Credo SRMO Optics Standard for AI Data Centers 2026

Credo Technology Group Holding Ltd (NASDAQ: CRDO) said on October 12, 2026, that it is launching a new standards effort aimed at one of the thorniest wiring problems in modern AI infrastructure: how to connect tens of thousands of accelerators inside a single data center without drowning the system in copper, heat, and power loss. The San Jose-based connectivity company announced the Short-Reach Micro-Emitter Optics Workstream, a project housed inside the Open Compute Project Community that will attempt to define a new class of high-speed interconnect built specifically for AI data-center scale-up and scale-out applications.

The announcement, distributed through Business Wire’s product and technology newsroom and picked up by outlets including Street Insider, StockTitan, and MarketBeat, pulls together seven named component and materials suppliers around a single technical goal: a common, standardized way to move light-based signals across short distances inside AI clusters. Credo has not published pricing, shipment dates, or channel-count specifications tied to the new workstream. What it has published is a roadmap, a partner list, and a stated target of 2027 for approval by the Open Compute Project, often shortened to OCP.

Google · Preferred Sources

Don't miss new tech stories on Google

Add TrendinTech once in the Google app and our stories appear in your news suggestions.

Add Now

Credo Unveils the SRMO Workstream for AI Data Centers

According to the companies’ joint statement, the Short-Reach Micro-Emitter Optics Workstream, or SRMO, is designed to produce a formal specification rather than a shipping product. Credo and its partners are proposing the work inside OCP because OCP specifications tend to become reference designs that hyperscalers and their suppliers build toward collectively, instead of every vendor shipping a slightly incompatible version of the same idea. That matters for AI data centers because accelerator clusters today increasingly mix hardware from multiple vendors, and a fragmented interconnect landscape raises costs for everyone building at scale.

Credo frames SRMO as addressing the connectivity layer that sits closest to the compute itself: the short links between accelerators, switches, and memory inside a rack or a small cluster of racks, as opposed to the longer links that connect clusters across a data hall. The company’s announcement ties the effort directly to AI workloads that have moved from single-digit thousands of accelerators to tens of thousands operating as one coordinated system, a scale that existing short-reach interconnect approaches were not originally built to serve.

What Is Short-Reach Micro-Emitter Optics?

The name itself describes the technical bet. Micro-emitter optics refers to arrays of extremely small light sources, built at a scale that lets many individual optical channels sit side by side on a single chip or module rather than relying on one laser per fiber pair. Short-reach signals that these components carry over distances measured in meters inside a rack, not the kilometers covered by telecom-grade optics. Combining the two concepts into a formal specification is the core of what the SRMO Workstream is chartered to do.

Credo’s own announcement lists the technical ground the specification is meant to cover: link power, efficiency, signaling, parallelism, physical form factors, module and connector footprint, link budget, and compliance. That is a fairly complete list of what any interconnect standard needs to define before hardware vendors can build interoperable products against it, which signals that Credo and its partners are trying to establish SRMO as a full reference specification rather than a narrow point solution.

Why Credo Chose the Open Compute Project

The Open Compute Project began as a Facebook-led effort to open-source data center hardware designs and has since grown into the venue where hyperscalers, chipmakers, and component suppliers hash out shared infrastructure standards, from server chassis to rack power delivery. Routing SRMO through OCP rather than developing it as a proprietary Credo specification is itself a strategic signal. It tells potential customers, most of them hyperscale cloud operators building AI infrastructure at enormous scale, that no single vendor will control the specification once it is finalized.

That structure also lowers the bar for broad industry participation. Component suppliers who might be wary of designing proprietary silicon or connectors around one company’s roadmap are more willing to commit engineering resources to an open specification that any OCP member can eventually build against. It is the same logic that has driven past OCP workstreams around rack power and liquid cooling, both of which became reference points well beyond the companies that started them.

Inside the Draft Specification: What SRMO Will Define

Because Credo has not released a draft document publicly, the exact numbers behind link budget, signaling rates, or connector pin counts remain unconfirmed. What the announcement does confirm is the category of problems the workstream intends to solve before it submits anything to OCP for a vote. Link power and efficiency sit at the top of that list, reflecting an industry-wide concern that AI data centers are running into power ceilings well before they run into compute ceilings. A short-reach optical standard that trims watts per gigabit at the rack level has a direct line to lowering total data-center power draw, which has become the binding constraint for new AI infrastructure buildouts, a dynamic already playing out around disputes like FERC’s recent approval of PJM’s data center registry.

Signaling and parallelism point toward how SRMO intends to scale bandwidth: by running many optical channels in parallel at moderate per-channel speeds rather than pushing a smaller number of channels to extreme individual data rates. Physical form factor, module footprint, and connector footprint are the mechanical side of the same equation, since a specification that different vendors can actually manufacture to needs agreed dimensions, tolerances, and mounting approaches. Compliance, the final listed topic, is what eventually lets a buyer certify that a module from one vendor will interoperate with a switch or accelerator from another.

The Seven Companies Building the Standard

Credo named seven participating companies alongside itself: AFL, ams OSRAM, BizLink, Foxlink, Heraeus Covantics, SENKO, and YOFC. The roster spans optical fiber and cabling specialists, connector manufacturers, and photonics component makers, which lines up with the breadth of the technical scope Credo laid out. The table below summarizes what is confirmed about each participant’s role in the workstream based on the companies’ own statements.

Company	Industry Focus	Confirmed Role in SRMO
Credo Technology Group	High-speed connectivity silicon and optics	Workstream originator; provided named executive quote on standard rationale
ams OSRAM	Photonics and optoelectronic components	Advancing micro-emitter, micro-receiver, and 3D-packaging technology for the standard
SENKO Advanced Components	Fiber-optic connectors	Leading work on a common connector interface as channel counts scale
AFL	Fiber optic cable and connectivity	Named participant; specific technical role not detailed in the announcement
BizLink	Cable and interconnect manufacturing	Named participant; specific technical role not detailed in the announcement
Foxlink	Connector and cable assemblies	Named participant; specific technical role not detailed in the announcement
Heraeus Covantics	Precision materials and photonics	Named participant; specific technical role not detailed in the announcement
YOFC	Optical fiber and cable manufacturing	Named participant; specific technical role not detailed in the announcement

The breadth of that list, covering everything from raw fiber manufacturing at YOFC to connector engineering at SENKO and Foxlink to photonics packaging at ams OSRAM, suggests Credo is trying to line up the full supply chain before the specification is finalized rather than publishing a standard and hoping component makers follow later. That sequencing mirrors how other successful OCP workstreams have operated, where supplier buy-in during the drafting phase tends to predict how quickly a standard actually ships in real hardware.

An Unnamed Hyperscaler Joins the Effort

Reporting on the launch also references a leading hyperscaler participating in the workstream, though its identity has not been disclosed in Credo’s materials or in subsequent coverage. The involvement of an unnamed cloud-scale operator is notable even without a name attached, because OCP workstreams gain credibility and momentum almost entirely from hyperscaler backing. Standards that lack a committed buyer tend to stall at the draft stage; standards with at least one hyperscaler actively shaping requirements tend to reach production silicon.

It is worth being precise about what is and is not confirmed here: no outlet covering the announcement has named the hyperscaler, and nothing in Credo’s statement specifies whether that company is one of the handful that already dominate AI infrastructure capital spending. Readers should treat the hyperscaler’s identity as an open question rather than assume it refers to any specific company, including the ones most often associated with large-scale AI clusters.

Executives Weigh In on the SRMO Launch

Credo’s own executives framed the launch around the scaling problem that AI clusters now face. L.K. Bhupathi, Associate Vice President of Product at Credo, said that as AI workloads scale from thousands to tens of thousands of accelerators, the industry needs a standardized approach to wide, parallel optical connectivity, a comment reported by Street Insider that gets at the central argument for why SRMO exists at all: ad hoc, vendor-specific optical links do not scale cleanly once accelerator counts move into five digits.

Ashkan Seyedi, Vice President and General Manager of Digital Photonics Interconnect at ams OSRAM, described the company’s work advancing micro-emitter, micro-receiver, and 3D-packaging technologies as part of its broader AI-data-center-connectivity strategy, positioning SRMO participation as a natural extension of research the company was already pursuing. Obie Roy, Vice President of Strategic Commercial Development at SENKO Advanced Components, said the workstream aims to establish a common connector interface as channel counts scale into the thousands, a framing that puts the connector problem, not just the optics, at the center of why a standard is needed now rather than later.

Scale-Up vs Scale-Out: Two Different Optical Problems

Credo’s announcement explicitly covers both scale-up and scale-out applications, and the distinction matters for understanding why one standard is trying to address both. Scale-up connectivity links accelerators to each other tightly, often within a single rack or a few adjacent racks, to behave as one large virtual processor. Scale-out connectivity links groups of those racks together into the broader cluster fabric that coordinates thousands of nodes running a single training or inference job. Historically, these two problems have been solved with different technologies: scale-up links lean on very low latency, often electrical or short optical connections, while scale-out links have already moved largely to pluggable optical transceivers because the distances involved exceed what copper can handle efficiently.

SRMO’s bet is that micro-emitter arrays, built with enough channel density and low enough power draw, can serve both jobs with one underlying optical approach rather than two separate technology stacks. If that holds up through the specification process, it would simplify sourcing and qualification for hyperscalers that currently have to manage distinct supply chains for scale-up and scale-out optics. That simplification is also precisely the kind of efficiency gain that cloud providers have been chasing as AI infrastructure spending has ballooned, a trend visible in how aggressively providers compared in the AWS, Azure, and GCP GPU pricing wars are racing to cut the cost of every layer of the stack beneath their accelerators.

From Copper to Light: A Short History of Data Center Interconnects

Data center interconnects have moved through fairly distinct generations. Copper backplanes and direct-attach cables dominated server-to-switch connections for years because copper is cheap and simple to manufacture, but copper’s signal loss increases sharply with distance and speed, which eventually forces a move to optics as data rates climb. Pluggable optical transceivers became the default for longer links inside data centers as Ethernet speeds moved through successive generations, with the industry’s current wave of deployments built around 800-gigabit-class modules at the higher end of hyperscale networks.

More recently, the industry has been experimenting with co-packaged optics, which places optical engines directly next to switch silicon instead of in a separate pluggable module, chasing lower power draw at the cost of more complex manufacturing and serviceability. SRMO enters at a moment when AI accelerator clusters have made the short-reach segment, the links inside and immediately around a rack, the new bottleneck, a problem that did not exist at the same scale when earlier optical standards were written for general-purpose cloud and enterprise networking rather than tightly coupled accelerator fabrics.

How SRMO Compares With Other Scale-Up Approaches

SRMO is not the only effort trying to solve AI scale-up connectivity. Proprietary interconnect fabrics from major accelerator vendors already link GPUs within a server or rack using high-speed electrical or short-optical links, and a separate open consortium effort has been pursuing a vendor-neutral scale-up link specification of its own. Credo’s SRMO Workstream differs from both in scope: rather than defining an entire switching and protocol fabric, it is narrowly focused on the physical optical layer, the micro-emitters, connectors, and link budgets that any higher-level fabric would eventually run across.

Approach	Primary Use Case	Governance Model	Status as of October 2026
Proprietary GPU-vendor interconnect fabrics	Scale-up, within rack	Single-vendor controlled	Shipping in production accelerator systems
Vendor-neutral scale-up link consortiums	Scale-up, within rack	Multi-company industry consortium	Specification in active development
Pluggable optical transceivers (800G-class)	Scale-out, between racks and halls	Established multi-source agreements	Widely deployed in hyperscale networks
Co-packaged optics initiatives	Scale-out and switch-adjacent links	Multiple industry alliances	Early deployments and pilots underway
Credo SRMO Workstream	Scale-up and scale-out, short reach	Open Compute Project Community	Workstream launched; targeting 2027 OCP approval

What stands out in that comparison is governance. SRMO’s open, OCP-based structure sets it apart from single-vendor interconnect fabrics, while its narrow focus on the optical physical layer sets it apart from broader fabric-level consortium efforts. Whether the market needs both a protocol-level standard and a physical-layer optics standard, developed in parallel by different groups, is itself an open question that will likely only resolve once hyperscalers start specifying which combination of standards they intend to actually deploy.

Market Impact: What SRMO Means for CRDO and Suppliers

For Credo, leading a standards workstream carries a different kind of market value than simply shipping a new product. Companies that author the reference design for an eventual OCP specification often end up with a first-mover advantage in supplying compliant components once the standard is finalized, since their engineering teams understand the specification’s edge cases before competitors do. That dynamic has played out before in networking, where companies that led standards efforts captured outsized shares of the first wave of compliant hardware sales.

For the named component suppliers, participation signals where they expect AI infrastructure demand to concentrate over the next two to three years. ams OSRAM’s framing of SRMO as part of its AI-data-center-connectivity strategy and SENKO’s focus on connector interfaces built for channel counts scaling into the thousands both point toward suppliers positioning balance sheets and R&D budgets around sustained AI data center capital spending, a backdrop that has already reshaped unrelated supply chains, including battery and materials makers like Taiwan’s ProLogium, which recently pivoted a large share of its sales toward AI data centers as end markets shifted. Credo itself has not disclosed financial projections tied to SRMO, and no pricing for any resulting compliant hardware has been announced.

The Road to 2027: OCP’s Approval Timeline

Credo’s stated target is OCP approval in 2027, which places SRMO on a roughly one-year runway from launch to formal ratification, assuming the workstream stays on schedule. OCP specifications typically move through internal draft review, member feedback cycles, and a formal community vote before they are published as approved standards, a process that can compress or stretch depending on how much consensus exists among participating companies going in. The breadth of SRMO’s initial partner roster, spanning fiber, connector, and photonics specialists, suggests the group entered the process with meaningful technical alignment already in place, which tends to shorten the runway compared with workstreams that start from a blank page.

Milestone	Timing	Confirmed Detail
SRMO Workstream launch announcement	October 12, 2026	Announced from San Jose, California, with seven named partners
Draft specification development	Late 2026 into 2027	Covers link power, efficiency, signaling, parallelism, form factors, and compliance
Hyperscaler input incorporated	Ongoing	At least one unnamed leading hyperscaler is participating
Target OCP specification approval	2027	Stated goal in Credo’s announcement
First SRMO-compliant hardware	Not disclosed	No shipment or availability date announced

That last row matters as much as the others. A 2027 approval target for the specification itself says nothing definitive about when buyers can actually order SRMO-compliant modules. Standards bodies typically ratify a specification well before any vendor ships volume-production hardware against it, so readers should expect a further gap, likely stretching into 2028, between OCP approval and broad commercial availability.

Why This Fits a Bigger AI Infrastructure Pattern

SRMO’s launch lands in the middle of a broader AI infrastructure story that has dominated technology headlines through 2026: data centers straining against power, water, and physical space constraints as accelerator deployments keep climbing. Utilities and grid operators have spent much of the year adjusting policy to keep pace with the FERC-approved PJM data center registry mentioned above, while operators have also faced scrutiny over resource use detailed in reporting on produced water storage risk at AI data center sites, and over Washington’s own AI oversight staffing covered in the White House’s absorption of three CAISSI AI officials. An optics standard that reduces power draw per link does not solve those pressures on its own, but it chips away at one input in a system where every watt saved at the interconnect layer is a watt available for additional compute.

The timing also tracks with how quickly frontier AI models have driven demand for exactly the kind of tightly coupled accelerator clusters SRMO targets. Model releases like Meta’s Muse Wonder, which reached frontier-level performance in just 14 months of development, depend on training runs that increasingly span tens of thousands of accelerators acting in concert, the same scale threshold Credo’s own executives cited as the reason SRMO needs to exist. Infrastructure standards tend to lag the workloads that demand them by a year or more, and SRMO’s 2027 target fits that familiar pattern.

Five Predictions for AI Data Center Optics

More named participants before ratification. OCP workstreams typically attract additional member companies once a draft specification begins circulating, so expect the SRMO partner list to grow beyond the seven companies named at launch before any 2027 vote.
The unnamed hyperscaler gets identified informally. Hyperscaler involvement in OCP efforts rarely stays confidential for long once draft documents start circulating among member companies, even without an official announcement naming the company.
Connector standardization becomes the most contested technical fight. Based on SENKO’s stated focus on a common connector interface, the physical connector footprint, not the optics themselves, is likely to be where competing vendor interests collide hardest during the drafting process.
Parallel standards efforts will need to reconcile. With a separate vendor-neutral scale-up link consortium already active, expect pressure from hyperscaler members, who tend to sit on multiple standards bodies at once, to ensure SRMO’s physical layer and other groups’ protocol layers end up compatible rather than competing.
Compliant hardware announcements will trail OCP approval by at least a year. Given the gap between standards ratification and volume shipment seen in past OCP efforts, commercial SRMO-compliant modules are unlikely to reach broad availability before 2028 even if the 2027 approval target is met on schedule.

What to Watch Next From Credo and OCP

Credo has a near-term opportunity to add technical detail to the SRMO story at industry events, and the company has already signaled plans to showcase related high-speed connectivity products, reported separately by StockTitan, at upcoming AI infrastructure trade shows. Those venues are the likely place where Credo and its partners will start filling in the specifics the launch announcement left open: expected link budgets, target channel counts, and a clearer picture of how SRMO interacts with existing scale-up and scale-out architectures already deployed in production. Until a draft specification circulates publicly through OCP, every number beyond what the companies have confirmed should be treated as undecided rather than assumed.

Investors watching NASDAQ: CRDO will likely treat the SRMO launch as a credibility signal rather than a near-term revenue event, given that Credo disclosed no pricing or financial guidance tied to the workstream. The more direct read-through, according to coverage from MarketBeat’s tracking of Credo’s stock news, is that the announcement reinforces Credo’s positioning as a connectivity vendor embedded in the standards-setting process for next-generation AI infrastructure, a position that tends to pay off commercially only once the underlying specification reaches the market in shipped silicon.

Frequently Asked Questions

What is Short-Reach Micro-Emitter Optics (SRMO)?
SRMO is a workstream launched by Credo Technology Group and seven partner companies inside the Open Compute Project Community, aimed at developing a standardized specification for short-distance optical interconnects used in AI data-center scale-up and scale-out applications.

Who announced the SRMO Workstream?
Credo Technology Group Holding Ltd (NASDAQ: CRDO) announced the workstream on October 12, 2026, from San Jose, California, alongside named participants AFL, ams OSRAM, BizLink, Foxlink, Heraeus Covantics, SENKO, and YOFC.

What is the Open Compute Project?
The Open Compute Project is an industry community, originally started by Facebook, that develops open, vendor-neutral hardware and infrastructure specifications used across hyperscale data centers. SRMO is being developed as a workstream within this community rather than as a proprietary Credo standard.

When will the SRMO specification be approved?
Credo has stated a target of 2027 for OCP approval of the specification. No date has been announced for when compliant hardware products will ship or become commercially available.

Is a hyperscale cloud provider involved in SRMO?
Reporting on the announcement references a leading hyperscaler participating in the workstream, but its name has not been disclosed by Credo or confirmed in subsequent coverage.

What does SRMO mean for scale-up versus scale-out networking?
Credo’s announcement states that SRMO is intended to serve both scale-up connectivity, which links accelerators tightly within a rack, and scale-out connectivity, which links groups of racks into a larger cluster fabric, using one underlying optical approach.

How is SRMO different from existing optical transceiver standards?
Existing pluggable transceiver standards were largely developed for longer scale-out links in general-purpose data center networking. SRMO is focused specifically on short-reach links using micro-emitter optical arrays, targeting the tighter distances and higher channel density needed inside AI accelerator clusters.

Has Credo announced pricing for SRMO-related products?
No. Credo’s announcement did not include pricing, shipment dates, or detailed channel-count specifications for any SRMO-compliant hardware.

Credo Launches SRMO Optics Standard With 7 Partners [2026]