On November 4, Mixx Technologies issued a statement best read as a positioning paper with a product attached, not a product launch. The San Jose company — a member of the Expanded Beam Optics Multi-Source Agreement (EBO MSA) and a contributing member of the Open CPX MSA — set out its view of high-density optical scale-up, with a single thesis: near-package optics is only as reliable as its connector. At the center is the SxC connector, the connectivity layer of Mixx’s HBxIO platform, launched in September: 0.2 dB typical and under 0.5 dB maximum fiber-to-fiber loss, and up to 24,576 fibers in a single Open Compute Project rack unit — four times the fiber density of MMC. No deployment, no customer, no volume, and no qualification timeline is disclosed. What is on the record is a standards argument and a claim about where the next bottleneck sits.
This is the third leg of a transition this desk has tracked all month: ficonTEC and HTSI industrialized CPO wafer test on October 1, and Tower and Innolight put a first NPO device into early production on November 3. Test decides whether dies can be proven, production whether engines can be built repeatedly, and the connector whether the system can be serviced at inference’s fiber counts. None of the three is a shipment.
What was actually announced
Mixx published its position on November 4, 2026, from San Jose. The statement is not a launch — SxC launched in September, with demonstrations at SEMICON Taiwan beginning September 2 — nor a customer announcement. It is a statement of support for open specifications, built around three claims: loss, density, and a detachable module on an open socket.
The loss claim: SxC is an expanded-beam design, and Mixx states it holds 0.2 dB typical and under 0.5 dB maximum fiber-to-fiber loss. The link budget is the total loss of laser power a link can tolerate between emitter and receiver, and every connection spends part of it. A connector that spends less leaves more for the rest of the link, which is why Mixx’s chief executive, Vivek Raghuraman, frames recovered decibels as architectural headroom: “Every recovered dB becomes architectural headroom, and ultimately, economic advantage.”
The density claim: in a single OCP rack unit, SxC carries up to 24,576 fibers, four times the fiber density of MMC. Per the September launch release, each connector scales up to 64 fibers, and twenty-four bulkheads bring one rack unit to 24,576 fibers — four times the density of MMC-VSFF, the highest-density connector platform qualified for deployment today.
The architectural claim is the one the statement is really about. Mixx is building an Open CPX module pairing its HBxIO optical engine with a detachable SxC connector, so engine and fiber can be tested and replaced separately. The Open CPX MSA standardizes the socket beside the package, so modules can come from multiple vendors. Its founding members are Ciena, Coherent, Marvell, Molex, Samtec, and TeraHop; the specification, released September 17, 2026, defines a pluggable form factor for optics beside the host ASIC at up to 7.2 Tbps per socket and 200 Gbps per lane. The EBO MSA, created in March 2026, now has more than 60 members and is developing a 128-fiber design specification — in development, not released.
Why the connector decides the link budget
The claim that the connector is the bottleneck sounds like marketing until you count what a near-package link contains. In a pluggable, engine and connector live together in a module at the front panel, where a technician can reach every port. In near-package optics, the engine moves onto the board beside the switch chip, inside the box, and the fiber leaves through a connector that is harder to reach, harder to clean, and repeated far more times than a front-panel connector ever was. At NPO channel counts — Tower’s release last week put the typical figure at 16 or more per device — the arithmetic compounds: a connector losing 0.2 dB instead of 0.5 dB saves that margin at every mating point, across thousands of fibers in a rack.
The loss figure needs its context stated plainly. Mixx’s 0.2 dB typical and under 0.5 dB maximum are the company’s stated specifications, not independently measured results, and no test conditions, sample size, or measurement methodology is disclosed. They are vendor figures, to be verified in qualification rather than treated as measured performance in a customer system.
Expanded beam, explained — and what it gives up
The physics of the SxC design is worth explaining, because it is the reason the loss and cleaning claims are connected rather than separate. A physical-contact connector presses two polished fiber end faces together, and single-mode light travels in a core about 9 microns across. At that scale, dust, oil, or a scratch can block the light path. Expanded-beam optics widens the beam — to about 80 microns in one design the EBO MSA has shown — carries it across a small air gap, and refocuses it into the receiving fiber; the end faces never touch.
A wider beam is less sensitive to dust or misalignment, because a particle blocks a smaller fraction of the beam. The EBO MSA reports about a 20-fold reduction in mating force, about an 85 percent reduction in install and clean time, and — at ECOC 2026 — that about 80 percent of network issues are cleaning-related and about 95 percent of maintenance requests are for cleaning. Those are the MSA’s figures, reported in the statement, not independently verified measurements — but they explain why a connector vendor would anchor its pitch on cleaning rather than raw bandwidth.
Expanded beam is not free: the lens elements add components, cost, and alignment tolerances, and the air gap adds a reflection mechanism a physical-contact joint lacks. No cost, mating-cycle, or temperature-range data is disclosed — the numbers a buyer would need before the cleaning argument becomes a procurement argument.
Radix is a fiber-count problem
The statement’s most interesting argument is not about loss or cleaning. It is about radix — the number of ports on each switch — and it reframes the inference network as a connectivity problem rather than a bandwidth problem. Inference needs all-to-all connectivity: every processor must reach every other. Higher radix means fewer switching layers — less latency, less power, fewer parts to fail. But radix also means fiber count: each additional port is another set of fibers leaving the box, each crossing a connector.
This is where the 24,576-fibers-per-rack-unit figure stops being a specification-sheet boast and becomes an architectural claim: if the constraint on radix is how many fibers fit out of a rack unit, a connector with four times the incumbent’s density is the difference between a radix the architect wants and one the connector allows. No switch vendor has disclosed building a system around SxC. The argument is a thesis about what will constrain inference networks, not a report of a system that has hit the constraint.
Termination: the bottleneck nobody photographs
If the connector is the bottleneck, termination is the bottleneck inside the bottleneck, and Mixx says so directly. Termination — fixing each fiber into its connector — grows with every fiber, and the statement calls it a bottleneck across the industry as fiber counts scale. Raghuraman’s formulation is blunt: “Termination grows with every fiber, and the industry has to address it before optics can scale inside the box. With one connector technology across the signal chain, we address termination in one design rather than four. Scaling optics means measuring production in units per hour, not hours per unit.”
Today’s signal chain uses different connector technologies at front panel, backplane, board, and package — each with its own termination process, tooling, and failure modes. One connector technology across the chain would collapse four termination problems into one, and one learning curve. No units-per-hour figure, yield, or manufacturing location is disclosed. Until Mixx or a customer discloses volumes or a qualification result, the termination claim is a design intention, not a demonstrated capability.
Open CPX and the detachable module
The Open CPX half is the more consequential standards bet: it determines whether Mixx’s connector becomes a product or a platform. The MSA standardizes the socket beside the package, so modules can come from multiple vendors. Mixx is building an Open CPX module for near-package optics using its HBxIO optical engine and detachable SxC connector. The detachability is the point: because the connector comes off, the engine and the fiber are tested separately in manufacturing and replaced separately in the field.
In the CPO vision, the optical engine is bonded to the switch package, and a failed engine can condemn a switch ASIC worth far more than the optics. Open CPX instead makes the near-package engine a socketed, field-replaceable module. If the engine fails, it is swapped; if the fiber fails, it is swapped; neither failure takes the switch with it. Raghuraman’s summary — “The measure of a solution is whether a hyperscaler can confidently deploy it at scale. At volume, that means parts that can be tested and replaced on their own. A detachable connector on an open socket supports both, which makes Open CPX the pragmatic step to adopt” — is a pitch to the operator’s risk calculus, not the engineer’s bandwidth table.
Open CPX is a pragmatic step precisely because it is not CPO: it keeps optics beside the package, accepting a longer electrical path for serviceability and multi-vendor supply. Mixx’s HBxIO engine was designed for CPO and is being delivered first in a near-package socket; the company also calls it the first optical engine whose fiber interface is fully integrated and tested at the wafer level. It leaves open whether the near-package socket is a permanent product or a stepping stone — and a vendor whose first product is a connector has an interest in the step lasting a while.
Two standards, one bet
The statement is a bet on two standards bodies at different stages. Open CPX has a released specification — September 17, 2026 — with defined performance and named founding members. The EBO MSA, created in March 2026, has more than 60 members and a 128-fiber specification in development. One is a socket standard you can design to; the other is a connector specification still being written. Mixx is betting both converge in time for near-package optics to become a multi-vendor market — and that its connector, designed to both, is positioned wherever the convergence lands.
Mixx is a venture-backed startup whose $33 million Series A closed in December 2025, led by ICM’s HPQC Fund with participation from TDK Ventures, Systemiq Capital, and Ajinomoto Group Ventures, among others. That is what this statement does: it argues the buying criterion should be the connector, judged against open specifications rather than a single vendor’s roadmap. The physics is real; the framing is interested.
What is undisclosed, and why it matters
An honest accounting of what this statement does not contain is part of the story, not a footnote to it. No customer, deployment, volume, or revenue figure is given. No qualification timeline is stated. No lane rate, bandwidth, or power figure for Mixx’s module is disclosed — the 7.2 Tbps and 200 Gbps figures describe the socket standard, not the module.
The pattern of omissions sets the limit of what can be claimed here: no customer, deployment, qualification, volume, or measured in-system performance figure is asserted, because none is on the record.
One more gap will determine whether this story matters in six months. The statement describes Mixx as “building” an Open CPX module — present-continuous, not completed. What has been demonstrated is the connector itself, shown at SEMICON Taiwan in September; the module is still being built. Until Mixx discloses a module, a customer evaluation, or a qualification milestone, the statement is the beginning of an effort, not evidence it has succeeded.
What to watch
The signals that will move this story from positioning to evidence are specific. First, a customer naming SxC or an Open CPX module. Second, a qualification with measured loss figures under stated test conditions. Third, publication of the EBO MSA’s 128-fiber specification, or an Open CPX socket in a shipping product. Fourth, SxC volumes, termination throughput, or yield. Fifth, whether other vendors ship interoperable EBO products — or keep their near-package connectors proprietary.
If near-package systems ship with socketed, field-replaceable engines, the pragmatic-step thesis gains ground. If the industry jumps straight from pluggables to bonded CPO, the connector’s moment passes before it arrives. Either way, the link budget is decided one mating at a time, and the connector is where it is spent.
The fair summary is narrow. On November 4, Mixx argued that near-package optics is only as reliable as its connector, backing open specifications from the EBO MSA and the Open CPX MSA with its SxC connector — 0.2 dB typical loss, up to 24,576 fibers per OCP rack unit — and a detachable Open CPX module. No customer, deployment, volume, or qualification timeline has been disclosed. That is a modest claim, and precisely because it is modest, it is believable.
Sources and further reading
Mixx Technologies, Inc., “Near-Package Optics Is Only as Reliable as Its Connector: Mixx Backs Open Specifications for Optics Inside the Box,” November 4, 2026 (PR Newswire) — SxC connector, connectivity layer of the HBxIO platform, launched in September; 0.2 dB typical and under 0.5 dB maximum fiber-to-fiber loss; up to 24,576 fibers in a single Open Compute Project rack unit, four times the fiber density of MMC; member of the EBO MSA and contributing member of the Open CPX MSA; Open CPX module pairing the HBxIO optical engine with a detachable SxC connector; EBO MSA created March 2026, more than 60 members, developing a 128-fiber design specification; at ECOC 2026 the EBO MSA reported about 80% of network issues cleaning-related and about 95% of maintenance requests for cleaning; beam widened to about 80 microns versus a single-mode core of about 9 microns, about a 20x reduction in mating force and about an 85% reduction in install and clean time; quotes from Vivek Raghuraman, CEO of Mixx Technologies. Mixx Technologies, “Mixx Launches SxC Connector for High-Radix Scale-Up Architectures,” September 2026 (PR Newswire) — demonstrations at SEMICON Taiwan, booth #I2923, beginning September 2, 2026; each connector up to 64 fibers supporting single-mode and polarization-maintaining fiber; twenty-four bulkheads per OCP rack unit reaching 24,576 fibers, four times the density of MMC-VSFF, described as the highest-density connector platform qualified for deployment today. Mixx Technologies, “Mixx Technologies Supports Open CPX MSA Specification for Pluggable Near-Package and Co-Packaged Optics,” September 22, 2026 — specification released September 17, 2026; founding members Ciena, Coherent, Marvell, Molex, Samtec, and TeraHop; up to 7.2 Tbps per socket and 200 Gbps per lane; HBxIO described as the first optical engine whose fiber interface is fully integrated and tested at the wafer level, and as designed for CPO and delivered first in a near-package socket. Mixx Technologies, “Mixx Technologies Raises $33M Series-A,” December 1, 2025 (PR Newswire) — $33 million Series A closed December 1, 2025, led by ICM HPQC Fund with participation from TDK Ventures, Systemiq Capital, Ajinomoto Group Ventures, and others. This article’s account of the ficonTEC–HTSI CPO wafer-test cooperation (October 1, 2026) and the Tower–Innolight NPO production milestone (November 3, 2026) is drawn from this publication’s own prior coverage. No customer, deployment, volume, qualification timeline, price, or measured in-system performance figure for the SxC connector is asserted here, because none is disclosed in the announcement. This article is analysis, not investment advice.