Huang Renxun said at Computex that the end of AI is connectivity - that's why I've been holding NOK all along Tell me something that I don't think the market has fully priced in yet In the past three years, the direction of speculation has been GPU, HBM, ASIC, and advanced packaging. There is only one logic behind it: how to fit more computing power into one cabinet. This stage is called Scale Up, and now it is basically a consensus. But Scale Up has a ceiling. The AI factory is getting bigger and bigger, and a single data center can no longer sustain the future training scale. Electricity, land, cooling, grid connection, and construction are all hard constraints that cannot be solved by spending money. So in the future, a large-scale AI cluster will not only exist in one building, but will be distributed across multiple data centers, parks, and even geographical locations to train the same model together. This direction has been appearing more and more frequently in the financial conference calls of major cloud providers recently, and there is a term called Scale Across. Once Scale Crosse becomes mainstream, the importance of connectivity begins to truly approach computation itself for the first time. Previously, GPUs determined everything, and networks were not as important. But the logic of Scale Crosse is reversed, where for every doubling of GPU, the connection requirements may increase by two or even more times. Because every time a new computing node is added, it needs to communicate with more existing nodes, which is an exponential increase in demand. So recently you will see a group of seemingly completely different companies starting to talk about the same thing: some talk about distributed AI training, some talk about data center interconnection, some talk about backbone fiber optics, and some talk about Pump Laser. These are not isolated stories, behind them lies the same logic unfolding. Murphy's mention of The Distance Free Datacenter in Computex is the ultimate form of this logic: GPU pooling, storage pooling, CPU pooling, not necessarily adjacent to each other, as long as the connection is fast enough, the entire system still looks like a supercomputer from the outside. And the infrastructure that enables all of this is light. Nowadays, many people are discussing optical communication, which is still at the first level The path from 800G to 1.6T to 3.2T to CPO is correct, but it is only the beginning of the entire story. The complete connectivity revolution is divided into three layers: the connection inside the cabinet is Scale Up, the optical module and switch solve the problem of Scale Out inside the data center, and the connection across data centers and campuses is Scale Akross. This layer requires coherent optical DSP and backbone optical transmission system. CPO solves the first layer, but Scale Crosse requires the third layer. This is an area where the market has not yet fully priced, with only a partial price in. The first level Scale up has a basic price in, the second level Scale out has most of the price in, and the third level Scale across only has a portion of the price in. Why is Nokia worth paying attention to Many people still have the impression of Nokia as a traditional communication equipment company, which is the fundamental reason why the market still prices it relatively low compared to CIEN (CIEN forward pe 139, NOK forward pe 43.8). Last year, Nokia acquired Infinera. The core asset of Infinera is the self-developed capability of coherent optical DSP, which happens to be the most critical technical link for data center interconnection and backbone optical transmission. After the acquisition, Nokia acquired coherent optical DSP chips, pluggable optical modules, and backbone optical transmission line systems, covering the entire chain from devices to systems. The 1.6T is already in mass production, while the 2.4T and 3.2T are currently undergoing testing. This vertical depth is something that most pure optical module manufacturers cannot achieve. On the performance level, the orders of AI and Cloud customers increased by 49% year-on-year in the first quarter of this year, with a single quarter increase of about 1 billion euros, and the full year guidance was raised. This is already being fulfilled. But the market is still pricing it using the valuation logic of traditional communication companies. There are more and more narratives stacked on Nokia now: optical communication, AI data center interconnection AI-RAN、 Edge AI, Physical AI, NVIDIA ecosystem collaboration. Most of these directions have not yet entered the valuation model. There is another supply chain bottleneck that has not been fully discussed When discussing optical communication, people usually focus on optical modules, DSPs, and EML lasers. But I think what is truly lacking is InP, indium phosphide. Pump Laser、Narrow Linewidth Laser、Gain Chip、 Coherent devices, optical amplifiers, and these key devices almost all rely on InP. The wafer production capacity of InP is not on the same scale as that of silicon, and the expansion cycle is also longer. In the past few quarters, Lumentum, Coherent, Nokia, Fabrinet, and Applied Optoelectronics have all mentioned similar issues in their financial reports: extended laser lead times, inability to keep up with InP production capacity, and backlog of orders. This is why Nokia's acquisition of Infinera is important. One of Infinera's core assets is its ability to vertically integrate InP wafers. It's not just about buying InP devices, it can make them on its own. The entire chain from InP wafers to coherent optical DSP chips to pluggable optical modules is interconnected internally. After Nokia acquired Infinera, it was equivalent to bringing in this capability together. This means that while the entire industry is clamoring for InP shortages and longer laser lead times, Nokia has an advantage that others do not have - it does not rely entirely on external supply chains, and some key components can be supplied to itself. In a tight supply chain environment, the value of this capability will be repriced. The chain of AI expansion is very clear, with an increase in GPUs, cabinets, data centers, campuses, and cross campus connectivity requirements. Coherent demand is exploding, and InP production capacity has become a new bottleneck. The market's focus will spread from HBM to optical modules, and further to coherent DSP, Pump Laser, InP production capacity, and fiber infrastructure. This chain is still extending outward and has not reached its end yet. There is another layer of longer-term logic Huang Renxun has been repeatedly discussing Physical AI recently. In the future, it will not only be cloud based agents, but also a large number of physical agents will emerge: robots, autonomous driving, industrial equipment, smart factories, each of which is an AI node. When computing power spreads from data centers to base stations, factories, cars, and robots, the demand for connectivity will take another step forward. Nokia's position here is very special. It was originally building base stations for telecom operators and has strong wireless integration capabilities and operator relationships. Nvidia wants to promote AI-RAN, turning base stations into miniature AI data centers, and Nokia happens to be standing at this intersection. This part is still very early, and there will be no short-term performance realization. It is enough to push the narrative forward. How do I view my position I bought LEAPS on January 21, 2028 at 13.8 and have a portion of stock purchased at 15.6, so I have no plans to move. I don't really care about the fluctuations of 10% and 20% in the middle. The real risk is that after the logic is established, it sells early before it erupts (for example, not knowing that Lao Huang suddenly makes another order call like MRVL) ， In the long run, I don't like this kind of direct increase in news coverage. I prefer a gradual increase, which is healthy and requires sufficient turnover of chips. Currently, the market is discussing AI, with most attention still focused on GPU and HBM, and the repricing of the connectivity layer has just begun. The first stage of AI focuses on computing power, the second stage on storage, and the third stage on connectivity. If this judgment is correct, the distance between Nokia's current valuation and its expected valuation is much larger than most people think. The logic hasn't changed, just keep holding on.