanings of 3.2T, packaged-optics deployment, and whether orders convert into durable margins and cash flow.
Scale-up
High-speed links among GPUs, memory and accelerators; ALAB and CRDO are key names.
Scale-out
Data movement across server racks; CRDO, ANET and COHR are key names.
Scale-across
High-capacity transport across campuses, metros and data centers; CIEN is the focus.
Optical components
Lasers, digital signal processors, modules and silicon photonics; COHR and CRDO are key names.
| Company | Position | Current view | Key evidence | Main risk |
|---|---|---|---|---|
| CIEN | Coherent transport and data-center interconnect | Best balanced risk-reward | Accelerating revenue and 2029 margin targets | Ambitious targets, customer concentration and rising expectations |
| CRDO | Active cables, SerDes, DSP and optics | Highest growth torque | Revenue up 114.7% year over year with strong margins | Customer concentration and volatility |
| ALAB | In-rack connectivity and switching | High quality, higher price hurdle | Revenue up 104% with standout margins | More future growth embedded in expectations |
| COHR | Lasers, optical components and modules | Diversified optical picks and shovels | AI data-center demand supports growth | Lower margins and complex mix |
| ANET | Ethernet network platform | Mature quality benchmark | Strong revenue and operating margins | Broad recognition and customer concentration |
02 · THESIS → EVIDENCE → UPDATE
What changed in the thesis?
The bottleneck is shifting from chip count to data movement
- Prior thesis
- AI infrastructure research focused mainly on GPU supply, available compute capacity and power constraints.
- New evidence
- The migration from 800G to 1.6T, preparation for 3.2T and rising inter-campus demand show bandwidth and power becoming system constraints.
- Updated view
- Networking and optical connectivity now deserve core coverage, but scale-up, scale-out, scale-across and components must be analyzed separately.
Industry growth does not create equal investment returns
- Prior thesis
- Expanding connectivity demand can lift revenue across the chain.
- New evidence
- The five companies differ sharply in margins, customer concentration, capital intensity, product purity and market expectations.
- Updated view
- Prioritize companies where revenue, margins and cash improve together; concentration or high expectations can dilute the value of fast growth.
03 · EVIDENCE & ANALYSIS
Evidence and analysis
01｜Ciena: The more balanced risk-reward profile
Ciena reported fiscal third-quarter 2026 revenue of $1.67 billion, up 37%, and adjusted EPS of $2.11, up 215%. It raised fiscal 2026 revenue guidance to about $6.42 billion and set fiscal 2029 targets of roughly 30% revenue CAGR, about 50% adjusted gross margin and around 20% free-cash-flow margin. Those long-term figures are management targets, not achieved results. Two large customers accounted for 41.7% of revenue in the same quarter, so concentration belongs in the risk case. The key test is whether hyperscaler demand produces sustained improvement in growth, margins and cash. WaveLogic 6 Extreme supports single-carrier 1.6Tb/s transmission, reinforcing its role in data-center interconnect and metro coherent optics.[1][2][3]
02｜Credo: The highest growth torque, with concentration as the constraint
Credo reported fiscal first-quarter 2027 revenue of $479 million, up 114.7% year over year and 9.6% sequentially, with a 68.0% non-GAAP gross margin. Its next-quarter revenue guide was $525 million to $535 million. Credo spans active electrical cables, SerDes, DSP and optics, allowing it to raise content across copper and optical routes. However, its top ten customers accounted for about 90% of fiscal 2026 revenue. Customer diversification is the critical test of durability.[4][5]
03｜Astera Labs: Closest to the compute platform, with a higher execution bar
Astera Labs reported second-quarter 2026 revenue of $392.4 million, up 104% year over year and 27% sequentially, with a 73.7% non-GAAP gross margin. Third-quarter revenue guidance was $540 million to $560 million. It has disclosed a sampling-stage 16×200G, 3.2T aggregate signal-conditioning product and demonstrated 1.6T Ethernet solutions; this does not mean 3.2T ports are in volume production. Interoperability testing is not the same as a customer order. When expectations are high, production timing, revenue delivery and margins all need to hold.[6][7][8]
04｜Coherent: The broader optical-infrastructure beneficiary
Coherent reported fiscal fourth-quarter 2026 revenue of $2.05 billion, up 34% on the company's reported basis, with a 40.2% non-GAAP gross margin. Management argues that AI data-center architecture is shifting connectivity from copper toward optics; that is a management view, not evidence that the industry migration is complete. COHR spans lasers, components, modules and manufacturing. A further re-rating requires AI data-center growth and margin improvement to arrive together.[9]
05｜Arista Networks: The quality benchmark at the systems layer
Arista reported second-quarter 2026 revenue of $3.036 billion, up 37.7%, with GAAP and non-GAAP operating margins of 45.4% and 49.9%. It also introduced 1.6Tbps AI fabric platforms. ANET shows that the Ethernet systems layer can produce high margins, software-like stickiness and strong execution. It is best used as the mature quality benchmark, while recognizing that its strengths are already widely understood.[10]
04 · INVESTMENT IMPLICATIONS
Industry and asset implications
Research priority
CIEN and CRDO lead: one for structural growth and margin re-rating, the other for growth and customer broadening.
Different roles
COHR offers diversified optical exposure, ALAB represents high quality with high expectations, and ANET defines the network-platform quality ceiling.
Price discipline
A constructive industry trend cannot replace valuation work. Higher research priority is not an instruction to chase short-term strength.
The highest-value evidence is not a product announcement. It is volume deployment, broader customers and eventual improvement in margins and free cash flow.
05 · VALIDATION & RISKS
What to verify next
CIEN | Structural re-rating
Track hyperscaler revenue, margins and free cash flow.
Failure signal: Revenue growth fails to improve margins and cash.
CRDO | Growth quality
Track optical growth, major-customer count and concentration.
Failure signal: Growth remains dependent on a few customers or margins erode.
ALAB | Production proof
Track the fabric-switch ramp and delivery against third-quarter guidance.
Failure signal: Programs slip or high growth no longer supports margins.
COHR | Profit conversion
Track AI data-center revenue, capacity and gross margin.
Failure signal: Demand grows without sustained margin improvement.
ANET | Platform quality
Track 1.6T adoption, AI network revenue and operating margins.
Failure signal: Competition or architecture shifts durably compress margins.
What would change our view?
The thesis would weaken if AI capital spending falls broadly, the 1.6T and 3.2T migration is materially delayed, or new packaged-optics architectures change the value of existing products faster than expected. Company-specific risks include customer concentration, premature capacity expansion, supply constraints and valuation compression.
06 · FAQ
Key questions
Why is ANET not ranked first?
ANET has outstanding business quality, but its strengths are widely recognized. This report compares incremental research value and risk-reward, not company quality alone.
Does optical upgrading mean copper disappears soon?
No. Distance, power, cost and reliability favor different technologies, so copper and optics should coexist. The key is whether a company can add content across technology routes.
Why can CIEN and CRDO both rank first?
They serve different roles. CIEN offers structural re-rating and more balanced risk-reward; CRDO offers greater growth torque with higher concentration and volatility.
Does a strong industry outlook mean all five stocks are cheap?
No. Industry direction, company execution and security price are three separate questions that require separate proof.
07 · TERMS & SOURCES
Terms, sources and related research
Key terms
- Scale-up
- High-speed connectivity among GPUs, memory and accelerators within a rack or compute domain.
- Scale-out
- High-speed networking across multiple server racks.
- Scale-across
- High-capacity links between campuses, metros and data centers.
- Active electrical cable
- A copper cable with embedded signal processing to extend reach and reliability.
- SerDes
- High-speed serializer-deserializer technology underlying chip and network connections.
- Coherent optics
- Optical technology designed for high-capacity transmission over longer distances.
- Pluggable, near-packaged and co-packaged optics
- Architectures that progressively move optical components closer to switch silicon, trading off serviceability, power and density.
- 800G, 1.6T and 3.2T
- Rate or bandwidth labels whose meaning depends on the product layer. They may refer to a port, module, coherent wavelength or aggregate chip bandwidth and are not directly comparable.
[1] Ciena | Fiscal third-quarter 2026 results ↗
[2] Ciena | Fiscal 2029 financial targets ↗
[3] Ciena | WaveLogic 6 Extreme 1.6T data sheet ↗
[4] Credo | Fiscal first-quarter 2027 results ↗
[5] Credo | Fiscal 2026 annual report ↗
[6] Astera Labs | Second-quarter 2026 results ↗
[7] Astera Labs | Taurus 3.2T product announcement ↗
[8] Astera Labs | Interoperability lab ↗
[9] Coherent | Fiscal fourth-quarter 2026 results ↗
[10] Arista Networks | Second-quarter 2026 results ↗
[11] IEEE 802.3 | 800GbE standard ↗
[12] IEEE 802.3 | 1.6TbE standards work ↗
[13] Open CPX | Pluggable packaged-optics specification progress ↗
The data come from different fiscal quarters, and GAAP and non-GAAP definitions are not fully comparable. Ciena's 2029 figures are company targets, not achieved results; the copper-to-optics shift is Coherent management's view. This public research is for education and is not personalized investment advice.