Co-packaged optics (CPO), a technology that integrates optical engines directly into networking switches, is emerging as a critical solution to the data center power crisis driving AI infrastructure expansion across the United States. The U.S. CPO market was valued at USD 50 million in 2025 and is projected to reach USD 467 million by 2033, growing at a compound annual growth rate of 31.8% percent. This explosive growth reflects a fundamental shift in how hyperscale operators are approaching the power and bandwidth challenges posed by GPU-intensive artificial intelligence workloads.

The problem CPO solves is deceptively simple but consequential. Traditional electrical interconnects between data center switches consume significant power and generate heat as data travels across networks. CPO technology replaces these electrical connections with optical ones, dramatically reducing power consumption, signal loss, and latency compared with conventional pluggable transceivers. For data centers running thousands of graphics processing units (GPUs) simultaneously, these efficiency gains compound across entire facilities.

Why Are Data Centers Suddenly Investing in Optical Networking?

The answer lies in the sheer scale of AI infrastructure buildout happening right now. The U.S. hyperscale data center market is expected to attract investments of USD 697.84 billion by 2031, growing at a compound annual growth rate of 14.66 percent. This investment surge is driven almost entirely by demand for GPU-dense infrastructure to support large language models (LLMs), AI training clusters, and high-performance computing environments. Companies like AWS, Google, and Microsoft are announcing multi-billion-dollar data center expansions, and power availability has become the primary constraint limiting where and how fast they can build.

CPO technology is emerging as a key enabler for next-generation Ethernet networks operating at 800 gigabits per second, 1.6 terabits per second, and future 3.2 terabit per second speeds. At these bandwidth levels, traditional electrical interconnects face thermal and electrical limitations that optical solutions can overcome. The technology shipped approximately 13,500 CPO-enabled networking systems in 2025 and is forecast to reach 151,000 systems by 2033, driven by rapid investments in AI infrastructure and high-performance data center networks.

Which Companies Are Leading the CPO Race?

The United States remains the global leader in CPO commercialization, and major semiconductor and networking companies are investing heavily in the technology. Broadcom, NVIDIA, Cisco, Intel, Marvell, and Coherent are all developing optical networking, advanced packaging, and silicon photonics technologies to capture this growing market. Federal support through the CHIPS and Science Act is further strengthening domestic semiconductor manufacturing and advanced packaging capabilities, supporting long-term adoption of CPO across next-generation AI and cloud infrastructure.

The geographic concentration of this investment is also noteworthy. The Midwestern United States is expected to lead hyperscale data center investments, accounting for over USD 21.54 billion in 2025, followed closely by the Southeastern United States. Virginia, Texas, Illinois, and Arizona are attracting significant capacity additions as operators seek locations with access to scalable power infrastructure. This geographic shift reflects the reality that power availability, not real estate cost, is now the primary driver of data center location decisions.

How Are Hyperscale Operators Addressing the Power Challenge?

Beyond optical networking, data center operators are deploying a range of complementary technologies to manage power consumption and heat generation from GPU-intensive workloads:

• Direct Liquid Cooling: Companies like Meta are deploying direct liquid cooling systems designed specifically for high-density NVIDIA GPU environments, as demonstrated by Meta's Catalina architecture for next-generation AI infrastructure.
• Renewable Energy Procurement: Major cloud providers have secured more than 40 gigawatts of wind and solar capacity through Power Purchase Agreements (PPAs), with AWS expected to add over 1 gigawatt of self-built capacity between 2025 and 2030.
• Nuclear Power Integration: Nuclear power is emerging as an alternative energy source for hyperscale campuses, with AWS planning a USD 5 billion data center development in Texas comprising around 18 data center buildings.
• Battery and Power Architecture Innovation: Lithium-ion batteries and rack-level power architectures are gaining wider adoption across hyperscale facilities, illustrated by Google's deployment of more than 100 million lithium-ion cells within its global data center portfolio.

These investments underscore a critical reality: the power infrastructure supporting AI is becoming as important as the chips themselves. Data center power capacity additions are now shaping development decisions across the entire U.S. market, with operators prioritizing locations that offer both immediate power availability and long-term scalability.

The scale of individual projects illustrates the urgency. AWS is investing USD 10 billion in two data centers in Mississippi, a project expected to create around 1,000 construction jobs and be completed by 2027. Google announced plans for two new data center campuses in Oklahoma as part of its broader USD 9 billion investment in the state. Lambda's AI data center project in Missouri is planned to deploy more than 10,000 NVIDIA Blackwell Ultra GPUs, requiring power and cooling infrastructure at unprecedented density.

CPO technology represents a crucial piece of this infrastructure puzzle. By reducing the power footprint of data center networking, CPO enables operators to pack more computing density into existing power budgets, effectively multiplying the return on investment in power infrastructure. As AI workloads continue to grow and power constraints tighten, optical networking is transitioning from a niche technology to a foundational requirement for next-generation data center design.