AIToday

Goldman Sachs: AI networking market to surge to $154 billion(約25兆円) by 2028

Top Companies AI — US (1/2)2h ago
Goldman Sachs: AI networking market to surge to $154 billion(約25兆円) by 2028

Key takeaway

Goldman Sachs predicts the AI networking market will balloon from $15 billion(約2.4兆円) to $154 billion(約25兆円) by 2028 as optical technology replaces copper in data-center infrastructure. The shift reflects a fundamental bottleneck: moving data between thousands of AI processors is becoming as critical as the processors themselves, creating years of potential demand growth for optical component makers.

Summaries like this, in your inbox every morning.

Sign up free →

3 Key Points

  • What happened

    Goldman Sachs projects the AI networking addressable market will grow from roughly $15 billion(約2.4兆円) during the GB300 cycle to approximately $154 billion(約25兆円) by 2028, driven by optical technologies replacing copper as processors require faster data connections. Networking content per computing unit is projected to expand nearly 29-fold, from roughly $315,000 in GB300 NVL72 architecture to approximately $9.4 million(約15億円) in Rubin Ultra NVL576 systems.

  • Why it matters

    As AI clusters outgrow traditional electrical connections, optical components—including pluggable transceivers and co-packaged optics (CPO)—are becoming essential infrastructure. Goldman estimates the pluggable optical modules market alone could increase about 10-fold, while the broader optical components market expands roughly 13-fold, making suppliers like Lumentum potential beneficiaries of this shift.

  • What to watch

    Goldman expects co-packaged optics to capture 59% of AI networking spend by 2028, though slower CPO adoption timelines remain the primary investor risk.

Context & Analysis

Goldman Sachs' April 2026 report identifies networking as the next major investment cycle in AI infrastructure, shifting focus beyond the GPU dominance of the past three years. The thesis rests on a fundamental constraint: as Nvidia transitions through successive chip generations—from Blackwell to Vera Rubin and eventually Rubin Ultra—the volume of data flowing between thousands of processors is expanding far faster than compute capacity alone. Networking value per system is projected to climb nearly 29-fold, forcing a wholesale migration from electrical to optical interconnects.

This transition has material implications for the optical components supply chain. The pluggable optical modules market alone could grow roughly 10-fold, while the total optical components market expands roughly 13-fold as optics migrate beyond traditional rack-to-rack networking into shorter-reach connections inside AI systems themselves. Co-packaged optics—where optical modules are integrated directly into chip packaging—are expected to capture 59% of AI networking spend by 2028, though the primary risk to investors is slower-than-projected adoption timelines. For companies like Lumentum, whose business focuses nearly entirely on optical communications, the addressable market expansion from $15 billion(約2.4兆円) to $154 billion(約25兆円) represents potential revenue sensitivity far greater than diversified competitors.

FAQ

How much will networking content increase per AI computing unit?
Networking content per computing unit is projected to expand from roughly $315,000 in today's GB300 NVL72 architecture to approximately $9.4 million(約15億円) in Rubin Ultra NVL576 systems, representing a nearly 29-fold increase.
What share of AI networking spend will co-packaged optics capture?
Goldman expects co-packaged optics to capture 59% of AI networking spend by 2028, though slower CPO adoption timelines remain the primary investor risk.
Why is optical technology replacing copper in AI data centers?
Copper connections increasingly struggle with higher bandwidths, longer distances, and rising power consumption as AI clusters grow, making optical technologies the preferred solution for moving data at light speed.

Discussion

No comments yet. Be the first to share your thoughts!

Log in to join the discussion

Related Articles

Stay ahead with AI news

Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.

Get Started Free

Free · takes 30 seconds · unsubscribe anytime

1 minute a day. The AI essentials.

200+ sources · Email / LINE / Slack

Get it free →