On August 17, Morgan Stanley's chief fixed income strategist Vishwanath Tirupattur highlighted in a new report that the AI investment cycle is shifting from a technology-driven narrative to one led by capital markets.

Hyperscale cloud providers have significantly boosted their capital expenditure plans, while public and private credit markets are now shouldering a larger share of AI infrastructure financing. Investors are beginning to draw finer distinctions among different borrowers and business models, and the financing structures themselves are evolving rapidly, reaching deeper into the value chain and increasingly focusing on individual components such as chips.

Tirupattur wrote: "The sheer scale of AI investment has already made it clear that traditional funding channels alone cannot sustain the momentum. What is truly impressive is the speed, breadth, and creativity of the market's response."

AI is no longer just a technology story; it is becoming a capital markets story. Understanding where capital flows will soon matter as much as understanding the innovation itself.

Widening Funding Gaps Keep Credit Issuance Under Pressure

Compute supply continues to lag behind demand, prompting hyperscalers to keep raising capital spending to lock in future capacity.

Morgan Stanley's equity research team estimates that combined capital expenditures from Microsoft, Alphabet, Amazon, and Meta will grow 57% in 2027 compared to 2026.

Behind these spending plans lies strong confidence in returns. Morgan Stanley internet analyst Brian Nowak, in a report titled "The Path to 25%-50% Generative AI ROI," notes that related investments could achieve returns on invested capital (ROIC) exceeding 25%.

The challenge, however, is the timing gap between deploying capital and generating returns, which is putting pressure on near-term free cash flow. Morgan Stanley analysts have been revising down their free cash flow forecasts for the four major hyperscalers for 2027. With capital spending rising and cash flows falling, the funding gap is expected to widen further in 2027. Analysts believe AI-related credit issuance will remain elevated and may need to increase even more.

Credit Spreads Diverge: Who Feels the Squeeze and Who Holds Up

This summer's credit market performance has exposed structural differences within the AI financing ecosystem.

AI-related credit spreads widened considerably: high-grade issuers saw spreads expand by roughly 35 basis points from the start of the year, while lower-rated names widened by about 50 basis points, before partially retracing over the past two weeks.

More notable is the divergence across funding channels. High-grade unsecured bonds have been hit the hardest, due to a sharp increase in issuance volumes and broader risk exposure tied directly to the overall uncertainty of the AI investment cycle.

Data center ABS and CMBS, by contrast, have proven more resilient. These structured products are backed by operating assets that are built, powered, and leased, with largely predictable contractual cash flows. Combined with disciplined issuance pacing, they have effectively insulated themselves from the volatility in the unsecured market.

Tirupattur further distinguished between two types of issuers and their behavior:

On the high-grade end, hyperscalers like Microsoft, Alphabet, Amazon, and Meta carry average ratings around AA. Their funding needs are substantial, but their rating headroom is ample. Leading semiconductor companies such as Nvidia and Broadcom also fall into this category. Given their expected ROIC, these issuers are relatively insensitive to small shifts in funding costs, and wider spreads will not materially slow their financing pace.

On the lower-rated end, names like Oracle (rated mid-to-low BBB by various agencies), along with former bitcoin miners and REIT-based data center developers, have limited balance sheet flexibility and are far more sensitive to funding costs. For these borrowers, wider spreads create a real constraint, and the cost of capital itself becomes a natural stabilizer for supply.

Financing Frontier Moves Forward: From Data Centers to Chips

Tirupattur notes that the next phase of AI financing will feature a clear structural shift.

The focus of incremental capital spending is moving from data center shells to compute equipment (servers and chips) and energy assets. These asset classes are naturally suited for asset-level financing arrangements, providing greater room for private capital to participate.

Recent market signals are already clear: Nvidia has announced the launch of a compute infrastructure financing platform; a Broadcom-backed $35 billion chip financing deal has closed, setting a record; and the Beignet transaction from late 2025, along with the recent Sopaipilla deal, both represent new directions in financing structure innovation.

Tirupattur believes the rise of large-scale component financing will depend heavily on high-grade issuers leveraging their ratings and balance sheet strength—by providing backstop arrangements, credit support, and residual value guarantees—to help private capital underwrite ever-larger pools of AI infrastructure assets.

Capital Flows Will Matter as Much as Technological Innovation

As AI evolves from a technology cycle into a capital cycle, understanding the details of financing is becoming increasingly important.

The AI financing ecosystem is expanding quickly, but not uniformly. Different credit channels are absorbing varying levels of risk, and issuer behavior will diverge further based on capital needs, rating headroom, and sensitivity to funding costs. The end result is a more complex and layered market, where financing outcomes will increasingly determine competitive outcomes.

As Tirupattur puts it: "In the next phase of the AI buildout, understanding the flow of capital will be almost as important as understanding the flow of innovation. AI is no longer just a technology story—it is increasingly a capital markets story."