AMD’s Advancing AI: AI Hardware Announcements & Neocloud Providers Take Center Stage

On June 12, 2025, AMD hosted its annual Advancing AI event in San Jose, California, where it announced new Artificial Intelligence (AI) chips, system designs, and partnerships to strengthen its position in the growing AI compute market.The announcements underlined AMD’s commitment to advancing its portfolio with the launch of the AMD Instinct MI350 Series Graphics Processing Units (GPUs), a roadmap toward CDNA 4-based Instinct MI400 GPUs, the rack-scale Helios platform, UALink interconnects, and continued enhancements to the ROCm software stack with the launch of ROCm 7.

As well as these hardware and software announcements, it was notable to see that, aside from major hyperscalers like Oracle Cloud Infrastructure (OCI), AMD is also expanding its relationships with neocloud providers. Vultr announced it will be among the first to deploy AMD’s upcoming Instinct MI355X GPUs starting in 3Q 2025, enhancing its AI-focused infrastructure offerings alongside its existing MI325X and MI300X deployments. Crusoe committed to purchasing roughly 13,000 MI355X units (investing US$400 million) to power a new liquid-cooled AI data center optimized for high-efficiency inference workloads. Meanwhile, AI server vendors/Original Equipment Manufacturers (OEMs) like Aivres are integrating the MI350 Series into new AI server designs, signaling how AMD's next-gen accelerators are enabling smaller, GPU-rich cloud platforms to compete on performance, scale, and cost.

This growing momentum among neocloud providers—caught between rapid adoption of AMD’s latest AI accelerators, increasing infrastructure demands, and a fluid vendor landscape—makes it a timely and important moment to examine whether this emerging segment can truly redefine the AI cloud market or if it’s headed for structural headwinds.

At the heart of it, the term “neocloud” is a market positioning label for specialized cloud service providers, focusing primarily on advanced microprocessor and accelerator fleets to rent and as-a-Service cloud solutions optimized for Artificial Intelligence (AI) workloads, addressing gaps, pushing boundaries, and answering custom market demands that hyperscalers have not fully captured. They differentiate through pricing, breadth and depth of AI infrastructure stack, compliance requirements, and local points-of-presence. As enterprises increasingly adopt large-scale AI workloads that demand more flexible, high-density, and low-latency compute environments, neocloud providers are often better positioned to meet these evolving requirements than traditional cloud models. Reflecting this shift, ABI Research in its recent report, Mapping the Neocloud Market Landscape: Definitions, Technology Characteristics & Business Models, forecasts the revenue from GPU-as-a-Service (GPUaaS) offerings by neocloud providers are projected to surpass US$65 billion, up from US$24 billion in 2024.

While this new group of providers certainly offers an attractive new channel for chipset manufacturers to meet growing enterprise demand, there are a few important aspects that need to be considered when assessing the realistic influence of these companies. First, across the industry, the success of neocloud providers is driven by their integration into the NVIDIA Partner Network, which gives them access to accelerators and technical support. This increases the strategic depth that NVIDIA has within these neocloud providers. As a result, it will take other vendors like AMD additional effort to break into this segment and establish themselves.

Second, neoclouds face Return on Investment (ROI) risks due to heavy Capital Expenditure (CAPEX) needs and lack the ad, e-commerce, or search revenue buffers of hyperscalers. Currently, they benefit from taking a very opportunistic approach and enjoy the benefits of being the center of everyone’s attention (i.e., very favorable agreements), which mitigate the CAPEX needs in the short term. Without strong ecosystems, enterprises are hesitant to commit long term, often using neoclouds for low-cost training and then shifting workloads to their own Virtual Private Clouds (VPCs). The strategic battle ahead extends beyond GPU access to orchestration, pricing power, and ecosystem control.

Neocloud providers have rapidly gained visibility as the AI compute market explodes, but their long-term role will depend less on current momentum and more on their ability to evolve into structurally relevant players in a cloud landscape still dominated by hyperscalers. In the short term, their success is driven by clear gaps in the market, outlined earlier in this ABI Insight. Over time, however, their success will depend on a few different factors like their continuous access to GPU supply chains, their ties to the broader ecosystem, different regional and regulatory dynamics, their integration capabilities, and the competitive response from hyperscalers. For chipset manufacturers like AMD, this leads to a few recommendations:

• Treat Neoclouds as Strategic Amplifiers: AMD and other chipset manufacturers should continue engaging neocloud providers as fast-moving partners to showcase performance leadership, test new architectures, and address gaps underserved by hyperscalers. These providers offer valuable early-mover credibility in Generative Artificial Intelligence (Gen AI) and inference markets. However, they should be treated as strategic amplifiers. Maintaining diversified go-to-market pathways through hyperscalers, OEMs, and enterprise integrators remains critical for long-term scalability and risk mitigation.
• Ensure Ecosystem Portability and Limit Overexposure Through Tiered Engagement Models: To avoid overreliance on any single segment, chipset vendors should build cloud-agnostic hardware and software designed for portability across neoclouds, hyperscalers, and hybrid enterprise environments. This includes standardizing the software stack, enabling interoperability with third-party components, and supporting open interconnect technologies. While solutions like UALink address smaller-scale and scale-up needs, scale-out approaches such as UEC may be better aligned with the distributed nature of neocloud architectures. At the same time, vendors should apply tiered engagement models that account for the financial maturity and operational resilience of each neocloud partner—offering flexibility where needed, but with safeguards against volatility, consolidation, or sudden shifts in demand.
• Recognize and Navigate Varying Depths of Strategic Integration Across the Ecosystem: In all of this, chipset manufacturers should recognize the different degrees to which each of them is embedded within the neocloud landscape. NVIDIA, for example, has established closer, often exclusive relationships through early access programs, proprietary software stacks (e.g., CUDA), and co-designed infrastructure. In contrast, AMD and others may benefit from a more open, multi-vendor positioning—but must work harder to build durable value beyond hardware alone. Strengthening developer tools, orchestration integration, and co-marketing programs can help close this strategic gap and create longer-term alignment with neocloud partners.