AI Infrastructure & Cloud Pricing: Examples & Companies

What is it

AI Infrastructure & Cloud pricing is pricing for AI compute infrastructure — GPU clouds, serverless inference, and training platforms.

This is the bottom layer of the AI stack: the raw silicon and orchestration that every model, agent, and application ultimately runs on. Vendors in this category sell GPU-hours, inference seconds, tokens, per-output runs, and database operations rather than finished features. The corpus tracks 42 in-corpus companies here — the broadest product category on the site — from pure GPU marketplaces like Vast.ai and RunPod, to serverless inference clouds like Groq, Together AI, and DeepInfra, to per-second compute platforms like Modal and Lightning AI.

The category has widened well beyond bare GPU rental. It now includes hyperscale-class clouds like CoreWeave, Lambda, and Nebius; managed vector databases such as Pinecone, Qdrant, and Weaviate; model gateways like OpenRouter and Portkey; agent sandboxes (E2B, Modal); and the observability layer that watches it all (Helicone, LangSmith, Arize). What unites them is that they all sell consumption of infrastructure, priced by a metered unit rather than a per-seat license.

For the GPU-cloud core the underlying product is fungible — an H100 is an H100 no matter whose data center it sits in — so pricing, not features, is the primary battleground. That economic reality pushes the raw-compute vendors toward intensely competitive, transparent unit rates and a wider set of discount levers (spot, batch, reserved, committed spend) than any other product segment in the corpus. The category also carries the highest unit proliferation: a single vendor like DeepInfra meters across per-token, per-image, per-minute media, and per-GPU-hour units at once, because customers arrive with different workloads that each have a natural billing unit. For a deeper grounding in how to pick those units, see the guide on choosing the right usage metric.

How it works

Infra-cloud pricing is built from a small set of billing units, layered with availability tiers and volume discounts. Almost every raw-compute vendor here is fundamentally pure-usage — you pay for what you consume — and then bolts commitment or hybrid options on top for larger buyers.

On top of those units sit three near-universal discount levers. Availability tiers trade reliability for price: on-demand capacity at a fixed rate, spot or interruptible capacity that can be reclaimed mid-job (Vast.ai’s auction, RunPod’s Community Cloud, CoreWeave’s HGX H100 spot at $19.71/hr vs $49.24 on-demand). Batch discounts trade latency for price — Groq and Together AI both cut serverless inference 50% for asynchronous jobs. Reserved and committed spend trade flexibility for price, locking a discounted rate against a commitment — Together AI’s reserved cluster H100 drops to $3.09/hr, and CoreWeave quotes reserved contracts at up to 60% off on-demand.

Unit math: A pure GPU-second bill is GPU-seconds × per-second rate. Running a job for 90 minutes on a Modal H100: 5,400 sec × $0.001097 = $5.92. The same job on a fixed-hourly H100 at $2.89/hr (RunPod, billed by the hour) costs 2 hrs × $2.89 = $5.78 — but you pay for the rounded-up second hour even if the job finishes early, which is exactly the gap per-second billing closes.

Companies using this

Forty-two in-corpus companies sell AI infrastructure and cloud compute — the broadest set in the product-categories axis. They span raw GPU marketplaces like Vast.ai and RunPod, hyperscale GPU clouds like CoreWeave and Lambda, inference-specialized chip clouds like Groq and Cerebras, and the surrounding layer of vector databases, gateways, sandboxes, and observability that AI applications depend on.

Patterns observed

• Pure-usage is the default; commitment is the upsell. Almost every raw-compute company here lists pure-usage as its primary model — see fal, Novita AI, and DeepInfra for no-subscription pay-as-you-go, and Lambda and Nebius for per-GPU-hour clouds that bill from the first hour. The larger clouds then layer commitment on top: RunPod, Groq, Together AI, DeepInfra, and Cerebras all carry a commitment tier for enterprise spend, and CoreWeave earns the bulk of its revenue from sales-quoted reserved contracts.

• The billing unit keeps shrinking. Hourly GPU rental was the original unit, but the competitive frontier has moved to the second. Modal bills H100 at $0.001097/sec, Replicate at $0.001525/sec, and Lightning AI bills its Studios “by the second” against a credit pool. Per-second billing is now a selling point precisely because it eliminates the rounding waste of hourly rates. Agent sandboxes like E2B inherit the same per-second compute model for micro-VM run-time.

• Availability tiering is how marketplaces compete. Vast.ai runs an actual auction — interruptible bids from $0.194/hr — RunPod splits Secure Cloud from a cheaper Community Cloud, and CoreWeave publishes on-demand and spot side by side. The same hardware sells at two or three prices depending on how much reliability the buyer is willing to give up.

• Batch is the standard latency-for-price trade. Groq and Together AI both advertise a flat 50% discount for asynchronous batch inference, mirroring the batch-API discounts seen at the foundation-model layer. It has become a near-default SKU for any token-metered cloud.

• Free tiers are starter credits or capped plans, not free compute. Where a free tier exists it is almost always a credit grant or a capped starter: Modal’s $30/month, Anyscale’s $100 ACU credit, Cerebras’s $10 Developer tier plus a rate-capped free tier, and Pinecone’s capped Starter plan (2 GB storage, 1M read units/month). Perpetual free GPU time is not viable when the underlying cost is real silicon.

• The adjacent infra layer prices on operations, not silicon. Beyond raw GPUs, the category’s databases, gateways, and observability tools meter their own units. Pinecone bills read units, write units, and per-GB storage; OpenRouter passes through model prices at no markup and monetizes with a 5.5% credit-purchase fee; and Helicone prices log volume with a free tier and $79 Pro plan. These vendors compete more on features than on fungible-hardware price, but still converge on transparent usage meters.

Counterexamples & variants

The clearest variant is the pure per-output model, which abandons compute-time billing entirely. fal has no seats, no subscription, and no free tier — it charges per generated image (Seedream V4 at $0.03, Flux Kontext Pro at $0.04) or per megapixel and per-second video. For a buyer this is the opposite of a GPU-hour deal: you never see the hardware or the runtime, only the output you asked for. It works because generative-media inference has a predictable cost per run, but it would fail for long-running training jobs where output is not a meaningful unit.

A second variant is the credit-pool subscription, which looks like SaaS on the surface but is usage underneath. Lightning AI bundles a monthly credit allotment into freemium seat tiers, then bills per-second compute against those credits plus pay-as-you-go overage. Anyscale does the same with Anyscale Compute Units (ACUs). This blurs the line with the hybrid pricing model and trades the radical transparency of a raw GPU-hour rate for budget predictability.

A third variant is the percentage-fee gateway, where the vendor sells no compute of its own. OpenRouter routes across 400+ models at pass-through per-token pricing with zero markup and instead charges a 5.5% fee on prepaid credit purchases (minimum $0.80; 5% on crypto), plus a 5% BYOK fee after 1M free requests a month. There is no GPU-hour, no spot tier, and no reserved discount — the meter is the transaction, not the silicon. It is closer to a payments rail than a cloud.

The category’s true edge case is the inference-specialized chip cloud. Cerebras and Groq sell custom silicon (wafer-scale and LPU respectively) rather than commodity NVIDIA GPUs, so they cannot compete on raw GPU-hour rates the way Vast.ai does. Instead they price per token on the strength of speed — pitching tokens-per-second throughput as the value metric, with hardware contracts reserved for the enterprise tier. It is the one corner of the raw-compute category where the product is not fungible.

What this means for buyers vs vendors

For buyers

Normalize every quote to a comparable unit before you compare vendors — convert per-second and per-hour rates to the same basis and confirm whether you are billed for actual runtime or rounded-up hours, because per-second clouds like Modal can beat a cheaper headline hourly rate on bursty workloads. Watch for hidden node-level pricing on the big clouds: CoreWeave publishes 8-GPU node rates, so an HGX H100 node listed at $49.24/hr is really about $6.16 per GPU — always divide down before comparing to a single-GPU marketplace like RunPod at $2.89/hr.

Decide upfront how much you will sacrifice for price. Spot and interruptible capacity (Vast.ai, RunPod Community Cloud, CoreWeave spot) and batch discounts (Groq, Together AI) can halve your bill if your jobs tolerate interruption or latency, and reserved rates go further still — Together AI’s reserved H100 is $3.09/hr against dedicated rates. Model your actual utilization before signing a committed-spend deal: those discounts only pay off above a usage floor that the pure-usage default already covers below.

Finally, remember the meter changes with the layer. For a vector database like Pinecone your bill is driven by read units, write units, and storage, not GPU-hours — and a $50/month Standard minimum can dominate a small workload. For a gateway like OpenRouter the cost is a percentage on top of pass-through model prices. The guide on usage invoicing and billing cycles is a useful primer on reading these mixed meters, and you can sanity-check competitive token rates against the live pricing calculator.

For vendors

Pure-usage is table stakes for raw compute — buyers expect to pay from the first second with no minimum, so lead with a transparent unit rate and layer commitment on top rather than gating entry behind a subscription. Compete on the dimension you actually win: marketplaces like Vast.ai win on raw price via auctions and spot capacity, hyperscale clouds like CoreWeave win on cluster scale and free egress, and chip clouds like Cerebras and Groq win on speed and price per token, not per hour.

The infrastructure required is non-trivial. Per-second metering, spot reclamation, batch queuing, credit-pool accounting, and multi-unit invoicing all demand precise usage tracking, covered in the guide on tracking and metering usage events. Vendors selling the adjacent layer — vector DBs, gateways, observability — differentiate on features but should still expose a clean, meter-first bill; opaque pricing reads as a red flag next to the radical transparency buyers now expect from the GPU clouds. This commitment-on-top-of-usage shape is documented further in the infra commitment as standard trend.

Explore this theme in the knowledge graph →