Recently, I’ve been paying attention to a very interesting phenomenon—market understanding of 5G may still be stuck in old perceptions.

In the past, everyone said 5G was a story of consumer upgrade, but after tracking the capital expenditure directions of telecom operators and network equipment vendors, it’s clear that three other engines are truly driving the global network demand: explosive traffic from AI data centers, low-latency requirements from edge computing, and enterprise private 5G networks moving from trials to large-scale deployment. These three trends stacking together are redefining the entire 5G industry chain.

There’s a straightforward data point—future global traffic demand could be three to seven times current levels, with over 30% driven by AI. This is not a speculative forecast; it’s based on actual orders and expenditures.

Why focus on this? Because AI inference has already surpassed training to become the biggest traffic driver. In 2023, inference accounted for only one-third of AI computing, but now it’s over half, and this year is expected to surpass two-thirds. Training is cyclical and concentrated, finishing once done, but inference is continuous, accumulating with each new user, and must be deployed in a dispersed manner to reduce latency. This means network infrastructure upgrades are not optional—they are a hard requirement.

The explosive growth of AI agents is also a key driver. Research firms estimate that this year, the number of AI agents worldwide will be about 50 to 100 billion, potentially skyrocketing to 2 to 5 trillion by 2036. Along with this growth, global bandwidth usage will jump from about 100 exabytes daily this year to over 8,000 exabytes in 2036. AI-driven network traffic has a compound annual growth rate of up to 51%, with overall traffic expected to grow 5 to 9 times.

This change in traffic structure directly determines the types and specifications of upstream network equipment procurement. Data centers are now linking multiple facilities within the same region, treating them as a single AI factory, with internal interconnections shifting from traditional fewer than 1,000 fibers to thousands. Long-term, the traffic in North American hyperscale data centers has maintained an annual growth rate of over 30%, all concentrated on optical transceivers, especially high-speed modules over 800G, with the 1.6T generation gradually entering mass production.

Private 5G moving from concept validation to mass production is also noteworthy. Analysts’ views are straightforward—private 5G is a core pillar for large-scale AI expansion in production environments, enabling automated systems to operate reliably with zero tolerance for errors. By the end of this year, the scale of private LTE and 5G network infrastructure could reach $6.4 billion, with about 40% dedicated to standalone 5G private networks.

The relationship between edge computing and 5G is a necessity. Many AI computations cannot be fully offloaded to centralized clouds due to high latency. Multimodal AI agents require at least 3 Mbps uplink bandwidth, with smooth experience needing 8 Mbps, and air interface latency below 160 ms; wearable devices like AI glasses need over 10 to 20 Mbps uplink and seamless global coverage; industrial intelligent systems demand over 20 Mbps with millisecond latency and reliability exceeding 99.99%. These data precisely explain why private 5G is an infrastructure that Wi-Fi cannot replace in production lines or remote operation scenarios.

Viewing 5G and AI infrastructure separately, there are four levels. The upstream tends to be more volatile and more directly reflect AI capital expenditures; the downstream is relatively stable but varies in sensitivity to economic cycles.

Upstream includes components and materials—optical communications, high-speed optical modules, silicon photonics, PCBs, heat dissipation, etc. The upgrade of AI data center interconnect specs from 800G to 1.6T causes the largest fluctuations, with order visibility directly impacting revenue. Midstream involves equipment and infrastructure—companies like Ericsson, Nokia, Cisco, Juniper—that handle private 5G networks, data center switches, and edge computing nodes. Their fluctuations are moderate and influenced by telecom and enterprise capex cycles. Downstream includes operations and services—companies like AT&T, Verizon, Chunghwa Telecom—that provide fixed fiber access, 5G fixed wireless access, and enterprise private network management. Their volatility is lower, focusing on dividends and cash flow, with slower growth. The extended layer involves applications and software—edge AI platforms, IoT, smart factory solutions—more uncertain and often composed of smaller stocks.

In trading, you might focus on different levels depending on the cycle. Short-term traders can watch for capacity news in upstream optical communications; swing traders can monitor midstream equipment companies’ earnings and capex guidance; long-term defensive investors might consider downstream telecom operators as satellite positions. It’s important to note that these four levels often have different stock price rhythms. Over the past six months, upstream optical communications surged first, followed by midstream equipment vendors, with downstream telecoms lagging. This isn’t about who is better, but rather about capital flowing into the most directly benefiting sub-sector at each stage.

If you’re unsure where to start, it’s recommended to observe midstream equipment vendors or upstream optical communications first, as their orders and AI capex are most directly related, making their stock movements easier to understand.

Taiwan’s supply chain also benefits clearly from the upgrade from 800G optical modules to 1.6T and the acceleration of private 5G networks. Fields like semiconductor wafer foundries, compound semiconductors, network switches, optical modules, and related Taiwanese firms are benefiting from data center bandwidth upgrades and telecom 5G-Advanced spending. In chip design, MediaTek’s 5G-Advanced M90 solution and data center ASIC revenues have surpassed $1 billion; RF and power amplifier companies like Richwave, Macroblock, and NewEdge are seeing steady orders for 5G base station PAs, with NXP’s exit leading to obvious order transfers; optical communication and fiber component firms like Lianya, Huaxing Optoelectronics, and Zhongda are experiencing increased demand for 800G data center optical modules and silicon photonics.

Regarding specific 5G stocks worth watching:

The first category is network equipment suppliers. Ericsson, founded in 1918, is already a key player in global 5G promotion, with about 40% of global communication traffic carried by its networks. It is now transforming from a traditional telecom equipment provider to a critical enabler of enterprise edge AI and private 5G. Recently, Ericsson signed a multi-year strategic partnership with NTT DATA, with NTT DATA acting as a global system integrator and managed service provider to promote Ericsson’s private 5G platform in manufacturing, mining, ports, airports, energy, transportation, and smart cities. In Asia, Ericsson also signed a three-year 5G-Advanced acceleration agreement with FarEasTone, aiming to push network evolution toward AI-native, 5G-Advanced, and 6G-Ready.

The second category is optical communications and fiber. Corning, tracking optical supply chain orders, finds that AI-driven demand for fiber and optical modules far exceeds market consensus. This year, transmission specs starting from 800G are accelerating toward 1.6T, and Corning is one of the main beneficiaries of this structural upgrade. The transition from traditional pluggable optical modules to low-power linear pluggable optics and silicon photonics integration is driving the next wave of optical communication growth. Although technically challenging, early movers’ potential orders and scale effects could be significant. Taiwanese firms in silicon photonics and LPO have competitive advantages, with a complete ecosystem from TSMC’s silicon photonics platform and laser chips to backend packaging.

The third category is telecom operators. AT&T, a major US telecom, provides 5G services in over 14,000 US towns. Traditional telecom stocks have long been valued for stable dividends, low growth, and defensive qualities. But after the surge in AI data traffic, fixed broadband, fiber access, and 5G fixed wireless access are regaining growth potential. Cloud giants’ demand for high-speed data center interconnects also boosts fiber backbone leasing in some regions. While profit curves are less steep than equipment vendors, as the market reassesses telecoms’ growth sustainability, their long-term discounted valuations may see slight upward revisions.

The fourth category is 5G ETFs. For investors who prefer not to single-stock bet, the Defiance 5G Next Gen Connectivity ETF tracks companies involved in 5G development and AI-driven network transformation, covering equipment, chips, and infrastructure providers. The First Trust Indxx NextG ETF focuses on infrastructure and application companies related to 5G and next-generation communication tech, with broader coverage. These ETFs’ holdings can effectively diversify risks associated with individual stocks or sub-sectors.

What are the biggest risks in investing in 5G concept stocks now? There are three key points to watch.

First, telecom profit realization remains sluggish. Although underlying network equipment and optical communication supply chain revenues and orders are booming, many telecom operators have yet to find effective business models to convert more traffic into significant profit growth. The high costs of infrastructure investments and the lag in end-user revenue conversion are still evident.

Second, the pace of 5G equipment deployment may fall short of expectations. While AI-driven demand for optical communications over 800G is clear, delays in data center and private network construction due to government approvals, land permits, power supply, tariffs, and supply chain destocking in some regions could slow progress. Tariffs have increased import costs for chips, RF modules, antennas, and routers, and North American and European suppliers dependent on Asian supply chains face longer procurement cycles and higher costs.

Third, the 6G narrative has already emerged. Some capital has shifted from 5G equipment to 6G concept stocks, believing that the next-generation standard will truly unlock new business models. While 5G-Advanced’s commercialization is a critical step toward 6G, this preemptive rebranding and capital shift may exert unnecessary selling pressure on still-growing 5G equipment stocks.

Fundamentally, the core network bottleneck has shifted from download bandwidth to uplink capacity, low latency, and reliability. AI inference demand has overtaken training, and AI agents being online 24/7 generate continuous, bursty, and highly elastic mixed traffic. 5G’s features of ultra-low latency, high reliability, and massive connectivity are naturally aligned with these AI needs.

If focusing on mid-term structural growth, equipment vendors and optical supply chains benefit directly from the structural growth in AI data center capex. For short-term technical trading, this year may see delays in data center deployment, early 6G funding, or telecom profit shortfalls, requiring flexibility. For systematic investors, starting with midstream equipment stocks and gradually expanding to upstream optical communications and downstream telecom operators can form a comprehensive industry chain allocation.