AI cloud: which startup is ahead?

Which startups actually count as AI cloud companies?

Eight private startups belong in the serious AI cloud comparison today, spread across three distinct parts of the market.

We include startups that sell direct access to GPUs, managed training, production inference, or purpose-built AI data centers. Fireworks AI, Together AI, and Baseten sit closest to the model-serving layer. Lambda and Runpod sell more direct access to compute. Crusoe, Nscale, and Fluidstack combine cloud services with large infrastructure projects.

We exclude AWS, Microsoft Azure, Google Cloud, Oracle Cloud, and other established hyperscalers. CoreWeave and Nebius also sit outside the ranking because both are now public companies. They remain useful benchmarks. CoreWeave generated $5.13 billion of revenue in 2025 and ended that year with a $66.8 billion backlog, according to its annual results. Every private startup remains below that operating scale today.

Funding needs a warning label. Software-heavy providers mostly raise equity, whereas data-center builders also use equipment loans, project finance, credit facilities, and customer-backed debt. The figures below show the approximate disclosed capital we could verify and flag the companies where debt makes the comparison messy.

Startup	What it does	Approximate disclosed funding	Main lane
Fireworks AI	Custom model training and production inference	About $1.83B in equity	Inference cloud
Together AI	Inference, fine-tuning, training, and GPU clusters	About $1.33B in equity	Full-stack model cloud
Baseten	Deploys and operates custom models across multiple clouds	More than $2B in equity	Inference platform
Lambda	On-demand GPUs, managed clusters, and private supercomputers	About $2.3B in equity, plus debt facilities	GPU cloud
Nscale	Builds AI factories and sells cloud infrastructure	At least $3.7B across equity and SAFE rounds, plus several debt facilities	Hyperscale infrastructure
Crusoe	Develops power, data centers, GPU clusters, and cloud software	About $3.9B across debt and equity	Vertically integrated AI cloud
Runpod	Self-service GPUs, serverless inference, and clusters	About $122M in equity	Developer cloud
Fluidstack	Builds and operates custom AI data centers	At least $830M in equity	Custom infrastructure

Is there a clear AI cloud leader today?

Fireworks AI is ahead overall today, with Together AI close behind and Lambda leading the more traditional GPU-cloud race.

The answer changes as we move across the stack. Fireworks currently has the best mix of disclosed recurring revenue, production usage, customer quality, and recent growth. Together is already close on bookings and covers more of the training-to-inference workflow. Lambda has the strongest broadly available private GPU product for customers that want clusters directly.

Crusoe, Nscale, and Fluidstack lead a separate contest: building enormous amounts of future capacity for hyperscalers and frontier laboratories. Their contracts are larger than those of the inference platforms, but much of the compute still has to be financed, built, energized, equipped, and accepted by the customer. We give operating workloads more weight than planned campuses when deciding who is ahead now.

Together could overtake Fireworks if its bookings turn into recurring revenue at the expected rate. Lambda could move up if its large Microsoft deployment produces a much broader cloud business. Crusoe or Nscale could eventually dwarf all three if their infrastructure pipelines arrive on schedule.

Part of the market	Leader now	Closest challenger	How clear is the lead?
Production inference	Fireworks AI	Together AI, Baseten	Fireworks has the strongest verified revenue and token figures
General GPU cloud	Lambda	Crusoe, Runpod	Lambda has the most mature large-cluster product
Future physical capacity	Nscale	Crusoe, Fluidstack	Nscale leads disclosed GPU commitments; Crusoe leads contracted power
Developer self-service	Runpod	Lambda	Runpod has crossed one million developers
Overall private AI cloud	Fireworks AI	Together AI, Lambda	A real lead, though still narrow enough to change quickly

If you want more recent data on this point, please see our latest AI infrastructure market report.

Who is making real money from AI cloud services?

Fireworks AI makes the most money we can verify in the private field.

In its latest financing announcement, Fireworks said it had passed $1 billion in annualized revenue. Its previous major disclosure put the figure above $280 million, which means the run rate grew by at least 3.6 times between the two announcements. The company also says more than 95% of its traffic now comes from models specialized for customer data and specific jobs. Customers are using Fireworks inside real products and workflows, where migration becomes harder than changing a model API during an experiment.

Together AI is operating in the same broad range, although its best public figure measures bookings. Its latest financing materials said annual bookings had crossed $1.15 billion and that the platform served thousands of paying customers. Sacra separately estimates roughly $1 billion in annualized revenue. We treat the estimate as plausible because it sits close to the company’s bookings, yet recognized revenue and signed bookings remain different measures.

Baseten disclosed 20-fold revenue growth over the last year, and Sacra estimates a run rate near $600 million. Lambda’s latest outside estimates place annual revenue around $500 million. Runpod disclosed more than $120 million in annual recurring revenue before announcing its latest funding round. These businesses have clearly crossed into commercial scale, even though private-company accounting makes an exact ranking below the top two difficult.

The infrastructure builders reveal less. Fluidstack reported about $66 million of revenue for 2024 before winning much larger projects. Nscale publishes extensive contract and financing news without a comparable current revenue figure. Crusoe mixes data-center development, energy projects, and cloud services, so one headline revenue number would hide more than it explains. Their strongest proof is contracted construction and capacity, while Fireworks, Together, Baseten, Lambda, and Runpod can point to recurring cloud usage today.

Which AI cloud startup is growing fastest now?

Fireworks is growing fastest on hard numbers today; Baseten and Runpod are sprinting from smaller bases.

Fireworks moved from more than $280 million to above the billion-dollar mark in annualized revenue, an increase of at least 257%. Daily usage climbed from more than 10 trillion tokens to more than 40 trillion over roughly the same period. Revenue and workload grew together, which is stronger evidence than a single giant contract.

Baseten reported 20-fold revenue growth and 40-fold inference-volume growth during its latest fundraising cycle. The company withheld the starting values, which prevents a clean scale comparison with Fireworks. Even without them, 20-fold revenue growth puts Baseten among the fastest-growing companies here. A 40-fold jump in inference volume shows that real model traffic drove the expansion, beyond any pricing effect or a few reserved clusters.

Runpod offers the clearest developer-growth pattern. It served more than 500,000 developers when it announced $120 million in recurring revenue, then crossed one million developers before its new $100 million round. Its disclosed user base roughly doubled within months. Serverless requests have passed ten billion, giving the developer count a useful second measure.

Together is also moving quickly. The business secured more than 500 megawatts of independently financed compute commitments as open-model usage across the industry tripled over twelve months. Its next test is turning that demand into durable, recognized revenue.

Who has won the best AI cloud customers and contracts?

Lambda has the largest disclosed cloud contract, while Fireworks, Together, and Baseten have the strongest collection of AI-native production customers.

Lambda signed a multibillion-dollar agreement with Microsoft involving tens of thousands of Nvidia GPUs. That deal proves Lambda can finance and operate very large clusters for one of the world’s most demanding buyers. It also creates concentration: Microsoft is a customer, a cloud rival, and a company with enough internal capacity to change suppliers when the economics shift.

Fireworks works with Cursor, Harvey, Samsung, Uber, DoorDash, Notion, Shopify, and Upwork. Together names Cursor, Cognition, Decagon, ElevenLabs, and Suno among thousands of customers. Baseten lists Cursor, Notion, Harvey, HubSpot, OpenEvidence, Abridge, and Decagon. These logos carry more weight than a generic partnership announcement because AI performance sits inside the customers’ products. A slow or expensive inference provider directly hurts a coding assistant, voice agent, medical application, or legal research tool.

The infrastructure specialists have landed the largest strategic relationships. Nscale has major commitments tied to Microsoft and OpenAI. Crusoe is building campuses for Oracle and Microsoft and supplies infrastructure used across the AI ecosystem. Fluidstack was selected to help Anthropic deliver a $50 billion US data-center program. The size is impressive, though customer concentration is far higher than at the inference platforms.

One pattern deserves attention: the same companies can be customers, suppliers, investors, and rivals. Microsoft buys capacity from Lambda and Nscale while selling Azure. Fireworks and Together can run workloads on infrastructure supplied by Crusoe while competing with Crusoe’s own managed AI products. Customers are assembling their AI stack from overlapping specialists, and several vendors appear on both sides of the same deal.

If you want more recent data on this point, please see our latest AI infrastructure market report.

Who controls the most AI compute and can actually deliver it?

Nscale controls the largest disclosed future GPU commitments, while Lambda has the strongest evidence of large clusters that ordinary customers can use now.

Nscale announced commitments covering roughly 200,000 Nvidia GB300 GPUs for Microsoft across several locations. It has since expanded its roadmap with more than 66,000 Rubin GPUs planned in Portugal, a 1.35-gigawatt West Virginia project, and further European deployments. The company has also added a $900 million revolving credit facility, a $1.4 billion GPU-backed loan, and other project financing. Few startups can assemble that combination of customer commitments, hardware orders, sites, and capital.

Nscale still has to turn this paper scale into live capacity. Some projects begin service in phases, and an investigation by The Guardian found that its proposed Loughton site was still at an early planning stage after heavy promotion. We count the contracted demand fully and discount the capacity that still depends on construction. Announced GPUs only become a cloud business after the buildings, power, networking, cooling, and hardware work together.

Crusoe measures scale in power. The company says it has contracted 4.9 gigawatts across data-center projects and Crusoe Cloud, with more than 40 gigawatts in its wider development pipeline. Its first buildings at the 1.2-gigawatt Abilene campus are operational, and additional campuses are under construction. Crusoe also manufactures long-lead electrical components and develops energy solutions, giving it more control over delivery than a provider that simply leases finished data-center space.

Lambda has the clearest purchasable product. Its self-service clusters range from 16 to more than 2,000 GPUs, while private superclusters are offered from 4,000 to more than 165,000 GPUs. Customers can already rent H100, B200, GB300, and other Nvidia systems. Runpod serves the opposite end of the market through on-demand GPUs across 31 regions, serverless endpoints, and multi-node clusters. Fluidstack has shown unusual speed by helping Poolside deploy more than 2,500 GPUs within 48 hours, although its largest Anthropic facilities are still rolling out.

Which AI cloud is strongest for training large models?

Lambda currently leads large-model training among private AI clouds.

Its advantage comes from a product that already spans single instances, self-service clusters, managed Slurm and Kubernetes, and private supercomputers containing thousands of GPUs. Lambda lists clusters of up to more than 165,000 GPUs and has a Microsoft agreement that confirms demand at the tens-of-thousands-of-GPUs level. That combination of availability and proven buyer trust puts it ahead of startups whose largest systems remain construction projects.

Crusoe is the closest infrastructure challenger. It controls more of the physical chain, including energy, campuses, modular electrical equipment, networking, cloud operations, and managed training software. The first Abilene buildings are live, and Crusoe Cloud supports Nvidia and AMD accelerators. A frontier lab that cares about a dedicated campus and long-term power supply may prefer Crusoe’s model to a conventional rented cluster.

Among the inference-led companies, Together has the best training offer. It provides managed GPU clusters, pre-training, fine-tuning, reinforcement learning, and production serving in one environment. A Together customer described training on a 72-GPU Blackwell cluster and moving directly into deployment without transferring model artifacts. That workflow will appeal to AI startups that need dozens or hundreds of GPUs, then want the same provider to serve the finished model.

Nscale could take this position later. Its planned systems are larger, newer, and increasingly tied to Microsoft and OpenAI. Today, Lambda has more evidence of a broadly available training cloud, while Nscale has the more ambitious future fleet.

Who is winning the AI inference cloud race?

Fireworks AI is the current inference leader, with Together close enough to make this a genuine two-company race.

Fireworks says it now processes more than 40 trillion tokens each day, up from more than 10 trillion at its previous major funding announcement. That is a fourfold increase in traffic. The company has also moved beyond serving generic open models: over 95% of its tokens come from models adapted to customer data and tasks. Its work with Cursor and Harvey places that customization inside high-value products and daily workflows.

Together reported more than $1.15 billion in annual bookings and says it serves over one million developers. Its customer examples add useful performance detail. Decagon reported inference costs falling sixfold and voice latency dropping below 400 milliseconds. Cursor uses Together’s Blackwell infrastructure for production inference and model iteration. These cases make Together more credible than a provider that relies mainly on provider-run benchmark charts.

Baseten is a serious third player. The company says inference volume grew 40 times over the last year and names Cursor, Notion, Harvey, Abridge, and OpenEvidence as customers. Its main strategic difference is infrastructure flexibility: Baseten can source compute across more than 20 cloud providers. Customers gain portability and supply resilience, although Baseten owns less of the physical stack than Lambda or Crusoe.

Direct token comparisons remain imperfect because providers count input, output, cached, and internal reasoning tokens differently. Call counts create a similar problem: Baseten’s reported one billion daily inference calls could represent many small requests or fewer expensive multimodal jobs. We use workload figures alongside revenue growth, customer evidence, and deployment depth. Taken together, the evidence puts Fireworks first, Together close behind, and Baseten clearly third.

If you want more recent data on this point, please see our latest AI infrastructure market report.

Which AI cloud gives customers the best economics?

Runpod usually offers the cheapest route to raw GPUs, while Fireworks and Together create larger savings when inference software dominates the bill.

Lambda’s public pricing provides a useful baseline. An eight-GPU H100 SXM instance currently costs $3.99 per GPU-hour, while a single H100 SXM costs $4.29. Lambda charges no egress fees. Runpod uses a broader and more variable supply base, so customers can often find lower hourly prices, especially for flexible workloads that can move between regions or tolerate different service levels.

Production inference has a different cost structure. GPU utilization, batching, quantization, caching, model choice, and engineering time can matter more than the sticker price of one GPU-hour. Together says Decagon cut inference costs sixfold after moving to its stack. Fireworks advertises large gains from its optimization software, and Baseten customers can mix open and closed models to lower costs on workloads where the quality difference is small.

The claims are difficult to standardize. Each provider chooses favorable models, latency targets, batch sizes, and hardware. A benchmark winner on one model can lose on another after a software update. Buyers should test their own traffic and compare cost per completed task, p95 or p99 latency, failure rates, and engineering effort. For a small team buying flexible compute, Runpod has the clearest price advantage. For a company serving millions of model calls, Fireworks or Together can save more by squeezing additional work from each GPU.

Who is using AI cloud funding most efficiently?

Runpod has built the most business with the least equity, even after its new $100 million round.

As noted above, Runpod had reached more than $120 million in recurring revenue after raising about $22 million. Its new $100 million round brings cumulative equity to roughly the same level as disclosed recurring revenue. That is still unusually strong for an infrastructure startup, especially beside companies that raised billions before publishing comparable revenue.

Using the Fireworks revenue figure discussed earlier, its rough revenue-to-equity ratio is around 0.55. Together’s outside revenue estimate against approximately $1.33 billion of equity produces a higher ratio near 0.75, though the underlying revenue figure remains unverified. Baseten’s exact current ratio is less reliable because the company disclosed growth rates without publishing revenue.

Infrastructure builders need a separate test. Nscale, Crusoe, and Fluidstack use capital to buy land, secure power, build facilities, and install equipment that can produce revenue for years. Comparing their current revenue with a software platform’s revenue would punish them for the normal construction cycle. We instead ask whether each dollar is backed by a signed customer, a financeable site, and a believable delivery schedule. Crusoe currently gives us the strongest physical evidence. Nscale has assembled the largest commitments, with more execution still ahead. Fluidstack has one exceptional customer and far less diversification.

What can the leading AI cloud startups do that competitors cannot easily copy?

Crusoe owns the hardest physical advantage to copy, while Fireworks makes customers hardest to pull away.

Crusoe works across energy, site development, data-center construction, modular electrical equipment, GPU infrastructure, and cloud software. Rebuilding that chain would require years of permits, power agreements, factories, construction knowledge, supplier relationships, and operating experience. Its contracted capacity gives that system commercial proof, and the first Abilene buildings show that part of the pipeline is already running.

Fireworks’ advantage sits closer to the customer. When a company fine-tunes a model, builds evaluation systems, connects proprietary data, and optimizes production traffic on one platform, moving becomes a real engineering project. Specialized models now account for almost all Fireworks traffic, suggesting that customers are going deeper than simple API calls.

Together combines respected systems research with training, post-training, clusters, and inference. Baseten offers a useful hedge against supply concentration by spreading workloads across more than 20 clouds. Lambda has years of experience operating Nvidia clusters and unusually close hardware relationships. Nscale is building leverage through sovereign projects, power access, and giant customer commitments. Runpod’s large developer base gives it distribution and a steady stream of smaller workloads that larger providers often overlook.

These advantages can still erode. Hyperscalers have their own chips, enterprise sales teams, and much larger balance sheets. Open-source serving software keeps improving, and Nvidia can influence which cloud receives scarce hardware. The startups must keep shipping faster, maintaining better utilization, and giving customers practical reasons to stay.

If you want more recent data on this point, please see our latest AI infrastructure market report.

Where could the AI cloud leaders fail?

Customer concentration and construction risk could break the infrastructure leaders, while price compression is the bigger threat for inference clouds.

Nscale depends heavily on Microsoft-linked projects. Fluidstack’s rise is closely tied to Anthropic’s infrastructure program. Lambda’s largest disclosed agreement also comes from Microsoft. Crusoe has more projects and a broader physical platform, yet a small number of hyperscale customers still drive the largest campuses. A delayed facility or changed customer plan can move billions of expected revenue into another year.

The physical economics are unforgiving. New GPU generations arrive quickly, while the debt used to finance a cluster can last several years. A provider that buys too early, overpays for power, or misses its delivery window may end up renting older hardware at lower prices before the financing is repaid. CoreWeave offers a useful public warning: it produced $5.13 billion of revenue in 2025, yet interest expense reached $1.23 billion and the company reported a $1.17 billion net loss.

Inference providers face falling software prices. vLLM, SGLang, TensorRT-LLM, model quantization, and hyperscaler discounts keep improving. Fireworks, Together, and Baseten need to create value through optimization, customization, reliability, and workflow integration. Plain access to a popular open model will become a commodity quickly.

Nvidia’s many roles complicate the picture. It supplies chips, invests in several providers, and sometimes participates in customer or financing arrangements around the same ecosystem. Those links help startups obtain hardware and capital, but they can make demand look more independent than it is. We trust diversified paying workloads more than deals where supplier, investor, and customer relationships overlap heavily.

Which AI cloud startup has the strongest momentum currently?

Fireworks has more momentum than any private AI cloud currently.

Baseten, Together, and Runpod are moving fastest behind it, but Fireworks has just combined a large funding round with revenue and daily token traffic multiplying since its previous major disclosure. Its valuation rose sharply alongside the operating business, with usage and sales providing the support.

Baseten’s latest update is nearly as striking: revenue grew 20 times, inference volume grew 40 times, and the company raised $1.5 billion to expand an already active production platform. Together added $800 million, secured more than 500 megawatts of future compute, and continues to win demanding AI-native customers. Runpod roughly doubled its disclosed developer base and added enough capital to move beyond its scrappy, low-funding phase.

Crusoe is moving fastest among the physical infrastructure companies. Lately it has taken contracted capacity close to five gigawatts, launched self-service inference and serverless fine-tuning, announced another one-gigawatt Texas campus, and expanded its power partnerships. Nscale is raising and financing capacity at an even larger announced scale, though delivery evidence still trails the size of its press-release pipeline. Fluidstack’s new $830 million round gives it more room to execute for Anthropic, but that single relationship still shapes much of the story.

Which AI cloud startups are actually ahead?

Fireworks AI is ahead overall today, Together AI is the closest challenger, and Lambda remains the leader for direct GPU infrastructure.

We give the top position to Fireworks because it has the cleanest combination of current revenue, production usage, growth, strong customers, and customized workloads. The company has passed $1 billion in annualized revenue, while daily traffic has reached more than 40 trillion tokens. The gap over Together is smaller than the gap between the pair and most of the field.

Together ranks second because it covers more of the model lifecycle. Annual bookings have already passed $1.15 billion, while its training, post-training, inference, and cluster products give it a broader offer than Fireworks. We keep it behind because bookings provide less certainty than recognized recurring revenue, and its most widely quoted revenue number remains an outside estimate.

Lambda takes third. It has the most mature large-scale GPU product, a multibillion-dollar Microsoft agreement, and clusters customers can buy now. Baseten ranks fourth after an extraordinary year of growth and a strong customer list. Its position could rise quickly once the company provides clearer absolute revenue and workload figures.

Crusoe comes fifth, ahead of Nscale, because more of its physical system is visible and operating. Nscale has the larger announced GPU program and could become the biggest company in this group, but several flagship projects still carry meaningful delivery risk. Runpod ranks seventh: it leads in developer reach, accessibility, and capital efficiency, while enterprise contract size remains smaller. Fluidstack ranks eighth because Anthropic has given it an enormous opportunity, with less evidence so far of a diversified cloud business.

The ranking can change under clear conditions. Together moves to first if its bookings convert into recurring revenue above Fireworks’ level. Lambda rises if the Microsoft deal broadens into a multi-customer cloud operating at several billion dollars of revenue. Crusoe or Nscale can jump into the top three after their contracted capacity becomes operational and produces repeatable cloud economics. Baseten can challenge the top two by pairing its growth rates with verified absolute figures.

Rank	Startup	Why it is here
1	Fireworks AI	Best verified mix of revenue, production traffic, growth, customer quality, and customized inference
2	Together AI	Similar commercial scale, broader platform, and strong customers; bookings are less conclusive than recurring revenue
3	Lambda	Strongest mature private GPU cloud and clearest large-cluster delivery record
4	Baseten	Exceptional growth and customer traction, with less public absolute data
5	Crusoe	Strongest physical advantage and substantial operating infrastructure
6	Nscale	Largest future GPU commitments, with higher delivery risk
7	Runpod	Large developer reach, strong accessibility, and impressive funding efficiency
8	Fluidstack	Huge Anthropic opportunity, with customer concentration and execution still defining the case

If you want more recent data on this point, please see our latest AI infrastructure market report.

Who is the author of this content?

NEW MARKET PITCH TEAM

We track new markets so founders and investors can move faster

We build living "market pitch" documents for emerging markets: AI, synthetic biology, new proteins, and more. Instead of outdated PDFs or hallucinated LLM answers, our clients get a clean, visual, always-updated view of what's really happening: key players, deals, regulations, and signals that matter. Learn more about us.