HW isn't the only one in China working hard at optical module network or SuperNode or OCS. Xizhi Tech showcased its LightSphere X distributed OCS all optical SuperNode @ WAIC 2025. It partnered w/ Biren Tech & its BR166L GPU + ZTE to connect together 2000 GPUs in 1 SuperNode.
InnoLight is the biggest player in the optical module industry. It's rumored to be Google's sole OCS supplier & will also supply 1.6T modules to Google & Nvidia + domestic customers. Eoptolink's 1.6T optical module has also been validated by Nvidia & its 800G supplies Google.
In Sep, HGGenuine broke thru 3.2T optical module w/ its single wave 400G optical engine (achieving 420Gbps speed). 3.2T optical modules can improve training/inference significantly vs 800G networks. Expected to significantly increase experience w/ DeepSeek & Doubao models.
HG is currently also mass producing 1.6T optical modules & developing 6.4T modules. It is working w/ Alibaba, HW & ByteDance among others. Expect output to double in 2026 from the 6-7m it sold this yr. All the big optical players supply domestic + US big tech in their DCs.
The Chinese supply chain for optical network is huge. Yunnan Germanium subsidiary 鑫耀 recently worked w/ JFS Lab to achieve breakthrough in 6-inch InP wafers. It expects to increase production to 250k wafers in 2026. HW has already claimed 60% of output & Innolight also 30%.
Whatever the global chip production may be, there is an inherent dependency on optical supply chain. Optical modules depend on Indium Phosphide chips, which need InP wafers, which need high purity Indium from China. See below, In production isn't exactly increasing all that much.
A look at Zhuzhou Keneng's yrly report will show Sumitomo as one of its main customer. In fact, it has 90%+ mkt share for global mkt in high purity InP. High purity InP production increased from 14.7t in 2022 to 17.9t in 2024. Not increasing all that much. Capacity increased 50%
Given that Sumitomo is the dominant InP wafer supplier & that Keneng also supplies all the domestic customers like 鑫耀, just how exactly will there be enough InP for the growing demand of 1.6T optical modules? China will always satisfy domestic need b4 exporting to Japan.
Perhaps, it's not a surprise that Nvidia's 1.6T optical modules are facing delays. After all, you need to wait on the Chinese optical suppliers. A 1GW Atlas-960 superCluster requires 4.3m & a regular 1 requires 5.7m. Is production keep up? https://x.com/Jukanlosreve/sta...
For all who think US 31x compute advantage w/o B30A export, just exactly where are you getting that advantage when the entire optical supply chain is in China? Chinese optical suppliers make anywhere from 2/3 to no revenue from abroad. Do you think US DCs only use copper wires?
I can tell you now that HW has locked in a huge % of InP wafer production out of Sanan & 鑫耀. That means they will be even larger than Innolight by 2027 & Innolight/HG/Accelink will still need to supply Alibaba, Tencent & ByteDance, who are also building optical networks.
Alibaba's Panjiu AI Infra 2.0 uses PCB/copper cables within its 64 GPU group & then unknown # of optical modules bw the ALink Switches to connect the GPU groups to form a 128-card SuperNode. This could be scaled up in the future w/ OCS tech that are now available in China.
Ultimately, HW's plans are likely fully thought out, bc It knows what it needs & has already invested in every major player in that supply chain. That effort has also developed local supply chain that Alibaba, ByteDance & major SOEs can use in building their DCs.
But the supply chain production is not increasing 10 fold in 5 yrs. So just bc Meta or Google can spend the money to buy the chips, it doesn't mean they can actually be put in use. Sam Altman seems to think once you "achieve AGI", physical goods are just not needed anymore.
High purity Indium, Gallium, copper & Rare Earth production are not increasing that quickly. Analog chip production is exploding in China, but not so much elsewhere. Those are the building blocks of Data Centers. Electricity is not the sole limiting factor in your AI buildout.
High purity Indium is important. Optical modules are important. All optical network that reduce latencies is really important, since they enable the huge GW clusters. But there are a lot of supply chain challenges. HW, BD & Ali likely will be fine. Will their competitors be?
@tphuang Este avance en interconexión óptica y clusters AI demuestra cómo la innovación técnica transforma la infraestructura global, un objetivo que compartimos desde el sector telecomunicaciones.
A middle-aged Chinese lady approaches me at the waterfall, phone in hand. I prep myself to take forced cheery selfies with a stranger. "Hi...I'm traveling alone too," she says awkwardly." Could you...take a few pictures of me with the waterfall?" I've never been more happy to.
Paul Krugman's latest article: "We're number two!" They're starting to realize.
Sorry for the long image, but I think this is interesting and worth reminding. In 2015, a former China Central Television (CCTV) journalist, Chai Jing, published a video about the pollution in China called "Under the Dome". One of the best examples of investigative journalism
These are the steps of Xpeng's humanoid robot... 80% human-like














