Salad undercuts Vast.ai on RTX 5090 distributed GPU cloud pricing

The aggressive pricing for consumer-grade hardware like the RTX 5090 provides streaming platforms with a cost-effective alternative to traditional enterprise data centers. By utilizing distributed networks, these providers allow developers to scale compute-heavy tasks such as transcoding and AI-driven personalization at rates well below standard cloud tiers. This shift forces a re-evaluation of the streaming technology stack, as decentralized resources become viable for production-grade workloads. As the market matures, watch for whether Vast.ai adjusts its auction-based bidding system to match Salad's lower fixed-rate benchmarks for high-demand NVIDIA hardware.

Salad has been expanding its distributed GPU network beyond gaming into enterprise AI and media workloads. In early 2025, Salad raised $20 million in a Series B round led by Grisham Robotics to scale its marketplace of consumer GPUs for inference and rendering tasks. The company's model aggregates idle consumer hardware, primarily NVIDIA GeForce cards, into a unified compute layer that undercuts hyperscaler pricing by a wide margin. Vast.ai operates a similar peer-to-peer marketplace but uses an auction-based bidding system where hosts set their own rates, creating more price variability across GPU SKUs.

The competitive dynamics between Salad and Vast.ai reflect a broader trend in decentralized compute infrastructure. Vast.ai secured $18 million in funding in late 2024 to expand its GPU marketplace and position itself as an open marketplace for AI training and inference. Both companies are targeting the same cost-sensitive segment that includes streaming platforms running transcoding pipelines and AI-driven content personalization. The pricing pressure from these distributed providers is also being felt by traditional cloud vendors, as enterprises increasingly evaluate hybrid approaches that blend hyperscaler capacity with cheaper distributed alternatives for burst workloads.

On the hardware side, the RTX 5090 and RTX 4090 have become the workhorses of distributed GPU clouds due to their high VRAM-to-cost ratio. NVIDIA's RTX 5090 launched in January 2025 with 32 GB of GDDR7 memory, making it attractive for inference workloads that previously required data-center-class cards like the A100 or H100. Salad's fixed-rate pricing at $0.25 per hour for the RTX 5090 represents a significant discount compared to on-demand rates at AWS or Azure for equivalent VRAM capacity. Vast.ai's marketplace model means RTX 4090 instances can dip below $0.20 per hour during low-demand periods, though availability fluctuates with host participation. For streaming engineers evaluating distributed GPU cloud pricing for transcoding or real-time AI features, the tradeoff between Salad's predictable rates and Vast.ai's variable but sometimes lower bids remains the central decision point.