Market backdrop: AI infrastructure demand defies the bears
The AI stock complex has wrestled with a tug of war all year, swinging between optimism about rapid AI adoption and concern that capacity will outpace need. Investors have focused on the risk of oversupply in GPU clusters and AI infrastructure, a concern that has squeezed neocloud providers like Nebius and its peers. Yet in a fresh, bottom up assessment, Morgan Stanley argues that the big story remains a massive buildout of compute capacity, not a glut of hardware waiting to be idle.
The note arrives as the industry calibrates how quickly hyperscalers will scale up their AI workloads in the wake of fast evolving models and growing demand for real time inference. The bank says the jump in compute needs over the next few years could redefine the capital spend profile for the tech sector and set a new trajectory for exposed infrastructure players.
One line that has circulated in trading rooms highlights a controversial angle: morgan stanley destroys bear narrative by reframing the risk around capacity and not scarcity. The takeaway is simple in theory and hard in practice: a multi-year, multibillion to multitrillion dollar buildout may be underway, supported by both software advancement and the hardware supply chain that underpins it.
Morgan Stanley thesis: compute capacity rises threefold by 2028
Morgan Stanley lays out a bottom up model that tracks compute capacity across the five dominant hyperscalers. The core finding is stark: compute capacity is set to climb from about 30.5 gigawatts in 2025 to roughly 116.6 gigawatts by 2028. That is an increase of 282 percent in just three years, and it implies a wave of capital spending to design, finance and bring online new GPUs, networking gear and data center space.
Key numbers from the MS model include a near fourfold gain in total capacity and an implied capital spending surge that could run into the trillions of dollars over the period up to 2028. The bank pegs the potential investment range for the AI hardware cycle at roughly 4 trillion to 8 trillion in the coming years as capacity scales to meet demand.
Compute capacity by company: a snapshot from the MS framework
- Amazon: 2025 capacity around 13.8 GW; 2028 capacity near 35.8 GW; growth about 159%
- Google: 2025 capacity about 5.0 GW; 2028 capacity about 31.6 GW; growth about 532%
- Microsoft: 2025 capacity roughly 7.5 GW; 2028 capacity around 20.3 GW; growth about 171%
- Meta Platforms: 2025 capacity near 3.5 GW; 2028 capacity around 21.2 GW; growth about 506%
- SpaceX: 2025 capacity about 0.7 GW; 2028 capacity around 7.8 GW; growth about 1,014%
Aggregate the five hyperscalers show a climb from 30.5 GW in 2025 to 116.6 GW in 2028, underscoring the magnitude of the infrastructure cycle. The 282 percent uplift over three years is the centerpiece of Morgan Stanley s argument that the next phase of AI rollout will be capital intensive rather than supply constrained.
Implications for Nebius and its rivals
Nebius Group and other neocloud players rely on renting high performance GPU clusters to AI developers and startups. The new MS framework suggests that the demand side for Nebius and peers will grow in step with hyperscaler capacity. If hyperscalers add thousands of megawatts of compute to support advanced models, the need for outsourced compute capacity and specialized data center services rises in tandem.
Industry observers note that the bear case for AI infrastructure stocks hinged on the risk of oversupply and fading demand. The Morgan Stanley analysis flips that script by stating that the coming years will be defined by capacity expansion and sustained capital expenditure. In that light, Nebius and similar firms could see more predictable utilization and stronger pricing power as customers compete for space in a tightened, purpose built data center ecosystem.
The analysis also points to a structural shift in how AI ecosystems are financed. Rather than a one time cycle of hardware purchases, the forecast shows a steady drumbeat of investment to scale compute capacity, with a large portion directed at building out data centers, cooling, and energy efficiency to support higher utilization at scale.
Investor reaction, risk and the path forward
Investors are parsing the implications for Nebius and other listed players. While the note does not provide a single stock price target, the emphasis on sustained, multi year capex suggests a longer horizon for equity holders to see meaningful earnings uplift from higher utilization and new service lines tied to AI workloads.
Analysts caution that execution risk remains. The supply chain for high end GPUs, energy costs, and regulatory considerations around data center siting and carbon footprints could modulate margins. Nonetheless, the core message is consistent with a broader macro trend: the AI hardware cycle is not a blip but a foundational wave that could lock in a long term demand envelope for Nebius and its peers.
In interviews after the note circulated, a number of market participants argued that the bear case underestimates the scale of the investment required to deploy the next generation of AI models. The sentiment is that Nebius and other neocloud providers stand to benefit from a combination of rising utilization, longer term contracts and a shift toward edge and specialized data center services that complement hyperscaler capacity.
The thrust of the Morgan Stanley case is as much about capital discipline as it is about demand. The bank contends that the coming years will see orderly, financed growth rather than a sudden glut that drags down prices and utilization. That view aligns with a broader trend in 2026 markets where investors are scrutinizing the economics of cloud and AI infrastructure, seeking evidence that scale can be achieved with sustainable margins.
Conclusion: a new lens on the AI infrastructure cycle
As the AI revolution moves from talk to deployment, the narrative around Nebius and other neocloud players is shifting. The Morgan Stanley bottom up framework presents a scenario where the market is preparing for a protracted, capital intensive expansion rather than a sudden plateau. If the hypothesis holds, Nebius could emerge as a beneficiary of a broader, multiyear wave of compute capacity, with stronger utilization and pricing dynamics as hyperscalers push the envelope on efficiency and scale.
For investors, the take away is clear: the horizon for AI infrastructure assets is longer than many anticipated. The focus is shifting from fearing a sudden lull to watching how the capital plan unfolds, how the data center footprint grows, and how the economics of AI compute evolve in a world where the 2028 target feels not like a peak but a stepping stone to even larger deployments. In this framework, the line morgan stanley destroys bear rises from a critique to a catalyst for a new, data-driven investment thesis around Nebius and the broader AI infra space.
Discussion