AI Investment Enters Stage 4: Citi Upgrades Cloud Capex, Reveals New Winners in Optical Communications

The investment wave in AI infrastructure continues to intensify. Citi’s latest research report has substantially upgraded its forecast for data center capital expenditures (capex) by global hyperscale cloud service providers (Hyperscalers), indicating that AI network infrastructure has reached a critical turning point—transitioning from the bandwidth expansion phase into the early adoption stage of 'Stage 4.' The next phase of growth will shift toward next-generation optical interconnect technologies such as Near-Packaged Optics (NPO) and Co-Packaged Optics (CPO).

Citi analyst Ronald Josey’s team highlighted in its latest industry report that AI infrastructure investment is forming a 'super cycle.' According to its updated model, data center capex from the top five cloud providers is projected to rise sharply from $400.2 billion in 2025 to $730.9 billion in 2026, reaching $978.1 billion in 2027 and surpassing $1.16 trillion in 2028. The 2026 forecast represents a 16% upward revision from the April estimate, reflecting that AI compute demand continues to exceed initial market expectations.

Citi believes AI investment is transitioning from the hundreds of billions of dollars to the trillion-dollar scale. The report forecasts that in 2026, combined data center capex from the four major hyperscalers—Amazon (AMZN-US), Microsoft (MSFT-US), Alphabet (GOOGL-US), and Meta (META-US)—will exceed $630 billion. Global AI capex is projected to reach $490 billion in 2026, with cumulative investment from 2026 to 2030 revised upward from $8 trillion to $8.9 trillion.

Citi notes that three to four of the four hyperscalers are expected to accelerate their data center capex growth in 2026 compared to 2025, indicating that the AI infrastructure race remains in an acceleration phase.

Although such massive investments will compress free cash flow in the short term, Citi views this not as a sign of deteriorating business health, but as a proactive strategic move to build long-term competitive moats. The report compares this to Amazon’s (AMZN-US) earlier heavy investment in AWS, suggesting that companies leading in AI infrastructure investment may gain stronger market advantages in the future. Citi expects Alphabet (GOOGL-US), Meta (META-US), and Amazon (AMZN-US) to deliver second-quarter revenue and earnings that exceed market consensus.

Another key insight from Citi’s report is that AI networking technology has now entered 'Stage 4.'

Citi divides AI networking development into five stages. Market focus has now shifted beyond basic network setup, bandwidth enhancement, and data center interconnects, moving into the early commercialization phase of scale-up optics—specifically, the integration of NPO and CPO into data center architectures.

In the past, the primary goal of AI data center construction was 'connecting devices.' However, as model sizes rapidly increase, data exchange volume and latency requirements are rising in tandem. Market demand is now shifting toward 'faster and denser connectivity.' NPO and CPO technologies, by improving optical interconnect efficiency between chips and racks, are expected to reduce power consumption and latency, becoming a key direction for the next phase of AI infrastructure upgrades.

Market data also reflects this trend. Citi notes that the front-end data center switch market is maintaining an annual growth rate of about 20%, while the AI backend Ethernet switch market is growing at over 100% year-on-year, with a market size exceeding $3.7 billion and expected to double for the full year. The optical transmission market grew approximately 20% year-on-year in Q1, with ZR/ZR+ technology growing over 30%.

As AI networking enters Stage 4, Citi believes optical communication and network equipment suppliers will be the primary beneficiaries.

The report names three companies as top picks in the communications equipment sector: Lumentum (LITE-US), Ciena (CIEN-US), and Keysight Technologies (KEYS-US). Among them, Lumentum is viewed by Citi as a key beneficiary of the Stage 4 trend, with a first-mover advantage in NPO and CPO optical interconnect technologies and benefiting from tight supply of silicon wafers and laser components.

Ciena benefits from growth in the optical transmission market, with its Q1 market share exceeding 25%, up over 300 basis points year-on-year, and ZR/ZR+ technology growing over 30%. Keysight is another favored beneficiary due to rising demand for AI network infrastructure testing.

Additionally, the spillover effects of AI capex will extend to semiconductors and power infrastructure. Citi has previously raised its 2026 AI capex forecast from $420 billion to $490 billion, believing that GPU suppliers like NVIDIA and networking chipmakers like Broadcom will continue to benefit.

However, Citi also warns that the AI investment boom is not without risks. One key challenge is whether the supply chain can support rapidly expanding demand.

The report points out that supply bottlenecks may emerge in critical components such as silicon switch chips, memory components, and lasers, potentially affecting gross margins and product deliveries in the second half. While capex upgrades are one thing, the actual deployment capability of the supply chain will determine whether the AI infrastructure cycle can continue smoothly.

Moreover, the market is beginning to question whether AI investments are overvalued. While Citi believes this AI investment cycle is fundamentally different from the 2000 dot-com bubble—due to real enterprise AI demand generating actual revenue and orders—markets must still monitor whether massive capex from 2027 to 2028 translates into sufficient returns.

On the other hand, traditional network equipment vendors that fail to keep up with the optical interconnect upgrade trend may face market share erosion. Citi notes that Stage 4 represents a technological shift, and the ability to deploy NPO and CPO and secure access to silicon wafers and lasers will be key competitive differentiators.

The significance of Citi’s report lies in redefining both the 'scale' and 'technical direction' of AI infrastructure investment.

In terms of investment scale, AI infrastructure is moving from the hundreds of billions to the trillions of dollars. From a technological evolution perspective, market focus is shifting from simply increasing compute power to enhancing data transmission efficiency.

Citi believes that the optical interconnect upgrade represented by Stage 4 will become the core of the next phase of AI network investment. For investors, the next key watchpoint will be whether cloud giants’ capex guidance in earnings reports meets expectations. If actual investment continues to be upgraded, it will further strengthen the AI super-cycle narrative; conversely, if capex falls short of expectations, the market may reassess the AI investment cycle.

The AI race is gradually extending from competition in models and chips to competition in internal data center network architectures. Citi’s proposed 'Stage 4' signifies that the next battlefield for AI infrastructure investment is shifting from compute power itself to the high-speed optical interconnect networks that support it.