Analyst Upgrades Signal Confidence in Neocloud Model
A wave of bullish analyst calls this week has put the spotlight on three AI infrastructure providers — CoreWeave, DigitalOcean and Nebius — as Wall Street bets on a new phase of artificial intelligence spending. Truist upgraded CoreWeave from Hold to Buy, while Baird launched coverage on both DigitalOcean and Nebius with Outperform ratings, arguing that the market is moving from training massive models to running inference workloads for real-world applications.
The analysts see a structural shift underway: enterprise customers are increasingly adopting open-source AI models and demand flexible, full-stack alternatives to the giant hyperscale clouds. CoreWeave, the leading neocloud operator, boasts roughly 1 gigawatt of online compute capacity and a large contracted backlog. Truist says that strong demand from beyond the large hyperscalers — driven by open-source and sovereign AI projects — will absorb any capacity freed up if big clients do not renew, reducing a key risk.
Baird's thesis for DigitalOcean centres on the company's evolution from a developer-centric platform into a holistic AI infrastructure provider. The firm highlights DigitalOcean's focus on digital-native businesses, its avoidance of rigid “take-or-pay” contracts, and its growing capabilities in inference and AI agents. For Nebius, the appeal is a full-stack cloud platform, self-designed hardware, and an active acquisition strategy aimed squarely at the inference market.
Behind the Ratings: The Strategic Logic of the Inference Shift
The Inference Pivot: From Training to Real-World Use
The common thread across the three ratings is a conviction that AI spend will increasingly tilt toward inference — the process of running a trained model to answer queries or generate outputs — rather than the enormously expensive training runs that dominated the first wave. Inference workloads tend to be more distributed, more sensitive to latency and cost, and better suited to specialised infrastructure. This opens a door for providers that can offer high-performance compute without the bundled services and premium pricing of hyperscalers.
CoreWeave: Scale and a Potential Boost from GPU Lifespan
Truist sees CoreWeave as the clear leader among neoclouds, with that 1 GW of online power giving it a scale advantage. The analysts also flagged a possible upside to profitability if the useful economic life of GPUs extends beyond the six-year depreciation schedule currently assumed. Longer-lived chips would lower replacement costs and boost margins, though this is contingent on technology cycles and how quickly model architectures evolve.
DigitalOcean: Betting on the SMB AI Opportunity
Baird’s initiation of DigitalOcean at $165 signals belief that the company can carve out a profitable niche serving small and medium businesses that the hyperscalers often overlook or under-serve. With an existing community of developers and a broadening product suite that spans infrastructure, data management and AI agents, DigitalOcean is positioned to capture customers who want an integrated solution but not the complexity — or long-term commitment — of a major cloud provider. Accelerating AI-related recurring revenue growth is a clear expectation.
Nebius: Vertical Integration and M&A as Accelerants
Nebius receives perhaps the most bullish structural call: Baird argues it is among the biggest beneficiaries of the training-to-inference shift. The firm’s ownership of most of its infrastructure, together with in-house hardware design and proprietary software like the Token Factory inference platform, creates a tightly integrated stack. An aggressive acquisition strategy adds another lever, allowing Nebius to bolt on new capabilities and scale more quickly. Baird expects Nebius to be one of the fastest-growing names in the sector as capacity expands and the customer base diversifies.
Risks: Hyperscaler Competition and Policy Headwinds
Not all of the picture is uniformly bright. Truist acknowledged two risks: the potential for Meta’s planned cloud business to add new competition, and the possibility of policy limits on data-centre development that could slow expansion. The counterargument is that the inference market is growing fast enough and is fragmented enough to accommodate multiple players, but the risk of margin erosion if hyperscalers aggressively price for inference cannot be dismissed.
What This Means for AI Infrastructure Investors
- CoreWeave: Focus on utilisation rates and backlog conversion trends in upcoming reports; any signs that GPU economic lives are extending would directly enhance long-term earnings forecasts.
- DigitalOcean: Track the company’s AI annual recurring revenue (ARR) disclosure to gauge whether the SMB full-stack strategy is translating into accelerated growth, particularly from inference and agent workloads.
- Nebius: Monitor capacity expansion timelines and M&A announcements, as these are key to Baird’s high-growth thesis; the Token Factory platform’s adoption will be an early indicator of commercial traction.
- Sector-wide: Watch for enterprise surveys and hyperscaler earnings commentary that confirm a rising share of inference in total AI workload spend — such data would support further re-ratings for neoclouds.
Risk & Opportunity Assessment
| Commercial Risk | Medium | Hyperscalers could intensify price competition for inference workloads, and policy restrictions on data-centre construction may limit expansion — though near-term demand is strong enough to absorb freed capacity as noted by Truist. |
| Competitive Risk | High | Meta’s planned cloud business and other neoclouds pose a direct competitive threat. Distinct business models such as Nebius’s vertical integration and DigitalOcean’s SMB focus provide some insulation but do not eliminate price pressure risk. |
| Regulatory Risk | Medium | Possible policy limits on data centre development, mentioned by Truist, could impede growth for power-intensive AI infrastructure providers, though no specific regulation has yet been enacted. |
| Reputation Risk | Low | No reputation concerns were raised in the analyst notes; all three companies are viewed as credible players in their respective niches. |
| Technology Disruption | Medium | The shift to inference is a positive disruption for these companies, but faster-than-expected innovation in chip efficiency or new model architectures could shorten GPU economic lifetimes, affecting CoreWeave’s depreciation thesis. |
| Commercial Opportunity | High | Broadening enterprise adoption of open-source models and inference workloads significantly expands the addressable market beyond the large hyperscalers, as flagged by both Truist and Baird. |
Comments 0