Huawei's Hubble Invests Across the Full InP Optical Chip Chain, Betting on AI Interconnect Infrastructure
Nashnova编辑部
Huawei's venture arm Hubble Technology has systematically invested across China's photonics supply chain since 2019 — from lasers and detectors to InP substrates — positioning for the explosive demand for 800G-to-3.2T optical interconnects in AI data centers.
What has Hubble invested in, and what picture does it form?
Hubble has backed at least four photonics companies: Vertilite (VCSEL laser chips — vertical-cavity lasers used for short-range, high-speed optical links), Yuanjie Semiconductor (DFB laser chips — the most common light source inside data-center optical modules), NOP / Northern Photonics (micro-nano optics and optical integration), and Milphoton Semiconductor (InP photodetectors).
This means → Hubble is not making scattered bets on isolated components. It is building along one chain: emit light → receive light → integrate the optical path → secure the base material.
In plain terms = for every critical component a light signal passes through from source to destination, Hubble has placed a portfolio company.
What gap does Milphoton fill?
In May 2026, Milphoton completed an angel round with Hubble participating. Funds go toward R&D and pilot-scale production lines.
Milphoton specializes in InP-based photodetectors — InP, or indium phosphide, is the core semiconductor material for high-speed lasers and detectors. Target applications span 800G, 1.6T, and 3.2T optical modules and AI data-center interconnects.
This means → Hubble already had the transmit side (lasers) and optical integration covered. The receive side (detectors) was the missing piece — Milphoton fills it.
Why has InP suddenly become a bottleneck?
InP is the essential material for high-performance lasers and detectors operating at 1310 nm and 1550 nm — the wavelengths that make low-loss, long-distance fiber transmission possible.
Lumentum CEO Michael Hurlston said the InP substrate shortage now exceeds that of DRAM and NAND, with some customers' orders queued to 2028.
Yole estimates 2026 global demand at 2.6–3.0 million 2-inch-equivalent wafers; effective capacity is only about 750,000 — a supply gap above 70%.
In plain terms = every AI data center on Earth is racing to buy optical modules, yet the raw material those modules depend on covers less than a third of demand. Whoever holds the supply holds the leverage.
How are global players responding to the gap?
JX Advanced Metals plans to spend ¥120 billion to grow InP capacity seven- to tenfold by FY2030. AXT has raised roughly $632.5 million for expansion. Sumitomo Electric targets a 3.1× capacity increase by FY2028 versus FY2024.
Nvidia stepped in directly: in March 2026 it committed $2 billion each to Coherent and Lumentum under long-term supply agreements. Coherent then broke ground on a new InP wafer fab in Texas.
This reflects a shift: InP is no longer just a materials issue — it is an infrastructure-level bottleneck in the AI compute arms race. Industry estimates put over 80% of current InP demand as AI-data-center-related.
How does Hubble solve the "designs but no material" problem?
In December 2020, Huawei Investment took a 23.91% stake in Yunnan Xinyao Semiconductor Materials, now one of China's largest InP wafer producers.
By 2025, HiSilicon had become Xinyao's largest customer, accounting for roughly 40% of its shipments.
This means → from chip design down to raw substrates, Hubble's vertical integration is now end-to-end: Xinyao supplies wafers → Milphoton makes detectors → Vertilite / Yuanjie make lasers → NOP handles optical integration — a closed loop.
Can this chain deliver?
The strategic logic is clear: during a window when global InP capacity is severely short and expansion cycles run years long, lock in upstream supply early to backstop Huawei's own optical-transmission, optical-access, and 800G data-center products.
The risks are equally clear: Xinyao's capacity still lags global leaders, and Milphoton is at the angel stage — the leap from pilot line to volume production requires time and yield verification.
In plain terms = the board is set, but not every piece has landed. Whether this chain can deliver stable domestic alternatives during the tightest supply years is the real test of the strategy.
Content is for reference only, not financial advice.