Morgan Stanley has sharply lifted its forecast for the artificial intelligence network infrastructure market, now estimating a global opportunity of about $70 billion by 2030, more than four times its earlier projection, as demand for moving data between AI chips becomes one of the biggest bottlenecks in modern computing.
The bank’s latest view points to a clear near-term winner: copper. While optical networking is expected to become increasingly important later in the decade, Morgan Stanley says copper cables are likely to keep dominating AI network buildouts through 2026 and 2027 because they remain cheaper, easier to deploy and more power-efficient for short-distance links inside AI clusters.
Co-packaged optics, known as CPO, is not expected to make a meaningful breakthrough until closer to the end of the decade. Morgan Stanley now expects CPO penetration to remain near zero through 2027, begin limited adoption in 2028 and reach roughly 20% to 30% of the market by 2030.
That timing matters because the AI infrastructure boom is not only about faster chips. It is increasingly about how those chips connect to one another, how quickly data can move across racks of servers and how much power is consumed in the process. As AI models grow larger and more complex, the physical network behind them is becoming just as important as the processors doing the calculations.
Morgan Stanley’s updated forecast suggests that the biggest spending wave over the next several years will not immediately go to futuristic optical designs. Instead, much of the near-term demand is expected to flow into copper cables, connectivity chips, electrical modules, testing equipment and other components that help existing server networks move more data across short distances.
The bank’s conclusion is straightforward: AI back-end networks are getting much larger, but the optical takeover is not imminent.
Copper remains the near-term winner
At the center of Morgan Stanley’s forecast is a practical point about distance. AI clusters are expanding from single-rack systems into multi-rack systems, but many of the most important connections still run across short distances. For links of up to about nine meters, copper continues to offer an advantage in cost and power efficiency.
That makes copper especially attractive for the current phase of AI data center construction. Cloud companies and large technology groups are racing to add capacity, but they are also watching costs, power use and supply-chain risk. Copper is a mature technology with established suppliers, known installation practices and lower complexity than newer optical systems.
Optical systems can move data over longer distances and are expected to become more important as bandwidth requirements rise. But for short-range connections inside racks or across nearby racks, copper remains hard to displace.
Morgan Stanley’s report suggests that copper will keep benefiting from engineering improvements that extend its useful life. Better cables, retimers, digital signal processors and active electrical components can help push copper further, even as network speeds increase.
This is important because AI networking demand is rising quickly. Data links are moving from 100 gigabits per second toward 200 gigabits and 400 gigabits per second, with even faster speeds already being tested. As those speeds climb, copper faces more pressure from signal loss, heat and power use. Even so, the bank expects copper to remain the preferred choice for many deployments until optical supply chains and platform designs are ready for broader use.
Optical adoption is expected to build slowly
CPO promises a major shift in how high-speed networks are built. Instead of placing optical transceivers separately at the edge of a switch or system, CPO moves optical components closer to the compute or switching chips. The goal is to reduce power use, improve bandwidth density and make data movement more efficient.
In theory, that approach is well suited for large AI systems. In practice, it is difficult to manufacture, test, replace and maintain at scale.
Morgan Stanley expects CPO adoption to stay limited through 2027 because the technology still faces several hurdles. These include thermal management, manufacturing yield, maintenance complexity and the need for tighter coordination between chipmakers, system builders and optical component suppliers.
Large cloud providers also have commercial concerns. Moving to tightly integrated optical designs could increase dependence on a smaller number of suppliers. It may also change maintenance costs and make it harder to swap parts in the field. For companies spending tens of billions of dollars on data centers, those risks are not minor.
The bank’s model shows CPO beginning to appear in limited form in 2028 before rising more meaningfully in 2029 and 2030. That shift is expected to coincide with bigger GPU domains, larger multi-rack architectures and rising demand for interconnects that can support far higher bandwidth.
Nvidia’s future systems are expected to play a major role in that transition. Morgan Stanley notes that future architectures could require far more optical engines for each GPU than today’s systems. According to the bank’s projections, each GPU could eventually require between 17 and 70 optical engines in large-scale deployments, compared with roughly two today.
That would mark a major change in component demand. But the report also makes clear that such adoption depends on platform decisions, supply readiness and the ability of cloud operators to justify the cost and complexity.
AI clusters are outgrowing old network assumptions
The AI infrastructure buildout is changing the structure of data centers. Traditional data center networks were designed around general computing, storage and internet traffic. AI training systems require something different: huge volumes of data moving rapidly between many processors, often with very low tolerance for delay.
Training large AI models requires thousands, and sometimes tens of thousands, of GPUs or custom accelerators to work together as one system. If data movement slows, expensive chips sit idle. That makes networking a direct driver of system efficiency.
Inference, the process of running trained AI models for real-world use, is also becoming more network-intensive. As companies deploy more AI services into search, software, advertising, coding tools, enterprise systems and consumer products, the need for fast, reliable back-end infrastructure continues to rise.
This is why Morgan Stanley’s forecast is focused on AI back-end networks rather than only front-end internet connectivity. The critical challenge is increasingly inside the data center itself, where servers, accelerators and racks must exchange data at high speed.
The shift from single-rack to multi-rack AI systems increases the need for advanced interconnects. A single rack may be able to rely heavily on short copper connections. Multi-rack systems require more complex network designs and eventually create stronger demand for optical links, especially as distances and bandwidth requirements increase.
The report suggests that this transition will happen gradually rather than suddenly. Copper will carry much of the early load, while optical technologies prepare for heavier deployment later in the decade.
Component makers face different timelines
Morgan Stanley’s forecast separates near-term beneficiaries from companies more dependent on later optical adoption.
Companies producing connectivity chips, electrical modules and components that support copper-based systems are expected to benefit first. These include Astera Labs, Broadcom and Semtech, which are exposed to demand for signal integrity, retimers, connectivity silicon and other tools that help copper operate at higher speeds.
The logic is simple. If copper remains the main technology through 2026 and 2027, spending should continue to favor companies that improve copper performance and extend its useful range.
Optical component suppliers such as Corning, Lumentum and Coherent are more tied to the pace of CPO adoption in the second half of the decade. These companies could benefit as cloud platforms and AI hardware makers shift toward optical-heavy designs, but the timing is less immediate.
That does not mean optical suppliers are out of the AI story. It means their largest opportunity may be delayed until CPO becomes easier to deploy at scale and more deeply embedded in major AI platforms.
Morgan Stanley also raised its rating on Keysight Technologies, citing rising demand for testing tied to new AI network topologies. Emerging interconnect standards and architectures, including NVLink, UALink and other high-speed links, require validation at 800 gigabits per second, 1.6 terabits per second and 3.2 terabits per second.
Keysight already generates a mid-teen percentage of its revenue from AI-related network testing, according to the report. As AI systems become faster and more complex, testing becomes more important because errors in high-speed networks can reduce performance, raise costs and delay deployment.
Fragmented platforms may slow a uniform shift
One reason Morgan Stanley is cautious about rapid CPO adoption is that AI networks are not developing around a single universal design. The market remains fragmented across different chip platforms, cloud architectures and proprietary interconnect systems.
Nvidia’s roadmap points toward greater use of optical connectivity over time, especially as GPU clusters get larger. But other ecosystems use different approaches. Google’s TPU systems, for example, rely on different network topologies and design choices. Other cloud providers and chip developers may also pursue distinct architectures.
This fragmentation can slow standard adoption. Suppliers may need to support several different designs at once, while cloud companies may hesitate to commit fully to one technology path before the market settles.
A broad move to CPO would also require changes across global supply chains. Optical engines, advanced packaging, thermal systems, testing equipment and maintenance procedures all need to mature together. If one part of that chain lags, large-scale deployment can be delayed.
That is why Morgan Stanley’s $70 billion forecast should not be read as a simple bet on optical components. It is a broader call on the growth of AI networking infrastructure, with copper dominating the first stage and optical systems gaining ground later.
Data center spending keeps rising
The updated forecast comes as the wider data center market continues to expand at an extraordinary pace. As of July 2026, outside industry estimates suggest the total network market could grow to about $154 billion by the end of 2028. Major technology companies spent more than $380 billion on data center construction and related infrastructure last year, reflecting the scale of demand created by AI workloads.
Power is also becoming a central constraint. Global data center power demand is expected by outside forecasters to rise sharply by 2030, with some estimates pointing to an increase of around 165%. That makes network efficiency a financial and operational issue, not only a technical one.
Copper’s advantage over short distances is partly about power. If a system can move data efficiently over a few meters without optical conversion, it can help control energy use and reduce cooling requirements. But as systems grow and distances increase, optical links may become necessary to keep power consumption from rising too quickly.
This creates a two-stage market. In the first stage, cloud builders focus on what can be deployed quickly and affordably. In the second stage, they move toward more advanced optical systems when copper begins to hit harder limits.
For traders watching AI infrastructure, the timing of that transition is critical. Short-term enthusiasm for optical networking may need to be balanced against the slower pace of real-world deployment. At the same time, ignoring optical technology entirely could miss the larger shift expected closer to 2028, 2029 and 2030.
Digital-asset markets may track the hardware cycle
The forecast also has implications for digital-asset traders focused on tokens linked to computing power, decentralized physical infrastructure, data networks and AI-related blockchain projects.
The main message from the hardware market is that usable infrastructure matters more than distant promises. Projects tied to current computing demand, practical server coordination and short-range network expansion may have a clearer near-term story than projects built mainly around future optical architectures.
Digital-asset traders may therefore pay closer attention to platforms that can operate within today’s hardware environment. That includes decentralized systems designed to coordinate compute tasks, manage data flows, support distributed workloads or connect physical infrastructure using existing network standards.
By contrast, tokens whose narratives depend heavily on complex light-based hardware may face a longer wait if CPO adoption remains limited until 2028 or later. Those projects may still have long-term potential, but the near-term hardware cycle suggests that widespread optical deployment is not yet the dominant market reality.
The same applies to decentralized physical network projects. Platforms that can scale using affordable, standard short-range links may be better aligned with the current data center buildout than models that require advanced optical supply chains to mature first.
For traders, the key indicators may include component shipments, cloud capital spending, power availability, data center construction trends and test-equipment demand. Monthly and quarterly reports from cable suppliers, optical component makers, chip connectivity firms and cloud infrastructure providers may offer early signs of when the market is shifting from copper-led expansion toward optical-led deployment.
The bottleneck is no longer just the chip
Morgan Stanley’s forecast highlights a broader truth about the AI boom: the hardest problem is no longer simply making faster processors. The challenge is connecting them efficiently, powering them affordably and keeping massive clusters running without wasting expensive computing capacity.
Copper is expected to capture much of the spending now because it solves today’s problem at today’s scale. Optical systems are expected to gain importance later because tomorrow’s AI clusters may be too large and too bandwidth-hungry for copper-heavy designs alone.
That creates a market with two clear phases. Through 2026 and 2027, the advantage remains with copper, electrical connectivity and testing tools. From 2028 onward, optical systems and CPO could begin taking a larger role, especially if Nvidia-led platforms and other large AI architectures require denser, faster and more power-efficient links.
The $70 billion opportunity described by Morgan Stanley is therefore not a forecast of an immediate technology replacement. It is a sign that AI networking itself is becoming one of the most important markets in the global technology supply chain.
For now, metal wiring remains at the center of the buildout. Light-based networking is coming, but it still needs time, scale and a more mature supply chain before it becomes the standard across the largest AI data centers.
Want more on AI’s market impact? Explore our guide to Web3, AI, and crypto reshaping digital infrastructure next.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
