To The Moon

AI infrastructure is entering a new phase.

In the first phase, the market was trading one question:

Who has the most GPUs?

Then the focus shifted to data centers.

Who can secure enough power, land, cooling, servers, and deployment capacity?

But now, the question is changing again.

As AI Mega Clusters move from hundreds of thousands of GPUs toward millions of GPUs, the bottleneck is no longer just GPUs. It is no longer just power. The next bottleneck is whether multiple data centers can be connected into one unified AI training cluster.

That is Scale-Across.

SemiAnalysis recently published a deep dive on this topic. Their core view is clear: future AI clusters cannot rely forever on the expansion of a single campus. More cloud providers will have to connect multiple data centers, multiple campuses, and even multiple regions into one larger AI training network.

This matters a lot for the market.

Because the AI infrastructure trade is moving from Scale-Up and Scale-Out toward Scale-Across.

In the first phase, the market traded chips.

If you had GPUs, you had compute.

In the second phase, the market traded data centers.

If you had power, land, cooling, and server capacity, you could scale faster.

In the third phase, the market will increasingly trade networks.

Because when a single data center can no longer hold enough GPUs, and when a single campus can no longer secure enough power, AI clusters have to expand across multiple data centers.

That is where optical networking, data center interconnect, low-latency transmission, DWDM, coherent optics, and optical transport equipment become much more important.

That is why I think Scale-Across is bullish for $诺基亚(NOK)$ and $Ciena科技(CIEN)$

The logic is not simply:

“AI is growing, so networks will grow too.”

The real logic is this:

The larger AI clusters become, the more data center interconnect stops being ordinary networking and starts becoming part of the training system itself.

In the past, data center interconnect was mainly about traffic transfer, cloud synchronization, disaster recovery, and content delivery.

But Scale-Across is different.

It is not just about moving data from Data Center A to Data Center B.

It is about whether multiple data centers can work together like one larger AI cluster.

That creates much higher demands on the network.

First, bandwidth has to increase.

AI training is not ordinary internet traffic. Large-scale model training involves parameter synchronization, gradient communication, and massive data exchange across nodes. The more GPUs you add, the heavier the communication burden becomes.

Second, latency has to fall.

If latency between data centers is too high, GPUs spend more time waiting. Utilization drops.

For cloud providers, the most expensive problem is not buying GPUs.

The most expensive problem is buying GPUs and then having those GPUs wait on the network.

Third, reliability becomes critical.

AI training clusters are not ordinary business systems. If a training job gets interrupted, the cost is not just lost bandwidth. It is lost time, wasted compute, and massive electricity costs.

At the scale of tens of thousands, hundreds of thousands, or even millions of GPUs, network stability directly affects training efficiency.

Fourth, power and density matter.

AI data centers are already constrained by power and cooling. If networking equipment consumes too much power or takes up too much space, it competes directly with GPU deployment.

So the next generation of optical networking equipment cannot just be faster.

It also has to be more power-efficient, higher-density, and more automated.

That is where $CIEN and $NOK come in.

Ciena is already a key player in data center interconnect and coherent optical transport. Its products are directly tied to the kind of high-capacity, low-latency optical infrastructure that Scale-Across requires.

Nokia has a similar logic.

Nokia has a full optical networking and data center interconnect portfolio, including DWDM, coherent optical engines, optical line systems, and intelligent automation.

So this is not just another AI buzzword trade.

It reflects a real structural shift in AI infrastructure.

Previously, the key question was:

Who can provide the GPUs?

Now the question is becoming:

Who can make more GPUs work together efficiently across a much larger physical footprint?

That shift moves part of the value chain from chips alone into networking, optical modules, optical transport, and data center interconnect.

Of course, the stock prices may not reflect this immediately.

$NOK and $CIEN are not the kind of pure AI names that usually move 30% in a single day.

They are more like second-layer AI infrastructure beneficiaries.

The market first traded Nvidia, Broadcom, and Arista because they are closer to the core of compute and AI networking.

But as AI Mega Clusters continue to expand, the market will eventually have to ask a new question:

If one data center is no longer enough, who connects the next one?

Ciena and Nokia are part of that answer.

The key point is not simply that “optical networking benefits from AI.”

The key point is that AI infrastructure is moving from single-site expansion to multi-data-center coordination.

That creates a new capital expenditure direction.

GPUs need to be connected.

Data centers need to be connected.

Cities need to be connected.

Regions may need to be connected.

The bigger AI becomes, the more important the network becomes.

In the past, the market traded compute itself.

In the next phase, the market will increasingly trade the efficiency of connecting compute.

This is not the end of the AI infrastructure cycle.

It is the expansion of the AI infrastructure cycle from chips into networks.

From Scale-Up, to Scale-Out, and now to Scale-Across.

Phase one was about GPUs.

Phase two was about data centers.

Phase three will be about who can connect multiple data centers into one true AI super cluster.

That is the opportunity for $NOK and $CIEN.