The 800G-to-1.6T Transition: How to Verify Chinese Optical Module Suppliers’ Capacity Claims
Chinese vendors hold over 60% of the global optical-module market and an estimated 70%+ share of 800G-and-above AI interconnects. But as the industry swings from 800G to 1.6T in 2026, booth demos and annual reports tell different stories. Here is a field-tested framework for separating real production capacity from press-release promises before you sign a multi-year supply agreement.
1. Why 2026 Is the Make-or-Break Year for Optical Modules
If AI accelerators are the brain of a modern data center, optical modules are the optic nerve. Every time a GPU exchanges activations, weights or checkpoint data with its peers, the signal has to leave the chip on an electrical trace, be converted into photons by an optical transceiver, race through single-mode fiber to a Top-of-Rack or Spine switch, and be converted back to electrons on the other side. A single 100,000-GPU training cluster consumes five times as many modules as a traditional data center of comparable server count, and the industry is building dozens of such clusters this year alone.
That is why the September 9–11, 2026 edition of CIOE (the China International Optoelectronic Exposition, held in Shenzhen) felt less like a trade show and more like a futures market. According to on-the-ground reporting from China Science and Technology Innovation Daily (科创板日报) and Yicai (第一财经), exhibitors across module, chip, fiber, connector and test-equipment aisles used the same three phrases: “demand is larger than expected,” “orders queued into next year,” and “we are short on everything upstream.” One top-tier module maker was quoted as telling a buyer that orders under RMB 20 million (roughly USD 2.8 million) would simply not be accepted — a remarkable statement in an industry that, as recently as 2023, was still fighting price wars at 400G.
The structural backdrop is straightforward. Per LightCounting’s May 2026 vendor ranking, Chinese firms hold seven of the top ten global positions and a combined share of roughly 62–65%. In 800G-and-above high-speed datacom modules, their share already exceeds 70%. Zhongji Innolight (中际旭创) has held the number-one spot for four consecutive years, with an estimated 28–30% global share and 50–70% of the 1.6T silicon-photonics segment; Eoptolink (新易盛) has climbed to number two at 15–18%; HG Genuine (华工正源), Accelink (光迅科技), Hisense Broadband and Cambridge Industries (剑桥科技) fill out the next tier. For any overseas cloud provider, AI builder, telecom operator or systems integrator, sourcing 1.6T modules in 2026 is, in practice, sourcing from China.
2. The Numbers: An Industry Running Hot
To understand why capacity claims have to be read carefully, start with the sector-level figures that management teams and equity analysts are working from in Q3 2026.
2.1 Market size and share
- LightCounting projects the global optical-module market will exceed USD 26 billion in 2026, up more than 60% year-on-year. The datacom segment is doing most of the heavy lifting; telecom remains comparatively flat.
- China’s exports of integrated circuits (a category that includes optical module sub-assemblies) reached USD 103.5 billion in the first four months of 2026, up 83.7% by value (Xinhua / Qiushi English, June 2026). Export order books for optical modules were reported as stretching into 2028.
- Silicon-photonics-based modules are on track to exceed 50% of the market for the first time in 2026, meaning that process yield and packaging know-how — not just bill-of-materials — now determine who can ship at volume.
2.2 Capacity trajectory (industry-wide estimates)
Aggregating announcements from the publicly listed module makers and on-the-record commentary from CIOE exhibitors, the industry’s monthly capacity ramp over 2026 looks roughly as follows. These are consolidated China-plus-overseas figures; we will break down the domestic-vs-overseas split later.
| Generation | Est. monthly capacity early-2026 | Est. monthly capacity end-2026 (announced target) | Primary demand driver |
|---|---|---|---|
| 800G (DR8/FR8/2×FR4, EML & SiPh) | ~350,000–400,000 units | ~800,000–1,000,000 units | NVIDIA H200/B200 & in-service AI clusters; ongoing hyperscaler refresh |
| 1.6T (OSFP-XD DR8/2×FR4, 3nm DSP) | ~40,000 units | ~200,000 units | NVIDIA Rubin/Vera, Google TPU v6/v7, Microsoft Maia 2 ramp |
| 3.2T / 6.4T NPO/CPO | Demo / sample only | Pre-production validation | 2027–2028 next-gen platforms; Google 12M-unit NPO tender already placed |
Sources: LightCounting May 2026, Goldman Sachs Global Optics Report (Sept 7, 2026), Citi, CIOE 2026 exhibitor disclosures, and company H1 2026 filings.
2.3 Cambridge Industries: the poster child for multi-country expansion
Cambridge Industries Group (剑桥科技, SSE: 603083 / HKEX: 6166) is useful to look at in detail, because its 2026 disclosures spell out exactly what a genuine ramp looks like on paper:
- H1 2026 revenue reached RMB 2.705 billion (+32.9% YoY), with net profit of RMB 328 million (+171% YoY). High-speed optical modules accounted for RMB 1.675 billion of revenue, or more than 60% of the total.
- The company’s existing annualized optical-module capacity stood at roughly 3.5 million units at mid-year. Management stated at its August 25, 2026 results briefing that it targets no less than 6 million units of annualized capacity by year-end 2026.
- Geographic footprint is deliberately diversified: the Jiashan (Zhejiang) Phase-2 fab entered operation in March 2026; Malaysia Phase-2 is in production with yield up 15 percentage points year-on-year; the Vietnam line passed customer audit in H1 2026; Mexico cleanroom design is complete, with optical-module production targeted for early 2027. Management told Goldman Sachs at the September 2026 Asia Leaders Conference it expects overseas factories to account for more than half of capacity going forward.
- Product status: 800G modules are already shipping in volume to North American “core clients” (the company has publicly acknowledged supplying Microsoft and is widely reported to be qualifying with additional hyperscalers); 1.6T modules completed customer testing and verification in H1, with mass volume targeted in H2. Its 3nm-DSP-based 1.6T OSFP has a published power envelope of around 10 W.
The contrast between a company like Cambridge — publishing specific factory-by-factory milestones, yield improvements and customer audit status — and a second-tier vendor that issues a single press release announcing a “1.6T production line” could not be sharper. That contrast is exactly what a good due-diligence process is built to surface.
3. Three Capacity Distortions Every Buyer Must Unpack
When a Chinese module vendor tells you “we have 200k/month of 1.6T capacity by year-end,” there are three specific ambiguities you need to resolve before you treat that number as deliverable product.
Built Capacity vs. Planned Capacity
Built capacity means a cleanroom is certified, SMT lines are installed, test equipment (Keysight, EXFO, etc.) is calibrated, operators are trained, and the line has produced qualified modules at stated yield over multiple weeks. Planned capacity means a factory building may be leased, a press release issued, and some equipment on order — but the line is not yet shipping saleable product.
In 2026 the gap between the two is commonly six to nine months, and sometimes longer when 3nm DSP chips, indium phosphide (InP) lasers, or automated coupling stations are in short supply. Yunnan Germanium (云南锗业), a key InP substrate supplier, told CIOE 2026 attendees that the industry-wide InP supply gap is in the “millions of wafers” range — a bottleneck that sits squarely in front of any vendor’s headline capacity number.
Domestic Capacity vs. Overseas Capacity
The FCC’s July–September 2026 rulemaking process, which added optical transceivers to its Cover List while stopping short of a direct ban on major Chinese vendors, has sharpened every buyer’s focus on country of origin. A shipment from a Thai or Mexican subsidiary has different tariff, FCC and customer-acceptance implications from one exported directly from Jiangsu or Hubei.
When a vendor quotes “annual capacity,” always ask: how much of that can be delivered from a facility outside mainland China, in the form that your import compliance team will accept? As of Q3 2026, only a handful of Chinese vendors — most notably Innolight (Thailand/Tongling twin bases) and Eoptolink (Thailand Phase-2) — have overseas lines running at multi-million-unit annual scale. Several others have overseas facilities in various states of “construction,” “equipment installation” or “trial production,” which can mean anything from three months to eighteen months away from revenue.
Locked Capacity vs. Elastic Capacity
High-end optical modules are not generic widgets. A line that builds an 800G DR8 for one hyperscaler — using that customer’s qualified BOM, firmware, test vectors and labeling — cannot simply be repurposed to build a 1.6T 2×FR4 for a different buyer on Monday morning. Customer qualifications typically take 6–12 months for 1.6T, and once a line is qualified and reserved for a given customer, its output is effectively locked.
The “elastic” slice of a factory — capacity not yet committed to a specific customer, SKU and firmware — is the only slice you can realistically contract for on a short lead time. CIOE 2026 reporting and analyst commentary from Goldman Sachs both suggest that, at top-tier vendors, elastic 1.6T capacity is already mostly spoken for through 2027. The RMB 20-million-minimum-order anecdote refers specifically to this: small buyers are being turned away because the uncommitted slice is too thin.
4. The Major Chinese Vendors in 2026: Who Actually Ships?
Below is a pragmatic snapshot of the Chinese optical-module landscape as of Q3 2026, based on public filings, CIOE 2026 disclosures and analyst coverage. It is not investment advice; it is a starting map for procurement teams trying to decide which vendors deserve a deep diligence look.
| Vendor | Approx. global position | 800G status | 1.6T status | Overseas footprint (2026) |
|---|---|---|---|---|
| Zhongji Innolight 中际旭创 (SHE: 300308) |
#1, ~28–30% share | Large-scale shipping; >40% of global 800G SiPh | Scaling now; ~50–70% of global 1.6T SiPh; won 60% of Google’s 12M-unit NPO tender | Thailand + Tongling twin bases, >8M units/year combined |
| Eoptolink 新易盛 (SHE: 300502) |
#2, ~15–18% share | Bulk shipping; LPO route at ~45% gross margin | Small-batch shipments ramping; 3-year revenue targets RMB 50B/90B/150B | Thailand Phase-2 releasing; Chengdu base under expansion |
| HG Genuine (Huagong Zheng Yuan) 华工正源 (SHE: 000988 sub.) |
Top-5 tier | Bulk delivery; Wuhan Optics Valley park in full production | Claimed world’s first 1.6T mass production (Feb 2026); 12.8T XPO demo at CIOE | Wuhan main base; overseas channel expansion ongoing |
| Accelink 光迅科技 (SHE: 002281) |
Top-5 tier | Stable bulk delivery | Scaling shipments; world’s first 6.4T SiPh single-mode NPO demoed at CIOE | RMB 3.5B private placement for capacity expansion (4.99M new units/year) |
| Cambridge Industries 剑桥科技 (SSE: 603083) |
Rising first tier | Bulk to North American core clients; Microsoft acknowledged | Customer validation complete; mass ramp H2 2026; 6.4T/7.2T NPO sample | Jiashan + Malaysia operational; Vietnam audited; Mexico for early-2027 |
| Hisense Broadband, Source Photonics, TFC (天孚通信), Luxshare Tech, NeoPhotonics/Sols | Second tier / specialists | Varies; many shipping 800G | Mostly sampling / customer qualification; TFC active in optical engines | Mixed; some leveraging Malaysian/Thai partners |
Sources: LightCounting May 2026; company H1 2026 reports; China Daily / Xinhua Feb–Sept 2026; CIOE 2026 on-site coverage.
5. A Seven-Step Verification Playbook
Regardless of which vendor you are evaluating, these seven checks will take you from slide-deck confidence to a grounded view of what can actually be delivered.
Pull the official business license and registration file
Start with the AIC (Administration for Market Regulation) registration record for the contracting entity. Confirm the registered capital matches the scale of capacity being promised; check the registered address against the factory the vendor claims to operate from; look for any recent increases to business scope covering optical transceivers, silicon photonics, or optoelectronic device manufacturing. Mismatches between a vendor’s marketing materials and its registered scope are an early warning sign.
Verify overseas subsidiary registrations
If the contract will rely on Thailand/Mexico/Vietnam/Malaysia production, pull the local business-registration filings for those subsidiaries (Thai DBD, Mexican RFC/SAT, Vietnamese National Enterprise Registration Information System, Malaysian SSM). Confirm incorporation date, registered capital, listed business activities and the names of directors — these should all align with what the parent company has publicly disclosed.
Cross-check disclosed capacity against annual-report fixed assets
A 6-million-unit annual module line is a capital-intensive asset: cleanroom construction, automated coupling stations, burn-in chambers, 800G/1.6T test gear from Keysight/EXFO, and 24/7 staff all show up on a balance sheet. If a vendor claims massive new capacity but its H1 fixed-asset line and capex guidance do not support it, ask pointed questions.
Check patent and IP ownership
For 1.6T, key patents sit around silicon-photonics coupling, DSP driver interfaces, thermal packaging and LPO/NPO architectures. Cross-reference CNIPA filings with the product names the vendor is promoting. If a “1.6T OSFP” vendor has almost no patents in the relevant CPC subclasses filed in the last 24 months, it is likely integrating reference designs rather than controlling the process — which affects yield, customization and long-term support.
Ask for customer qualification evidence (not just logos)
Virtually every vendor lists the same handful of hyperscaler logos. Request the customer part number, qualification-revision letter, and first-article-inspection date for the exact SKU you would be buying. Be explicit about whether the qualification covers the domestic line or an overseas subsidiary line — a qualification on a Jiashan-built module does not automatically transfer to a Mexico-built unit.
Run administrative-compliance and litigation checks
Check the National Enterprise Credit Information Publicity System for operational anomalies, administrative penalties, pledged equity, customs penalties or environmental non-compliance — all of which can halt or delay an export-focused production line. For listed vendors, review stock-exchange inquiry letters (问询函), which often surface capacity, customer-concentration or related-party issues ahead of mainstream reporting.
Map the upstream single-point dependencies
Ask the vendor to disclose (under NDA if necessary) which DSP, laser, silicon-photonics die and fiber-array suppliers sit on the BOM for your SKU. Cross-verify that those suppliers are not already fully allocated to other module makers — and that the vendor has a qualified second source for each critical component. The 2026 shortage in InP substrates and 3nm DSPs means that a single-source dependency on one of these can quietly erase half the “headline capacity.”
6. What ChinaBizInsight Can Help You Verify
Public annual reports and CIOE press releases will get you only so far. Many of the data points that actually decide whether a module vendor can meet your commitment — the precise business-scope wording of a Mexican subsidiary, a customs penalty issued six months ago, a pledge on the parent company’s equity, a patent family recently transferred between related parties, or a winning bid record in a Chinese state-owned tender — live in fragmented Chinese government databases that are difficult to access and interpret from overseas.
This is the gap ChinaBizInsight’s due-diligence work is designed to close. We pull official registration records, shareholder and ultimate-beneficial-owner structures, administrative-compliance histories, IP filings, annual-report disclosures and bidding records into a single English-language report, so your procurement and legal teams can make capacity commitments based on primary-source evidence rather than PowerPoint claims. For multi-year optical-module supply agreements in particular, we recommend combining an official enterprise credit report with a focused on-the-ground verification of the specific factory that will produce your units.
You can browse the full range of our company-verification, document-retrieval and IP-check services on our products page, or contact us directly to scope a vendor-diligence engagement for a specific 800G or 1.6T sourcing project.
References
- LightCounting, Global Optical Module Vendor Ranking, May 2026 edition.
- Goldman Sachs, Cambridge Industries Management Meeting Notes — Asia Leaders Conference, September 4–7, 2026.
- Cambridge Industries Group (剑桥科技), 2026 Interim Results Presentation and Investor Briefing, August 25, 2026; institutional research minutes, September 8, 2026.
- Zhongji Innolight (中际旭创) and Eoptolink (新易盛), 2026 Interim Reports, filed August 2026 with the Shenzhen Stock Exchange.
- HG Genuine / HGTECH, “HG Genuine becomes leading optical module supplier,” China Daily / Optics Valley of China, February 13, 2026.
- China Science and Technology Innovation Daily (科创板日报) / Yicai, on-site reporting from the 27th China International Optoelectronic Exposition (CIOE), Shenzhen, September 9–12, 2026.
- Xinhua / Qiushi Journal English, “From optical modules to chips: China’s tech supply chains sustain global AI growth,” June 7–8, 2026.
- Goldman Sachs Global Investment Research, Global Optical Communications Sector Report, September 7, 2026; Citi and LightCounting 1.6T supply-demand estimates cited in Chinese business media September 2026.
ChinaBizInsight
Your strategic bridge to transparent business in China.