Neocloud AI compute providers target 20% market share by 2030

The rise of specialized GPU clouds forces a shift in how streaming and AI firms architect their backends, prioritizing raw performance for training while keeping data planes on legacy clouds. For the streaming ecosystem, this fragmentation introduces complexity in security and identity management that may offset the 70% cost savings on compute. As hyperscalers deploy custom silicon like Trainium and Maia, the price advantage of NVIDIA-distributor neoclouds will face structural pressure. Watch for whether CoreWeave can diversify its revenue beyond Microsoft and OpenAI to prove long-term vendor viability as GPU hardware rapidly depreciates.

CoreWeave has become the most visible neocloud provider, but its concentration risk is drawing scrutiny from analysts and investors alike. In May 2025, CoreWeave reported that Microsoft accounted for 62% of its revenue in the first quarter, a dependency that underscores the structural vulnerability of GPU-cloud startups that rely on hyperscaler contracts for scale. Lambda, meanwhile, has pursued a different path by securing a $480 million Series C round led by TWG Global Partners in early 2025 to expand its on-demand GPU cluster capacity, signaling that investor appetite for independent AI infrastructure remains strong despite the competitive overhang from AWS, Google Cloud, and Microsoft Azure.

The hyperscalers are not standing still. AWS has accelerated deployment of its Trainium2 chips, and in March 2025 Amazon confirmed that Trainium2-based EC2 Trn2 instances reached general availability with up to 4x better price-performance over comparable GPU instances, directly targeting the cost narrative that neoclouds use to win training workloads. Google Cloud followed with its Ironwood TPU generation, which the company described as delivering 4x the peak compute performance of its predecessor TPU v5p at inference-optimized power efficiency, positioning Google to undercut neocloud pricing on inference workloads where streaming platforms are most likely to need sustained compute. Microsoft's Maia 100 accelerator, still in limited internal deployment, represents a longer-term threat to NVIDIA-dependent providers.

Independent benchmarking adds nuance to the cost advantage narrative. In a June 2025 analysis, SemiAnalysis found that CoreWeave's effective GPU-hour pricing for H100 clusters was 30-40% below AWS on-demand rates but only 10-15% below reserved-instance pricing, suggesting that the headline savings narrow considerably for customers willing to commit to longer terms. Storage and data-egress costs further erode the gap. VAST Data and WEKA, both mentioned as storage-layer partners for neocloud deployments, have reported that data gravity and egress fees add 15-25% to total cost of ownership for multi-cloud AI pipelines, a friction point that streaming companies running large media libraries will recognize acutely. The net effect is that neoclouds hold a genuine but bounded edge on raw GPU throughput, while the full-stack economics remain contested.