Top Optical Transceiver Manufacturers in 2026

The Ethernet transceiver market grew 93% in 2024 and 82% in 2025. LightCounting's April 2026 forecast puts 2026 growth at 65%, and the reason for the deceleration is not demand - orders currently exceed InP EML and laser chip output by roughly 30%.800G and 1.6T modules together are expected to account for about $14.6 billion in 2026, close to 64% of total optical module revenue.

 

Mordor Intelligence values the 2026 market at $15.42 billion and Fortune Business Insights at $17.15 billion. Those figures are lower because their segmentation folds in telecom and access optics and excludes the AI-cluster interconnect categories that LightCounting models separately. Three firms, three numbers, three different definitions of "the market" - check the scope line before comparing any two of them.The one thing every model agrees on is that the growth ceiling in 2026 sits on thesupply side, not the demand side.

 

 

One number explains why. Industry analysts estimate that for every $1 spent on AI compute hardware (GPUs), approximately $0.15 to $0.20 now goes to networking and optical connectivity-a ratio that has roughly doubled since 2023. LightCounting projects the AI cluster optics segment specifically will double from $5 billion in 2024 to $10 billion in 2026. That concentration of spending has reshaped which manufacturers gain share, which technologies win design slots, and how procurement teams evaluate their supplier relationships.

 

The table below uses each company's most recent reported period rather than a common calendar year, because the reporting cycles no longer line up and the gap between a 2025 annual figure and a Q1 2026 figure is now material.

 

Company	Latest reported period	Revenue	YoY	What it tells a buyer
InnoLight (300308.SZ)	Q1 2026	¥19.50B	+192%	Gross margin 46.1%; silicon photonics now over half its shipment mix
Coherent (NYSE: COHR)	Q3 FY26, ended Mar 31 2026	$1.81B	+21%	Datacenter & Communications $1.4B; order book extends into calendar 2028
Lumentum (NASDAQ: LITE)	Q3 FY26, ended Mar 28 2026	$808.4M	+90.1%	Components 66% of revenue - it supplies rivals as much as it competes
Eoptolink (300502.SZ)	Q1 2026	¥8.34B	+106%	Optical interconnect gross margin 47.8% in FY2025, highest of the listed set
Applied Optoelectronics	Q1 2026	$151.1M	+51%	First volume 800G shipments; ~100k units/month capacity exiting Q1

 

Full-year 2025 for the two Chinese leaders: InnoLight posted ¥38.24 billion in revenue (+60.3%) with net profit up 108.8%, and transceiver gross margin climbing from 34.65% to 42.61%. Eoptolink posted ¥24.84 billion (+187%) with net profit up 236%. Both improvements came from mix rather than price - 800G and 1.6T displacing 400G in the shipment blend. On share: LightCounting has InnoLight first in global transceiver shipments for the third consecutive year, at 23.4% in 2025, roughly seven points ahead of Coherent at about 16%. The top five suppliers together hold close to 56%. That leaves a large remainder, and most of it is not hyperscale volume.

 

Sources: InnoLight and Eoptolink 2025 annual reports and Q1 2026 filings; Coherent Q3 FY2026 (May 6, 2026); Lumentum Q3 FY2026 (May 2026); Applied Optoelectronics Q1 2026 (May 7, 2026); LightCounting market models. Segment definitions differ between companies and figures are not directly comparable line for line.

 

How to Read This List: Manufacturer, Supplier, or Switch Brand?

 

When people search for the top optical transceiver companies in 2026, they are often mixing three different categories. Some companies are true optical transceiver manufacturers that design and assemble modules at scale. Some are upstream component specialists supplying lasers, DSPs, drivers, TIAs, or silicon photonics engines. Others are networking OEMs that sell branded optics mainly to protect platform compatibility. For procurement teams, that distinction matters more than the logo on the module. A switch vendor may control the qualification process, but the actual optical module manufacturer behind a high-volume 400G, 800G, or 1.6T deployment may be a specialized datacom optics supplier.

 

A practical list of optical transceiver manufacturers should therefore be read by capability, not just by company size. Hyperscale buyers care about 800G and 1.6T production capacity, AI cluster qualification, thermal performance, and allocation guarantees. Enterprise and regional data center buyers care more about MSA compliance, EEPROM coding, lead time, warranty terms, and whether the supplier can support mixed Cisco, Arista, Juniper, NVIDIA, or white-box switch environments. That is why the "best" optical transceiver supplier in 2026 depends heavily on the buyer profile.

 

Coherent Corp.

 

Coherent Corp. (formerly II-VI, which acquired Finisar in 2019) has one of the broadest transceiver portfolios in the industry, spanning SFP/SFP+ modules through 400G QSFP-DD transceivers and 800G platforms for AI cluster interconnects.

 

The reason Coherent keeps winning allocation battles: vertical integration. The company makes its own indium phosphide lasers, VCSELs, modulators, and photonic integrated circuits in-house. Most competitors buy at least some of these components from third parties. During the 2023–2024 supply crunch, when lead times stretched past 30 weeks for some module types, Coherent's internal supply chain gave it an obvious edge.

Coherent's Q3 fiscal 2026 results (May 6, 2026) put revenue at $1.81 billion, up 21% year-over-year, with the Datacenter and Communications segment at $1.4 billion and datacenter revenue up 37%. Bookings set a record and the order book now extends into calendar 2028. In March 2026 NVIDIA invested $4 billion across Coherent and Lumentum - $2 billion into Coherent - explicitly to accelerate indium phosphide capacity. Coherent expects to double internal InP output by the end of 2026, a quarter ahead of plan, and to more than double it again through 2027. At OFC 2025, Coherent was the only company demonstrating 1.6T transceivers across three different laser sources-silicon photonics, EML, and 200G VCSEL-with the SiPh and EML variants already ramping into production. The VCSEL-based 1.6T is expected to follow in mid-calendar 2026. Coherent has also shipped optical circuit switching (OCS) systems to seven customers, a product line that Google's Ironwood architecture is helping to validate at scale.

 

InnoLight (Zhongji Innolight)

 

Zhongji InnoLight was founded in 2008 with venture capital from the U.S. and China. Seventeen years later, it's arguably the most consequential transceiver company in the market. The Suzhou-based manufacturer posted revenue of ¥23.86 billion in 2024-up 122.6% year-over-year. Net profit rose 137.9%, with margins at 20–22%.

 

InnoLight currently holds over 50% of NVIDIA's optical module wallet share and is a primary supplier to Google's AI infrastructure as well. TrendForce estimates that InnoLight will capture roughly 80% of Google's orders for modules above 800G in 2026-driven partly by Google's Ironwood TPU architecture, which uses an all-optical Apollo OCS network and requires every cross-cabinet link to run on 800G or 1.6T optics. With Google forecasting nearly four million TPU shipments in 2026, that translates to demand for more than six million high-speed optical modules from a single customer. A 2023 partnership with Tower Semiconductor gave InnoLight access to a proven silicon photonics process platform (PH18), further strengthening yields at 400G and above.

 

LightCounting ranks InnoLight first in global optical transceiver shipments for the third consecutive year, with a 23.4% share in 2025. The company's 2024–2025 performance in high-speed datacom, where record market expansion was concentrated, is what moved it there. The product line spans 100G through 800G across QSFP28, QSFP-DD, and OSFP form factors, with 1.6T OSFP-XD modules in production. InnoLight expanded manufacturing capacity fast enough to avoid the allocation constraints that plagued competitors throughout 2024.

 

Which Vendors Are Shipping 800G and 1.6T for AI Fabrics?

 

For AI data center fabrics, the short answer is that production-ready 800G solutions are already concentrated among a small group of high-speed Ethernet optics vendors. InnoLight, Coherent, Eoptolink, Cisco/Acacia, Lumentum, Broadcom-linked ecosystems, and selected contract manufacturing partners are the names most often associated with 800G OSFP, QSFP-DD, coherent DCI, and next-generation 1.6T roadmaps. The dividing line is no longer whether a vendor can show a sample at OFC, but whether it can pass hyperscale qualification, hold power within the rack budget, and ship stable volume across multiple switch refresh cycles.

 

The 1.6T optical module producer landscape is still narrower. 1.6T deployments depend on 200G-per-lane PAM4 electrical interfaces, OSFP-XD or OSFP224 form factors, high-yield silicon photonics or EML designs, and reliable DSP or LPO/LRO architecture choices. For buyers asking which vendors are developing or shipping 1.6T optical solutions, the most credible candidates are those already shipping 800G at scale and already aligned with AI switch platforms, liquid-cooled rack designs, and 2026–2027 hyperscale qualification windows.

 

Cisco Systems

 

Cisco built its optical transceiver business through two acquisitions: Luxtera (2019) brought silicon photonics expertise, and Acacia Communications (2021) added coherent DSP technology. Acacia's WaveLogic platform leads in 400G+ coherent pluggable shipments, and the 2024 expansion to 800ZR and 800G ZR+ in both QSFP-DD and OSFP pushed optical segment revenue to approximately $2.3 billion with a 28.4% operating margin.

Where Cisco pulls ahead of component-only vendors is platform compatibility. Catalyst and Nexus switches natively support Cisco optics, eliminating the interop testing overhead that comes with mixing vendors. Routed Optical Networking, which Cisco claims can cut DCI space, power, and cooling needs by over 80%, adds another lock-in incentive.

 

Lumentum and the NeoPhotonics Consolidation

 

Lumentum completed its acquisition of NeoPhotonics in August 2022, then added Cloud Light Technology in October 2023 - the latter is what actually moved it into cloud datacenter transceivers rather than components alone. By November 2025, Lumentum's 800G coherent bookings had surpassed 100G orders for the first time-a milestone that validates the speed-grade transition across the carrier and DCI segments. Revenue reached $808.4 million in the quarter ended March 2026, up 90.1% year-over-year, with components making up 66% of the total. Lumentum is currently the only supplier shipping 200G-per-lane EMLs in volume, and it guides EML unit shipments to grow more than 50% between the December 2025 and December 2026 quarters. The company's R64 Optical Circuit Switch, a 64×64-port MEMS-based platform consuming under 150W while handling over 100 Tbps, positions Lumentum in the emerging OCS market that Google's Apollo architecture is accelerating.

 

Worth noting: Lumentum's influence extends beyond its own branded modules. A significant number of the DFB lasers and photodetectors inside other companies' transceivers come from Lumentum fabs. That upstream position gives the company leverage that doesn't show up in module-level market share numbers.

 

Chinese Manufacturers: Eoptolink, Accelink, and Hisense Broadband

 

Chinese optical transceiver manufacturers account for an estimated 40–50% of global shipments in some product categories.

Eoptolink posted 179% revenue growth in 2024, driven by direct sales to cloud operators following the model InnoLight pioneered. Net margin hit 33%-the highest among publicly traded transceiver vendors. The company has built a manufacturing facility in Thailand specifically to serve North American demand outside of Chinese export-control exposure, and management has signaled that this plant will support 1.6T volumes starting in late 2025 into 2026. Eoptolink is also pushing multimode approaches for short-reach AI cluster links, launching 1.6T OSFP modules using 200G VCSELs to offer a cost advantage over single-mode alternatives.

 

Accelink Technology, headquartered in Wuhan, covers transport, access, and data segments. Its Malaysian subsidiary Phabritek (opened November 2023) manufactures high-end optoelectronic modules closer to international customers, while Accelink's North American R&D center continues work on silicon photonics and hybrid integration for next-generation 800G transceiver modules.

Hisense Broadband leads in access network transceivers for FTTH, GPON, and XGS-PON, and has been expanding into datacom. LightCounting notes that Hisense, Accelink, and HGGenuine all benefited from surging demand by Chinese cloud companies in late 2024, with purchase volumes expected to double in 2025.

 

China's optical supply chain runs deeper than the publicly traded names. Below the top tier, mid-size manufacturers handle a significant share of global module production with their own packaging, testing, and OEM coding lines. FB-LINK, based in Shenzhen, ships 100G through 800G modules to customers in over 50 countries. Its Wuhan R&D center develops 400G QSFP-DD DR4 silicon photonics transceivers and DCI BOX subsystems.

 

This middle layer of fiber optic transceiver manufacturers is where many non-hyperscale projects actually get executed. A cloud operator may negotiate directly with InnoLight or Coherent, but an enterprise network upgrade, carrier edge buildout, university data center, or regional ISP deployment often needs smaller batch flexibility, faster coding turnaround, and practical compatibility support. In those cases, an optical module company with in-house EEPROM programming, burn-in testing, and multi-brand switch validation can be more useful than a top-tier supplier that is fully booked by hyperscale contracts.

 

For PAM4 optical transceiver manufacture, the hard part is not only assembling the module. The supplier has to control optical eye margin, DSP configuration, thermal behavior, DOM reporting accuracy, and firmware compatibility at the same time. That is why mid-size manufacturers with their own testing workflow can compete in 100G, 400G, and selected 800G projects even if they are not ranked among the global top 10 by revenue.

 

For buyers outside the hyperscale tier, manufacturers at this level tend to offer shorter lead times and more flexible coding than OEM brands.

 

Where FB-LINK Fits: The Independent Module Layer

 

Start with what this tier does not do. FB-LINK does not fabricate InP lasers, does not design PAM4 DSPs, and does not compete for hyperscale frame agreements - the allocation and capital requirements of that business belong to InnoLight, Coherent, Eoptolink and Lumentum. The work that sits in the independent layer is different: taking qualified optical and electrical components, packaging and testing them to MSA specification, coding them for the switch platforms a specific customer actually runs, and delivering in batch sizes that a Tier-1 supplier will not quote.

 

FB-LINK has run that business from Shenzhen since 2012, across three brands - FB-LINK, Aocclink and PUSMAI - with a Wuhan technology centre opened in 2018 to work on coherent transmission. The product path is visible in the dates: 10G CWDM and DWDM modules in 2013, DWDM transport for domestic data centres in 2015, 40G and 100G from 2016, a 400G QSFP-DD DR4 silicon photonics transceiver developed at the Wuhan centre in 2021, and a DCI BOX subsystem in 2022 supporting 1.6T in 1U and 3.2T in 2U. Current shipping range runs from 1000BASE SFP through 1.6T OSFP, alongside DWDM/CWDM frames and cards, EPON and GPON OLT/ONU, MPO/MTP assemblies and DAC/AOC.

 

What the credentials actually cover

 

Certification claims are worth checking rather than counting, so here is the specific scope. ISO 9001:2015 quality management, certified by SGS in 2020 and separately by ISA in 2022. ISO 14001 environmental management and ISO 45001 occupational health and safety, both certified in 2022. A Chinese telecommunications equipment network access licence (进网许可证) granted in 2019 - relevant mainly to carrier-facing deployments, since it requires type approval rather than self-declaration. On the IP side, more than 65 invention patents and over 90 software copyright registrations, under national high-tech enterprise status. Customers in more than 50 countries, with a base of 180+ small and mid-size enterprise accounts rather than a handful of large ones - which matters for supply behaviour, because no single customer can absorb a quarter's output and displace everyone else's schedule.

 

Where this tier is the right answer, and where it is not

 

It is the right answer when the deployment is 10G to 400G, when the switch estate mixes Cisco, Arista, Juniper, NVIDIA and white-box platforms and every port needs different EEPROM coding, when order quantities sit below what a Tier-1 supplier will quote, when a DWDM or DCI subsystem has to interoperate with equipment already in the rack, and when someone needs to map an existing OEM part number list to MSA-equivalent modules before a refresh.

 

It is the wrong answer when the requirement is 1.6T volume for an AI training fabric, when the buyer needs an integrated warranty that covers both the switch and the optic as one system, or when procurement policy prohibits non-OEM optics outright. Those cases belong with the vendors listed earlier in this article, and saying so is more useful than pretending otherwise.

 

Before you send an RFQ

 

The fastest way to get a usable answer out of any module supplier - this one included - is to lead with four things rather than a part number alone: the switch model and its current OS version, the actual link distance and fibre type, the quantity and required delivery date, and whether the modules will sit in an air-cooled or liquid-cooled rack. Those four inputs determine whether a proposed module will pass qualification. A part number on its own does not. If the starting point is an existing bill of materials, map an existing OEM part number list to MSA-equivalent modules first.

 

Send that combination through and the response should come back as a mapped part list with reach, coding target and lead time per line, not a price sheet.

 

Broadcom, Sumitomo Electric, and Other Key Players

 

Broadcom's transceiver business operates through its Avago division, primarily serving hyperscale Ethernet switching. The longer-term bet is CPO-mounting optical engines directly onto switch ASIC packages for 30–40% power consumption reductions. Sumitomo Electric maintains strength in Asia-Pacific, with a March 2025 Point2 Technology partnership targeting next-gen 25G modules for 5G/6G carrier upgrades.

 

A notable supply-chain shift: Applied Optoelectronics completed its first volume 800G shipments to a hyperscale customer in Q1 2026, after a firmware issue pushed the ramp back a quarter - a reminder that a signed frame agreement and a shipping product are separate milestones. It exited Q1 with roughly 100,000 units per month of 800G/1.6T capacity, targeting more than 500,000 by year end, and took a separate 1.6T order worth over $200 million in March 2026 for Q3 delivery. This deal-alongside similar hyperscaler arrangements with Fabrinet-signals deliberate diversification toward U.S.-based optical manufacturers. Marvell has also climbed rankings on strong 400ZR demand at Microsoft.

 

AAOI Competitors and the Public-Market Optics Basket

 

AAOI competitors in optical transceivers are not limited to module brands with similar product pages. In the 800G AI data center cycle, Applied Optoelectronics competes against high-speed module specialists such as InnoLight, Eoptolink, Coherent, and Lumentum-backed ecosystems, while also depending on the same constrained pool of DSPs, EML lasers, drivers, and advanced packaging capacity. That makes AAOI part of a broader optical interconnect supply chain rather than a standalone 800G module story.

 

For readers tracking top optical transceiver stocks in 2026, the cleaner way to think about the sector is by exposure type. Coherent and Lumentum offer stronger leverage to lasers, components, and vertically integrated optics. AAOI is more directly tied to high-speed datacom module ramps. Broadcom and Marvell sit closer to switch silicon, DSPs, and optical electrical interfaces. InnoLight and Eoptolink are closer to high-volume AI cluster module shipments. The ranking of optical transceiver market leaders can change quickly when one hyperscaler shifts allocation, so customer concentration and production capacity matter as much as headline revenue.

 

Three Technology Shifts Worth Watching

 

 

Reading the 2026 Forecasts: LightCounting, TrendForce, and Dell'Oro

 

Forecasts for the 2026 optical transceiver market can look inconsistent because each research firm cuts the market differently. A LightCounting-style forecast tends to focus closely on Ethernet optics, AI cluster interconnects, shipments, pricing, and supplier cycles. TrendForce often frames the market through AI server infrastructure, 800G/1.6T demand, component shortages, and regional supply-chain shifts. Dell'Oro forecasts are frequently used by buyers who want to connect optical transceiver demand with broader data center networking, DCI, router, and cloud capex trends. The numbers may differ, but the direction is similar: AI data center optical connectivity is pulling 800G and 1.6T adoption forward faster than traditional telecom upgrade cycles.

 

For this reason, a 2026 optical transceiver market forecast should not be read as a single TAM number. A 100G enterprise refresh, an 800G AI leaf-spine deployment, a 1.6T scale-out fabric, and an 800G ZR/ZR+ DCI link all sit inside the same broad market, but they use different suppliers, price curves, chipsets, fiber plans, and qualification cycles. The most useful forecast for procurement is the one that matches the actual application.

 

Which manufacturers end up on top in 2027 depends partly on how three technical transitions play out.

 

Silicon photonics is finally shipping at volume. Intel, Cisco, Marvell, and Coherent all have production SiPh-based transceiver modules on the market. Silicon photonics has already taken roughly 72% of 1.6T designs. InnoLight reported in Q1 2026 that SiPh accounts for more than half its own shipment mix and named it, alongside product mix and yield, as one of three drivers behind gross margin rising to 46.1%. CMOS-compatible fabrication on 300mm wafers plugs into existing semiconductor infrastructure at cost curves that III-V compound fabs can't match.

 

The 800G-to-1.6T transition is compressing faster than expected. TrendForce data shows the global shipment share of modules above 800G rising from 19.5% in 2024 to over 60% by 2026. The IEEE 802.3dj task force covering 200G-per-lane PAM4 reached draft D2.4 in March 2026 and is now tracking to complete in late 2026, with early 200G/lane products shipping ahead of ratification. IEEE 802 chartered a separate 400 Gb/s per lane study group on March 13, 2026, covering electrical interconnects and single-mode optical reaches up to 500 metres, and the OIF's 1600ZR/ZR+ implementation agreements are running in parallel. OSFP-XD has emerged as the dominant 1.6T form factor, specified in 92% of hyperscale contracts according to OCP data. Beyond 1.6T, Amazon, Google, and Meta are already funding OIF work on 448G PAM4 for 3.2T, demonstrated at networking events in 2025.

 

Co-packaged optics (CPO) is harder to time, but the commitments are getting concrete. NVIDIA declared Spectrum-X Ethernet Photonics in production on May 31, 2026 - the first co-packaged optics Ethernet switch built on 200G SerDes, with CoreWeave, Lambda and Oracle Cloud Infrastructure as launch adopters and broad availability guided to the second half of 2026. Broadcom's 51.2T Bailly CPO switch is also in volume manufacturing and has completed deployment validation at Meta, cutting optical power draw by up to 70% versus pluggables. TrendForce expects the real volume window in 2027–2028, gated by optical engine yield and advanced packaging capacity rather than by demand. Google's Apollo OCS architecture, which reduces per-switch power consumption by roughly 95% versus traditional electrical switching, is already validating the case for tighter optical integration. One underappreciated factor: the rise of liquid cooling in AI data centers is changing the physical environment that transceivers operate in, requiring coherent optical modules hardened for harsher thermal conditions than air-cooled racks demanded.

 

One point that gets lost in the "CPO replaces pluggables" framing: a co-packaged switch presents bare single-mode fibre at the faceplate. The server and NIC side of the same fabric still terminates in pluggable 400G, 800G or 1.6T modules, and in-rack GPU-to-GPU links still run on DAC, ACC, AEC and AOC. CPO changes what sits inside the switch chassis. It does not remove the transceiver from the bill of materials for everything attached to it.

 

Why 2026 Is a Supply Problem, Not a Demand Problem

 

Every speed grade above 400G is now rationed at the laser, not at the module line. Electro-absorption modulated lasers and continuous-wave lasers both require indium phosphide substrates, and there is no volume alternative material. InP substrate supply is concentrated in two or three producers holding roughly 75% of global output, with AXT alone at an estimated 60–70%. Epitaxial yields on those substrates run anywhere from 15% to 50% depending on wafer generation and design complexity, which is why capacity cannot simply be bought - it has to be qualified through production runs before anyone will commit volume against it.

 

NVIDIA's response was to pre-allocate EML capacity at the leading suppliers and then invest $4 billion into Coherent and Lumentum in March 2026 to expand it. The side-effect for everyone else is that EML lead times moved past 2027. McKinsey models 800G module output running 40–60% below demand through 2027 and 1.6T running 30–40% short through 2029. LightCounting is more constructive, putting the current demand-supply gap at about 30% and expecting shortages to clear by the end of 2026 as new InP capacity qualifies. Both views describe the same market: allocation decides who ships, and price is no longer the clearing mechanism.

 

That has an unequal effect across the product range, and it is the single most useful thing for a mid-size buyer to understand about 2026.

 

Lead Time and Availability by Speed Grade, mid-2026

 

Speed grade	Typical lead time	Constraint	Practical read for a non-hyperscale buyer
1G / 10G / 25G (SFP, SFP+, SFP28)	8–14 weeks	None material	Normal procurement. Price and coding are the only variables
40G / 100G (QSFP+, QSFP28)	8–14 weeks	None material	Widely available. Cost curve is at its most favourable point
400G (QSFP-DD, OSFP)	10–20 weeks	Competes for 100G EMLs, not 200G	The pragmatic choice for most enterprise and regional builds
800G (OSFP, QSFP-DD)	30–40+ weeks	200G EML allocation, hyperscale priority	Expect allocation questions before pricing questions
1.6T (OSFP-XD, OSFP224)	40+ weeks	200G EML plus 2.7–2.8× the InP substrate per unit	Effectively hyperscale-allocated through 2026

 

Note that a 1.6T module consumes roughly 2.7 to 2.8 times the InP substrate area of an 800G module. Doubling the port speed does not double the upstream demand; it nearly triples it. That single ratio explains why 1.6T allocation is tighter than the headline shipment forecasts suggest.

 

The procurement conclusion runs against instinct. When a project cannot get 800G allocation inside its build window, the usual reaction is to wait for a quarter that never quite arrives. The better move in a supply-constrained year is usually to re-cut the topology: a 400G leaf-spine fabric with more ports, built from modules that are actually available in 12 weeks, will carry an enterprise or regional workload for the two to three years it takes for 800G supply to normalise. The links that genuinely require 800G today are AI training fabrics with GPU-to-GPU east-west traffic. A colocation build, a carrier edge node, a university cluster or a regional ISP core rarely does - and paying an allocation premium plus a 40-week wait for headroom that will not be used for three years is the most expensive mistake available in this market.

 

Which optical transceiver speed grades are actually available in 2026?

 

10G through 400G modules are available on normal 8–20 week lead times; 800G and 1.6T are allocation-constrained at 30–40+ weeks and are effectively prioritised to hyperscale customers through 2026. The constraint is upstream of the module line: InP substrate supply is concentrated in two to three producers, epitaxial yields for EML lasers run 15–50% by wafer generation, and NVIDIA's pre-allocation of EML capacity pushed lead times past 2027. LightCounting puts the current demand-supply gap at roughly 30% and expects it to clear by end-2026; McKinsey models 800G output 40–60% below demand through 2027.

 

This suits enterprise, colocation, carrier edge and regional ISP projects, where a 400G fabric built from available modules will carry the workload for the two to three years the 800G supply picture takes to normalise. It does not suit AI training fabrics with GPU-to-GPU east-west traffic, where 800G or 1.6T is a topology requirement rather than a headroom preference - those projects need allocation commitments written into the purchase agreement, not spot quotes.

 

One ratio to carry into any 1.6T conversation: a 1.6T module consumes 2.7–2.8× the InP substrate area of an 800G module. Doubling port speed nearly triples upstream demand, which is why 1.6T allocation is tighter than headline shipment forecasts imply.

 

Comparison of Leading Optical Transceiver Manufacturers

 

Company	Product Speed Range	Core Technology Advantage	Primary Customer Segment	Headquarters
InnoLight	100G – 1.6T	High-volume silicon photonics and advanced datacom optics	Hyperscale cloud and AI infrastructure providers	Suzhou, China
Coherent Corp.	10G – 1.6T	Vertical integration across lasers, modulators, and photonic components	Hyperscale data centers, telecom carriers, and networking OEMs	Saxonburg, Pennsylvania, USA
Cisco (Acacia)	100G – 800G+	Coherent DSP technology and deep integration with switching platforms	Enterprise networks, service providers, and DCI deployments	San Jose, California, USA
Lumentum	100G – 800G+	Advanced laser sources, coherent optics, and upstream component supply	Telecom equipment vendors, hyperscale data centers	San Jose, California, USA
Eoptolink	100G – 1.6T	Cost-efficient high-speed datacom modules and VCSEL-based solutions	Hyperscale cloud operators and AI clusters	Chengdu, China
Accelink Technology	25G – 800G	Broad telecom and datacom portfolio with hybrid photonic integration	Telecom carriers, networking equipment vendors	Wuhan, China
Hisense Broadband	10G – 400G	Strong access network optics including GPON and XGS-PON	FTTH operators and broadband equipment providers	Qingdao, China
Broadcom (Avago)	400G – 800G+	Switch silicon ecosystem and development of co-packaged optics (CPO)	Hyperscale data centers and networking OEMs	Palo Alto, California, USA

 

Choosing a Transceiver Supplier in 2026

 

Price still matters, but supply chain reliability, roadmap credibility, multi-vendor interop testing, and thermal behavior in production environments have all moved up the checklist. Liquid-cooled AI pods create harsher operating conditions than air-cooled racks; a module that tests fine in a vendor's climate chamber can behave differently at 45°C ambient with restricted airflow.

The calculus differs sharply depending on buyer scale.

 

Hyperscale and large cloud operators tend to qualify 2–3 transceiver vendors per speed grade and negotiate multi-year frame agreements with price locks and allocation guarantees. At this tier, roadmap alignment is the primary filter-buyers need suppliers who can deliver 800G in volume today and have 1.6T production-ready for 2026 switch platform refreshes. Coherent, InnoLight, and Eoptolink currently dominate this bracket. The Amazon-AAOI deal signals that even hyperscalers are actively diversifying their supplier base beyond the established top three.

 

Mid-size enterprises and regional data center operators face a different set of trade-offs. They're typically deploying 100G or 400G-not 800G-and their volumes don't command the allocation priority that hyperscalers enjoy. For this buyer profile, third-party compatible vendors offer a practical path: the 400G manufacturing process has matured enough that DR4 modules now cost $400–700 from qualified third-party sources, versus $1,500–3,000 from OEM brands.

 

Which supplier tier matches which buyer?

 

Supplier tier	Typical MOQ	Coding turnaround	Best fit	Where it fails
Switch OEM (Cisco, Arista, Juniper, NVIDIA)	1 unit	N/A	Single-vendor estates under an integrated support contract	2–4× unit cost; optic roadmap tied to the platform roadmap
Tier-1 module maker (InnoLight, Coherent, Eoptolink)	Frame agreement, 5–6 figure units	Not offered	Hyperscale and large cloud, multi-year allocation commitments	Will not quote enterprise volumes; no mixed-brand coding service
Independent manufacturer with in-house coding	[MOQ by speed grade – to confirm]	[coding sample turnaround – to confirm]	Mixed-vendor estates, 10G–400G, regional and enterprise builds	No integrated switch warranty; not a route to 1.6T volume in 2026
Distributor / stockist	1 unit	Rebadged only	Emergency replacement, single-unit spares	No control over the coding source; traceability breaks at the reseller

 

The distinction that matters in an RFQ is not whether a company calls itself a manufacturer or a supplier. It is whether the party quoting controls four specific things: the EEPROM coding step, the burn-in and test record for the batch being shipped, the warranty decision, and the delivery date. A company that controls all four can answer a compatibility question in a day. A company that controls none of them is relaying a question down a chain, and the answer arrives after the maintenance window.

 

For deployments below roughly [port-count threshold – to confirm], the independent tier is usually the only one that will engage at all - Tier-1 makers decline the volume and OEM pricing distorts the project budget. Above that threshold, running a dual-source strategy across an independent supplier and one OEM-branded line is common practice, and it is worth setting up before the first order rather than after the first allocation problem.

 

Procurement Checklist for Non-Hyperscale Buyers

 

For buyers comparing an 800G optical transceiver price in 2026, the quoted module cost is only one part of the decision. A realistic RFQ should also ask about MOQ, lead time, switch compatibility coding, DOM/DDM reporting, burn-in duration, temperature rating, return policy, and whether the supplier has tested the exact switch operating system version in use. A low unit price can disappear quickly if modules fail qualification, trigger switch alarms, or require manual troubleshooting during a maintenance window.

 

Market leaders in 800G/1.6T high-speed Ethernet validation are usually strong because they treat qualification as a system problem. The module, switch ASIC, firmware, fiber plant, airflow, and rack temperature all interact. For regional data centers and enterprise networks, the safest optical transceiver supplier is often the one that can review the bill of materials, map OEM part numbers to compatible MSA modules, and recommend SR, DR, FR, LR, ZR, DAC, or AOC options based on actual link distance and switch platform constraints.

 

Companies like FB-LINK ship MSA-compliant, burn-in-tested modules with multi-vendor coding support-the same underlying components, packaged and tested to the same standards, at significantly lower cost. The trade-off is that you don't get the OEM's integration warranty with your switching platform, but for organizations running heterogeneous environments, that's less of a concern than it once was.

 

In both cases, thermal validation deserves more attention than it typically receives in procurement checklists. A growing number of 800G field failures trace back not to component defects but to thermal derating in rack environments that exceed the module's specified operating range.

The demand outlook through 2026 looks firm. The four largest U.S. hyperscalers have committed over $650 billion in combined capex, with about 75% going to AI infrastructure. A significant chunk ends up in optical interconnects. The vendors that come out ahead will be the ones shipping volume on their 800G and 1.6T roadmaps-not still qualifying samples.