Global chip sales recorded a 6.4% month-on-month growth: The AI boom is spreading to these five sectors.

A notable new shift worth paying attention to is emerging amid the current boom of AI chips.

Latest data from the Semiconductor Industry Association (SIA) shows that global semiconductor sales reached 146.8 billion U.S. dollars in July 2026, rising 6.4% from June. More notably, this marks the 17th consecutive month-on-month growth in global semiconductor sales.

From a longer time horizon, the change is even more prominent: global semiconductor sales hit 403.3 billion U.S. dollars in the second quarter of this year, up 35.1% from the first quarter.

The chip market is evidently red-hot.

But the more critical question now is no longer "whether the AI chip boom will continue", but:

After GPUs have become the most familiar AI hardware in the market, where will this boom propagate next?

The answer is becoming increasingly clear.

From memory and advanced packaging, to semiconductor equipment and materials, and further to power supply, heat dissipation and high-speed interconnection, AI is gradually expanding the previous market trend centered on "a single chip" into industrial chain demand covering "an entire AI server" and "a whole data center".

This is likely the most important clue to understand the AI hardware cycle in the coming period.

Behind the 146.8 Billion U.S. Dollars Figure: Look Beyond GPUs

Let's start with an easily overlooked data detail.

The global semiconductor sales figure of 146.8 billion U.S. dollars announced by SIA for July is higher than the 137.9 billion U.S. dollars recorded in June, representing a 6.4% month-on-month growth.

Besides, SIA adopts a three-month moving average statistical caliber, which means the figure does not reflect sudden orders on a specific day, but relatively sustained industrial changes over a period of time.

Why can AI drive such strong growth?

Because training and running large models require far more than just a single GPU.

Behind a high-performance AI server, at least several key issues need to be addressed:

First, computing power support is required.

Second, a huge amount of data needs to be quickly transmitted to the computing chip.

Third, sufficient storage capacity is required.

Fourth, different chips need to be packaged together.

Fifth, stable power supply for the entire server and efficient heat dissipation of massive heat are required.

Therefore, after AI infrastructure officially enters the large-scale construction phase, bottlenecks in the industrial chain will keep shifting continuously.

At the earliest stage, the main concern of the industry was:

Is there enough GPU supply?

Later, the problem gradually evolved into:

Is there enough HBM supply?

Is the production capacity of advanced packaging sufficient?

Is the high-speed network performance up to standard?

Is there sufficient power supply?

Can the heat dissipation system keep up with demand?

This is the very interesting feature of the AI industry.

When one bottleneck is solved, the next bottleneck is likely to become the new center of demand.

As a result, capital, orders and capital expenditure continue to spread outwards along the industrial chain.

The First Obvious Beneficial Segment: Memory

One of the most prominent areas where the AI boom spreads is the memory sector.

In the era of traditional personal computers, a PC requires not only a CPU, but also memory.

The same applies to AI servers, except that the data throughput is multiplied by many times.

Especially for large model training, chips need to read massive amounts of data continuously. If the computing power is strong but data cannot be transmitted in time, even the most expensive GPU can only be in a waiting state.

Thus, High Bandwidth Memory (HBM) has become one of the most critical components of AI servers.

This also explains why in this AI cycle, the memory industry has seen a change different from the traditional consumer electronics cycle:

In the past, the largest source of demand for memory was usually mobile phones and PCs.

Now, AI servers are becoming a key variable affecting the supply and demand of high-end memory.

This change has even begun to affect ordinary consumer electronics products.

Reuters reported on September 16 that some small and medium-sized mobile phone and laptop manufacturers are preparing for the memory supply shortage that may last for years. The industry is not only worried about price increases, but also about whether they can obtain sufficient production capacity in the coming period.

This has formed a very notable transmission chain:

AI servers grab high-end memory production capacity → memory manufacturers adjust capital expenditure and product structure → the supply of some general-purpose memory is affected → traditional terminals such as mobile phones and PCs recalculate their cost structure.

In other words, the prosperity of AI servers may even indirectly change the cost structure of an ordinary mobile phone.

The industrial impact of AI has begun to spill over from data centers to the consumer electronics sector.

The Second Segment: Advanced Packaging

If GPUs solve the computing problem, advanced packaging solves the problem of how to "integrate" increasingly powerful computing capabilities.

In the past, a very important growth logic in the chip industry was to continuously shrink the size of transistors.

However, as the manufacturing process becomes more and more advanced, it is increasingly difficult and costly to improve performance solely by relying on manufacturing processes.

As a result, the importance of another technical route has risen rapidly:

Chips with different functions are connected together through advanced packaging technologies.

AI chips especially require this capability.

Extremely high data transmission efficiency is required between GPUs, HBM and other chiplets.

This is also why advanced packaging has gradually evolved from a back-end link in semiconductor manufacturing to an important segment that determines the production capacity of high-end AI chips.

The changes it brings are very direct.

In the past, when the market evaluated chip companies, people often asked:

"Can it design advanced chips?"

Now an additional question must be raised:

"After the design is completed, can the chip be successfully packaged?"

For the industrial chain, this means that demand not only spreads to wafer manufacturing, but also continues to transmit to segments such as packaging equipment, testing equipment, substrates, materials and inspection.

The more complex the AI chip is, the longer the corresponding industrial chain will be.

The Third Segment: Semiconductor Equipment and Materials

There is another often overlooked logic:

When chip sellers see booming business, it will eventually drive the "shovel sellers" to receive more orders.

After global chip sales continue to grow, will chip manufacturers expand their production capacity?

If the answer is yes, they will increase capital expenditure in the next step.

What does a wafer fab need to expand production capacity?

Lithography machines, etching equipment, thin film deposition equipment, cleaning equipment, inspection equipment, and a large number of high-purity materials.

Therefore, there is a classic transmission chain in the semiconductor industry:

Terminal demand grows → chip demand rises → wafer fabs increase capacity utilization → capital expenditure is raised → demand for equipment and materials increases.

Recently, this transmission trend has shown obvious signals.

Reuters reported on September 14 that as major global chip manufacturers push forward next-generation manufacturing processes, ASML's new High NA EUV equipment is being adopted by more and more clients. The unit price of such equipment is as high as about 400 million U.S. dollars, and manufacturers including TSMC, Samsung, SK Hynix and Intel are all making relevant layouts.

This shows that AI investment has begun to further change the structure of semiconductor capital expenditure.

Apart from equipment, there is another huge market that is easily overlooked:

Materials.

The manufacturing process of a single chip may involve hundreds or even thousands of working procedures.

Silicon wafers, photoresist, electronic special gases, wet electronic chemicals, high-purity quartz and other materials do not seem to be as eye-catching as GPUs, but they directly determine the manufacturing yield.

The more advanced the AI chip is, the higher the requirements for material purity and equipment accuracy will usually be.

Therefore, the complete AI semiconductor industrial chain never consists of only "chip companies".

Behind it stands a long list of equipment and material suppliers.

The Fourth Segment: Power Supply and Heat Dissipation

Taking one step further outwards, we will enter another huge market:

Electricity and heat.

AI servers have a very practical problem — they consume huge amounts of power and generate massive heat.

The stronger the computing power of a chip is, the higher its power consumption usually is.

As a result, some previously relatively inconspicuous components in servers are becoming increasingly important.

For example, the power supply system.

Servers need to deliver power to GPUs, CPUs and storage systems in a stable, safe and highly efficient manner.

Another example is the heat dissipation system.

In the past, an ordinary server could basically meet heat dissipation needs with air cooling technology.

However, as the power density of a single cabinet of AI servers continues to rise, traditional air cooling is gradually approaching its efficiency bottleneck, and liquid cooling technology is being adopted by more and more data centers.

This means that a very interesting value migration is taking place in the AI industrial chain:

In the past, the most concerned question for investors was:

"How strong is the computing power of the chip?"

Now two more questions need to be added:

"How much power do these chips need?"

And:

"How to dissipate the generated heat?"

This is why power modules, liquid cooling systems, cold plates, CDU, pumps and valves, as well as data center infrastructure, have begun to become important parts of the AI investment chain.

These facilities themselves do not produce AI capabilities.

But without them, GPUs cannot work continuously.

The Fifth Segment: High-Speed Connection

There is one last easily overlooked segment — interconnection.

No matter how strong the performance of a single server is, it cannot complete ultra-large-scale AI training tasks.

Modern AI data centers often require tens of thousands of GPUs to work collaboratively.

At this time, the data transmission speed between chips, between servers and between cabinets will directly determine the efficiency of the entire system.

If the computing speed of the GPU is very fast but the network transmission cannot keep up, a very costly waste will occur:

The chips are waiting for data to be transmitted.

Therefore, AI data centers are driving continuous upgrading of high-speed networks, optical modules, switches and related optical communication devices.

In short, in the past, most of the value of data centers may be concentrated on "servers".

In the future, more and more value will be distributed to:

Chips + memory + networks + power supply + heat dissipation.

This is why the AI hardware market trend can no longer be simply summarized by the phrase "GPU industrial chain".

It is evolving into a complete AI infrastructure industry.

What Are the Real Opportunities for China's Supply Chain?

This is also the part of the global semiconductor data that is most worth paying attention to for Chinese enterprises.

Chinese enterprises do not necessarily need to directly participate in the competition for the most advanced GPUs in every segment.

The extension of the industrial chain itself means more participation opportunities.

For example, the memory industry requires equipment, packaging and testing, and materials.

Advanced packaging requires equipment, substrates, inspection and materials.

AI servers require PCBs, power supply, heat dissipation and high-speed interconnection.

Data centers also require power supply and distribution systems, liquid cooling facilities, cabinets and even energy infrastructure.

The more expensive an AI chip is, the more expensive the entire system built around it usually is.

What is really worth observing is not "whether there is another GPU company in China", but whether Chinese enterprises can obtain stable orders and increase their value share in the increasingly long AI infrastructure supply chain.

However, a very common misconception must be warned against here:

The outward spread of the AI boom does not mean that all companies in the industrial chain will make profits simultaneously.

The growth of global chip sales cannot directly lead to the conclusion that any Chinese enterprise will see profit growth.

To turn AI demand into actual profits, an enterprise in the industry needs to pass at least several key checks:

Has it been included in the supply chain of core customers?

Has the value of its products been improved?

Has its capacity utilization rate risen?

Has the product price been suppressed?

How much profit can be retained from the newly added revenue?

This is also the most critical distinction that needs to be made when analyzing the AI industrial chain:

The prosperity of the industry does not equal the performance growth of individual companies.

To Analyze AI Hardware in the Future, You Can Look for Opportunities Along These Five Layers

Therefore, the 6.4% month-on-month growth of global semiconductor sales in a single month is not only significant for the fact that "the chip market continues to grow".

What is more noteworthy is that the growth is spreading to more segments.

To observe this AI hardware cycle in the future, you can verify it layer by layer along the five-tier industrial chain:

The first layer: Computing chips.

Check whether the core computing demand of GPUs, ASICs and other products remains sustained.

The second layer: Memory.

Track the price, production capacity and capital expenditure of HBM, DRAM and other memory products.

The third layer: Advanced packaging.

Pay attention to the capacity utilization rate of packaging and the order status of related equipment and materials.

The fourth layer: Equipment and materials.

Observe whether wafer fabs continue to expand capital expenditure.

The fifth layer: Power supply, heat dissipation and high-speed interconnection.

Verify whether the power density of AI servers and the construction scale of data centers continue to increase.

If new orders and capacity bottlenecks appear in these five layers of demand in turn, this AI investment cycle will no longer be a boom limited to a small number of GPU companies.

It is evolving into a larger-scale infrastructure upgrading campaign.

Over the past two years, the market has been constantly asking:

Who can develop more powerful AI chips?

In the coming period, a question that may become increasingly important is:

To make these chips run smoothly, how much more equipment, memory, power supply, heat dissipation facilities and interconnection resources does the entire industry need to add?

The global semiconductor sales figure of 146.8 billion U.S. dollars may only be a node in this process.

What is really worth paying attention to is that more and more industries are being revalued by AI after the GPU sector.