Optical networking test isn’t one big TAM — it’s TWO different control points in the AI reliability stack.
Two companies. Two choke points:
• Upstream → Kill defects early (Aehr Test Systems)
• Downstream → Validate full AI fabric (VIAVI Solutions)
As silicon photonics + CPO scale, spending is splitting:
Convex vs recurring
Why test intensity is exploding?
Optics are moving closer to compute
Pluggables → Optical I/O → CPO
• TSMC → COUPE (2026)
• Broadcom → 200G/lane CPO
• NVIDIA → optical inside switches
Now ONE defect can break:
packaging + fiber + entire racks
Cost of escape = 💥 massive
Enter Aehr Test Systems ($AEHR):
The upstream bet → wafer-level burn-in
Kill failures BEFORE advanced packaging
Recent catalysts:
• Follow-on SiPh order
• New global networking customer
Power scaling:
1kW → 3.5kW per wafer
Screening is getting brutal
If burn-in becomes standard → convex upside
Now VIAVI Solutions ($VIAV):
The downstream play → validate the FULL AI fabric
Not just links. Not just devices.
👉 The entire system under real load
Think:
• Thousands of NVIDIA GPUs
• 800G → 1.6T optical links
• Switches, NICs, congestion
• Real AI workloads
Simple question they answer:
👉 “Will this AI cluster run at full speed without breaking?”
Why this matters (and why it’s BIG):
AI systems are fragile at scale:
• One slow link → bottlenecks thousands of GPUs
• Packet loss → retransmissions → latency spikes
• Congestion → idle accelerators
You don’t lose 5%…
You lose cluster efficiency
A $100M cluster at 80% utilization
= $20M wasted
That’s why validation = mission critical
What VIAVI Solutions actually does:
• Simulates REAL AI traffic (not synthetic tests)
• Stress-tests before deployment
• Certifies networks at install
• Monitors live systems (AIOps)
Spirent adds:
→ 1.6T testing
→ RoCEv2 congestion
→ AI workload emulation
They don’t just test —
they protect utilization
Recurring revenue = moat
How they fit together:
Upstream (Aehr Test Systems):
👉 “Is the chip clean?”
Downstream (VIAVI Solutions):
👉 “Does the system run perfectly at scale?”
Better screening → fewer failures
But BOTH are needed as 800G / 1.6T + CPO ramps
Different layer. Different dollars.
Bottom line:
“Optical test” is splitting into TWO economies:
• Upstream → defect prevention (AEHR)
• Downstream → utilization assurance (VIAV)
AI data centers now run on optics
The winners?
Whoever owns the reliability chokepoints
So — what’s your bet?
Early kill ⚡ or full-fabric assurance 🔁
Or both? For me Both
@TigerObserver @TigerPM @Tiger_comments @TigerStars @Daily_Discussion
Comments