From 800G to 1.6T to 3.2T — The Breakneck Speed of Optical Transceiver Technology Upgrades
If you think optical transceivers are just another commodity component, think again. In the AI era, they’ve become the critical bottleneck — and the pace of their evolution is unlike anything the industry has ever seen.
Just a few years ago, the optical transceiver industry moved at a predictable, almost leisurely pace. A new generation of products arrived every three to four years[reference:0]. Engineers had time to perfect designs. Supply chains had time to adjust. Customers had time to plan.
Not anymore.
Today, the industry is hurtling forward at breakneck speed. Product iteration cycles have compressed from 3–4 years to roughly 2 years[reference:1][reference:2]. The transition from 800G to 1.6T is happening in what feels like the blink of an eye. And 3.2T is already on the horizon, with the first products already demonstrated and validated[reference:3].
For overseas buyers, investors, and technology partners evaluating Chinese optical transceiver companies, understanding this accelerated roadmap isn’t optional — it’s essential. Here’s what you need to know about where the technology is today, where it’s going, and what it means for your business.
Iteration cycle: 3–4 years → ~2 years · Rate doubling: every 2 years vs. every 4 years[reference:4]
800G — The Workhorse of the AI Era
If there’s one product that defines the current optical transceiver market, it’s 800G. 800G has become the core growth driver for the entire industry[reference:5], and 2026 is shaping up to be its biggest year yet.
According to LightCounting, 800G shipments are expected to more than double in 2026[reference:6][reference:7]. The market is so strong that Goldman Sachs raised its 2026 800G volume forecast from 25 million units to 33.5 million units — a 58% upward revision[reference:8].
Demand is being driven by the hyperscalers. NVIDIA has reportedly increased its 800G procurement forecast for 2026 by 35%, with Chinese suppliers 中际旭创 (InnoLight) and 新易盛 (Eoptolink) together capturing 58% of the增量 orders[reference:9]. Overall, Chinese manufacturers now account for an estimated 70% of the global 800G market[reference:10].
800G at a Glance (2026)
- Shipments: Expected to more than double year-over-year[reference:11]
- Market forecast: 33.5 million units (Goldman Sachs estimate)[reference:12]
- Chinese share: ~70% of global market[reference:13]
- Key drivers: AI cluster buildout, hyperscaler capex[reference:14]
1.6T — The Commercial Takeoff
2026 is the year 1.6T goes from pilot to production. The technology has been in development for years, but this is the year it crosses the threshold into full-scale commercial deployment[reference:15][reference:16].
LightCounting now expects 1.6T shipments to grow to tens of millions of ports in 2026, up from a small base in 2025[reference:17][reference:18]. The ramp is happening faster than almost anyone predicted. Coherent reported in August 2026 that its 1.6T ramp is “even faster than expected three months ago”[reference:19]. Lumentum’s first 1.6T modules have begun shipping, with significant demand growth expected through 2027[reference:20].
On the Chinese front, 光迅科技 (Accelink) has announced that its 1.6T optical modules are now capable of mass delivery[reference:21]. 中际旭创 reported that its Q1 2026 revenue grew 207% year-over-year, driven primarily by 1.6T shipments[reference:22]. The company is also preparing 6.4T NPO and 12.8T XPO products for R&D and customer testing[reference:23].
Market estimates for 1.6T demand in 2026 range from 8.6 million to 20 million units[reference:24]. At typical pricing, the North American market alone is already seeing orders exceeding $20 billion in value[reference:25]. This isn’t a niche product — it’s the next mainstream.
Key takeaway: 1.6T is not a future technology — it’s shipping today. If your supply chain or investment strategy doesn’t account for 1.6T, you’re already behind.
3.2T — The Next Frontier
While 800G and 1.6T dominate today’s headlines, the industry is already looking further ahead. 3.2T is the next major milestone, and the technology pathways are already becoming clear[reference:26].
LightCounting expects 3.2T optical modules to begin gradually ramping from 2028 onward[reference:27][reference:28]. The market is projected to reach $1.4 billion by 2028 and $24 billion by 2031[reference:29] — a trajectory that mirrors the explosive growth we’ve seen with 800G.
The technology path for 3.2T is taking shape along two parallel tracks:
- EML-based single-wave 400G方案 — expected to be the first to market, leveraging mature EML laser technology[reference:30]
- Silicon photonics single-wave 400G — approaching the threshold of commercial viability[reference:31]
Chinese companies are already staking their claims. 光迅科技 (Accelink) unveiled the world’s first 3.2T silicon photonic single-mode NPO module at OFC 2026 in March, and has completed system verification with leading Chinese cloud service providers[reference:32][reference:33]. 华工正源 (Huagong Tech) also announced a 3.2T NPO product at its AI strategy conference, describing it as “the world’s first” and noting it has already been deployed with industry-leading customers[reference:34].
3.2T Market Outlook
- 2028: ~$1.4 billion market[reference:35]
- 2031: ~$24 billion market[reference:36]
- First movers: 光迅科技 (Accelink), 华工正源 (Huagong Tech)[reference:37][reference:38]
- Key enabler: Single-wave 400G modulation technology[reference:39]
The Packaging Evolution — From Pluggable to CPO
Speed isn’t the only thing changing. The physical architecture of optical transceivers is undergoing a fundamental transformation — one that will reshape the entire industry over the next decade[reference:40].
The evolution follows a clear trajectory: Traditional Pluggable → LPO → NPO → CPO. Each step brings the optical engine closer to the switching ASIC, shortening electrical paths, reducing power consumption, and enabling higher bandwidth density.
| Architecture | Key Feature | Power vs. Traditional | Status (2026) |
|---|---|---|---|
| Traditional Pluggable | With DSP, long electrical paths | Baseline | Mainstream |
| LPO | No DSP, linear drive | ↓ ~50% | Early deployment[reference:41] |
| NPO | Optical engine near ASIC | ↓ significant | Commercial orders[reference:42] |
| CPO | Co-packaged with ASIC | ↓ ~80% | Mass production[reference:43] |
LPO — The DSP Killer
LPO (Linear Pluggable Optics) removes the DSP chip from the module, shifting signal processing to the host ASIC. The benefits are dramatic: power consumption drops by approximately 50%, latency falls by up to 83%, and cost decreases significantly[reference:44][reference:45].
Chinese cloud service providers have already incorporated 800G LPO into their 2026 volume deployment plans[reference:46]. 索尔思光电 (Source Photonics) claims to be the world’s only manufacturer shipping 1.6T LPO modules in volume, with Meta and NVIDIA certification already secured[reference:47].
NPO — The Bridge to the Future
NPO (Near-Package Optics) moves the optical engine from the front panel to a location close to the switch ASIC, shortening electrical paths from decimeters to centimeters[reference:48]. It offers many of the benefits of CPO while maintaining better serviceability and leveraging existing PCB and thermal management systems.
The commercial momentum is real. In early 2026, Google placed a formal order for 12 million NPO optical modules for its next-generation TPU v7/v8 clusters, with delivery concentrated from Q3 2026 to Q2 2027[reference:49][reference:50]. The order is valued at approximately $12–15 billion, with Chinese suppliers capturing nearly all of it — 中际旭创 secured 60% (7.2 million units) and 新易盛 40% (4.8 million units)[reference:51].
Key takeaway: NPO is no longer experimental — it’s being deployed at hyperscale by the world’s largest cloud providers. The Google order alone is a game-changer.
CPO — The Ultimate Destination
CPO (Co-Packaged Optics) represents the ultimate integration — the optical engine and switch ASIC are co-packaged on the same substrate, minimizing electrical paths, maximizing bandwidth density, and dramatically reducing power consumption[reference:52].
In August 2026, NVIDIA announced that its Spectrum-X Ethernet silicon photonic switch — the world’s first mass-produced 200G/lane CPO Ethernet switch system — has entered full production[reference:53]. The results are staggering: laser count reduced by 75%, power consumption cut by 80%, and mean time between failures improved 10x[reference:54][reference:55]. The flagship SN6800 model can be configured with 512 800Gb/s ports or 2,048 200Gb/s ports, delivering total bandwidth of 409.6Tb/s[reference:56].
CPO Impact — NVIDIA Spectrum-X
- Laser count: ↓ 75%[reference:57]
- Power consumption: ↓ 80%[reference:58]
- MTBF: ↑ 10x[reference:59]
- Total bandwidth: 409.6Tb/s[reference:60]
Silicon Photonics — The Tipping Point
Perhaps the most significant technology milestone of 2026 is happening in materials. 2026 is the year silicon photonics goes mainstream.
According to LightCounting, 2026 will be the first year that sales of optical transceivers using silicon photonic modulators exceed 50% of the total market[reference:61][reference:62]. The market for optical chips — including silicon photonics, InP, and other materials — is projected to grow from $4 billion in 2025 to $15 billion by 2031[reference:63].
Silicon photonics chips, which accounted for one-third of all optical chips in 2025, will contribute 42% ($6.3 billion) of the market by 2031[reference:64]. The deployment of NPO and CPO is accelerating silicon photonics adoption and driving demand for InP continuous-wave lasers[reference:65].
What’s driving this shift? LightCounting puts it bluntly: major CMOS foundries like Samsung and TSMC have entered the silicon photonics market — not for the $6.3 billion chip market, but because they see the trillion-dollar semiconductor industry shifting from copper to optical interconnects[reference:66]. We are, as the analysts say, “at the very early stages of this transition”[reference:67].
Key takeaway: Silicon photonics isn’t just a niche technology anymore — it’s becoming the default. Companies that haven’t mastered silicon photonics are at risk of being left behind.
What This Means for Your Business
The optical transceiver industry is in the midst of its most dramatic transformation in decades. Product cycles are compressing. Technologies are converging. And the competitive landscape is shifting faster than ever.
For overseas businesses evaluating Chinese suppliers in this space, the stakes are high. The companies that lead in 800G today may not be the ones that lead in 1.6T tomorrow. The ones that master silicon photonics and NPO/CPO today will have a decisive advantage in the 3.2T era.
But evaluating these companies requires more than just reading headlines. You need to know: Are they actually shipping 1.6T in volume, or just announcing samples? Do they have silicon photonics capabilities, or are they still dependent on EML? Have they secured NPO orders from hyperscalers, or are they still waiting?
These aren’t just technical questions — they’re business questions. And they require reliable, verifiable intelligence.
ChinaBizInsight — Know Your Chinese Partners. We provide authoritative company credit reports, business registration documents, due diligence investigations, and apostille/legalization services for Chinese companies — including the full spectrum of optical transceiver manufacturers and their supply chain partners.
Our reports dig deep into registered capital, shareholder structures, legal disputes, patent portfolios, and executive backgrounds — giving you the confidence to make informed decisions about technology partners and suppliers.
The Bottom Line
We are living through a rare moment in the optical transceiver industry. Product cycles have compressed from 4 years to 2. 800G is exploding. 1.6T is taking off. 3.2T is on the horizon. LPO, NPO, and CPO are reshaping the physical architecture. And silicon photonics has crossed the 50% threshold.
Chinese companies are at the center of all of it — leading in market share, driving technology development, and capturing the lion’s share of hyperscale orders. But with speed comes complexity. And with complexity comes risk.
Make sure you have the intelligence you need to navigate this fast-moving landscape. Know your Chinese partners. Before you commit.
ChinaBizInsight — Know Your Chinese Partners. Visit our website to learn how we can support your China business intelligence needs.
Data Sources
- LightCounting, “Optics for AI Clusters” (January 2026) and “Silicon Photonics, LPO/LRO and NPO/CPO” (May 2026)
- LightTrends Newsletter, February 2026
- Company announcements and investor relations: 光迅科技 (002281), 中际旭创 (300308), 华工科技 (000988)
- OFC 2026 conference proceedings
- 前瞻产业研究院, “2026年光模块行业蓝皮书”
- Various financial research reports (Goldman Sachs, CITIC Securities, Huatai Securities)
All data and projections are based on publicly available information as of August 2026.
ChinaBizInsight
Your strategic bridge to transparent business in China.