Can neoclouds really beat AWS?
In our updated market reports, you will find everything you need
SUMMARY
Yes, neoclouds can really beat AWS for dedicated AI computing, but they are nowhere close to replacing AWS as the default cloud for ordinary companies.
The competitive opening exists because the scarce product is no longer a generic server. It is a complete, powered AI cluster with thousands of accelerators, fast networking, cooling and a delivery date the customer can trust.
Neoclouds have already won the narrow contest. CoreWeave, Nebius and other specialists are securing multibillion-dollar commitments from companies that also buy heavily from AWS, Azure and Google Cloud.
The largest AI customers are not choosing one cloud winner. They are assembling portfolios of capacity across providers because their demand is bigger than any single construction pipeline and one delayed campus can hold back an entire model generation.
Neocloud economics look strongest when a customer can reserve and fully use a large fleet for years. AWS becomes harder to displace when workloads are smaller, unpredictable, globally distributed or closely tied to data and software already running inside its ecosystem.
CoreWeave shows both sides of the model. It is growing far faster than AWS and has an enormous backlog, but it also carries heavy debt, high interest costs, operating losses and major exposure to a small number of customers.
The long-term test begins when advanced GPUs become easier to obtain. Providers that mainly resold scarce NVIDIA capacity will lose leverage; those with better scheduling, inference, monitoring, networking and workload software may keep it.
Inference could become more important than training for neoclouds because successful AI products produce continuous demand. Dedicated production fleets can stay busy for years, although AWS has a major advantage when customers want AI connected directly to existing applications, security systems and corporate data.
AWS’s strongest defense is not its size alone. It owns a global cloud platform, generates substantial operating profit and can steer more workloads toward Trainium rather than paying NVIDIA’s margin on every accelerator.
The neocloud market can grow dramatically without producing another AWS. A plausible outcome is a handful of global AI infrastructure companies, several regional specialists and many facilities owned by infrastructure investors but operated through neocloud software.
The real divide is becoming clearer: AWS can remain the main home for business data and applications while losing selected training and inference contracts. Neoclouds are building a valuable new layer of the cloud market, not inheriting the whole thing.
Can Neoclouds Really Beat AWS?
Why are neoclouds suddenly taking business from AWS?
Neoclouds are taking AI work from AWS today because the scarce product has shifted from ordinary cloud servers to entire powered clusters containing thousands of advanced chips.
A neocloud is a cloud provider built mainly for AI computing. CoreWeave, Nebius, Lambda, Crusoe, Fluidstack and Nscale spend most of their effort on GPUs, dense data centers, fast networking and the software needed to run large AI jobs. AWS has to support all of that alongside databases, storage, cybersecurity, business applications and hundreds of other services.
That narrower focus has become valuable. Meta recently expanded its CoreWeave commitments by approximately $21 billion. It also signed a five-year agreement with Nebius that could reach $27 billion. Microsoft has bought dedicated capacity from CoreWeave, Lambda and Nebius despite owning Azure.
These companies have not suddenly decided that AWS or Azure are bad clouds. Their AI plans are simply growing faster than any one provider can build data centers, secure power and install new chips. A frontier AI company can consume the output of an entire campus, so buyers now reserve capacity wherever they can find the right hardware and delivery schedule.
AWS therefore faces a different kind of competitor. Neoclouds offer fewer services, but they can devote a whole facility to one customer and one workload. For the largest AI buyers, that can be more useful than access to the broadest cloud catalog.
What would it actually mean for neoclouds to beat AWS?
A neocloud can beat AWS on a major AI contract years before it has any chance of beating AWS as a cloud company.
There are several very different competitions hiding inside the title. A provider could win a large model-training project, become cheaper for high-volume inference, attract more AI startups or replace AWS as a company’s main technology platform. Those outcomes should not be treated as interchangeable.
Neoclouds are already credible contenders for dedicated training clusters. A buyer that needs tens of thousands of identical accelerators in one place may care more about availability and cluster performance than about the provider’s database products.
Inference is becoming another realistic target. Large AI applications generate steady workloads that can fill specialized infrastructure for years. The opportunity becomes much harder when demand is unpredictable, spread across many countries or connected to data and applications already running inside AWS.
Replacing AWS across an entire company would require global regions, storage, networking, identity tools, databases, compliance certifications and technical support covering almost every type of software. No neocloud currently offers anything close to that range.
| What “beating AWS” could mean | Neocloud position today | Our judgment |
|---|---|---|
| Winning a giant model-training contract | Already happening | Neoclouds can win |
| Running high-volume AI inference | Early but credible | Neoclouds are becoming serious |
| Becoming the main cloud for ordinary companies | Still rare | AWS remains far ahead |
| Building a larger and more profitable cloud business | No neocloud is close | AWS remains dominant |
If you want more recent data on this point, please see our latest GPU cloud market report.
How big are neoclouds today?
Neoclouds are already a $25 billion-plus industry, although AWS remains about five times larger by annual revenue.
Synergy Research Group estimates that neocloud revenue exceeded $25 billion in 2025. The category reached approximately $9 billion in the final quarter alone, up 223% from one year earlier. That pace is exceptional even within the booming cloud market.
AWS generated $128.7 billion over the same year. It remains in a different weight class, but the gap is smaller than it was when neoclouds were mostly renting GPUs to startups.
The latest broader cloud data tells a similar story. Neoclouds currently account for around 5% of worldwide cloud infrastructure revenue. AWS holds approximately 28%. Put another way, the whole neocloud category is now equivalent to almost one-fifth of AWS by annual revenue and roughly one-sixth of its current market share.
The comparison also understates the neocloud position in AI. AWS earns money from storage, databases, ordinary servers and many other products. Neocloud revenue is much more concentrated in accelerated computing, so its share of the market for large AI clusters is considerably higher than 5%.
Synergy expects the neocloud market to approach $400 billion by 2031. Forecasts that far ahead deserve plenty of caution, especially in a market shaped by chip cycles and private contracts. The recent revenue is already large enough to show that neoclouds have moved beyond a temporary side market.
| Measure | Neoclouds | AWS | What the comparison shows |
|---|---|---|---|
| 2025 revenue | More than $25 billion | $128.7 billion | AWS was about five times larger |
| Latest global cloud share | About 5% | About 28% | AWS remains the clear overall leader |
| Recent quarterly growth | 223% in the final quarter of 2025 | 28% in the latest reported quarter | Neoclouds are growing much faster |
| Main revenue source | AI computing | Broad cloud services | Neoclouds are stronger than their total share suggests inside AI |
Is CoreWeave already a real AWS rival?
CoreWeave is already a real rival for giant AI deployments, but its size and finances still look nothing like AWS.
CoreWeave generated $2.08 billion in its latest reported quarter, more than double the $982 million recorded one year earlier. Annualizing that quarter gives roughly $8.3 billion of revenue.
AWS generated $37.6 billion in its latest quarter. Its annualized pace is therefore above $150 billion, around 18 times CoreWeave’s current pace.
Growth tells the opposite story. CoreWeave expanded by 112% year over year, while AWS grew by 28%. CoreWeave also ended the quarter with $99.4 billion of revenue backlog. That backlog is almost 12 times its current annualized revenue, showing how much capacity customers have reserved before it has been fully built.
The recent Jane Street agreement widens the customer story. The trading firm committed approximately $6 billion to CoreWeave services and invested another $1 billion in its shares. That is unusually large spending from a company outside the small group of frontier AI laboratories and consumer internet platforms.
The financial comparison is brutal. CoreWeave reported a $144 million operating loss, $536 million of net interest expense and a $740 million net loss during the quarter. AWS produced $14.2 billion of operating income.
CoreWeave can beat AWS in a procurement contest involving one enormous GPU cluster. It would struggle far more in a prolonged price war, a lending downturn or a period of weak infrastructure utilization.
If you want more recent data on this point, please see our latest GPU cloud market report.
Why are Microsoft, Meta, OpenAI and Anthropic using several AI clouds?
The biggest AI buyers currently spread work across several clouds because their computing needs are larger than any one provider’s construction pipeline.
Microsoft’s behavior makes the point clearly. The company operates Azure, one of the world’s largest clouds, but has still signed multibillion-dollar infrastructure agreements with companies including CoreWeave, Lambda and Nebius.
OpenAI also works across several providers. It has purchased substantial CoreWeave capacity while agreeing to consume approximately two gigawatts of AWS Trainium capacity. Anthropic has signed a multiyear CoreWeave agreement while using AWS as its main cloud and training partner.
Each provider gives the buyer a different combination of chips, locations, prices and delivery dates. AWS may offer Trainium and direct access to Bedrock. A neocloud may offer a dedicated NVIDIA cluster that becomes available sooner. Another provider may have secured power in a region where the buyer wants to expand.
The contracts also shift construction work away from the customer. The cloud provider finds the site, arranges electricity, finances the hardware, installs the network and operates the facility. The customer commits to using the output.
This multi-cloud buying pattern probably sticks. AI laboratories do not want one delayed electrical connection, chip shipment or construction project holding back an entire generation of models.
Are neoclouds really faster or cheaper than AWS?
Neoclouds can be faster and cheaper for one large, predictable GPU workload, but there is no universal neocloud discount.
Their speed advantage begins with focus. A neocloud can design a cluster around one accelerator generation, one network and a limited set of workloads. AWS has to fit new AI infrastructure into a much wider global platform.
Power has become especially important. Nebius recently arranged 328 megawatts of behind-the-meter fuel-cell capacity for a US site, partly to avoid waiting for conventional grid expansion. Other neoclouds are pairing construction projects with dedicated generation or long-term utility agreements. Securing electricity can now save more time than securing the chips.
The cost comparison is messier. Public hourly prices rarely reflect the private deals signed by major customers. Large buyers negotiate reserved capacity, minimum spending commitments, hardware upgrades and service guarantees over several years.
Utilization can easily outweigh the advertised GPU price. A cluster that costs 10% less per hour becomes expensive when jobs repeatedly fail, the network slows communication between machines or the customer cannot obtain enough neighboring GPUs. A more expensive cluster may produce a trained model sooner and reduce the total bill.
Neoclouds are strongest when a customer knows that it will use almost every accelerator for months or years. AWS becomes more attractive when demand changes quickly, when the company needs several chip choices or when its data and applications already sit inside the AWS ecosystem.
Will neocloud growth survive when GPUs are easier to get?
The strongest neoclouds should survive easier GPU supply, but companies that mainly resell scarce chips will lose pricing power.
The early boom benefited heavily from shortage. Demand for NVIDIA accelerators rose faster than chip production, power connections and hyperscaler data centers. A provider that obtained a large block of GPUs could charge attractive prices without offering a particularly deep software platform.
Supply is still tight. Synergy Research Group says AI demand continues to exceed the capacity available from traditional clouds. Neoclouds are signing contracts several years ahead, and new sites are being planned at gigawatt scale.
The market will eventually become less forgiving. More chips will ship, hyperscalers will complete new campuses, and each hardware generation will deliver more computing power per dollar. Customers will have more alternatives when renewing contracts.
CoreWeave and Nebius are preparing for that change. They have added tools for scheduling, model development, inference, monitoring and workload management. Nebius is also allowing infrastructure partners to run its cloud stack in facilities financed and owned by those partners.
The providers that keep customers through better performance and software can remain useful after the shortage eases. The weaker ones may discover that access to GPUs was their entire advantage.
If you want more recent data on this point, please see our latest GPU cloud market report.
Can neoclouds win AI inference, not just model training?
Neoclouds are already moving into AI inference, and their early performance shows that they can compete for large production workloads.
Training made neoclouds famous because one job can occupy thousands of chips at once. Inference produces smaller requests, but those requests may arrive continuously from millions of users. A popular AI product can generate more stable infrastructure demand than a series of occasional training runs.
CoreWeave recently ranked first for both output speed and price-performance when Artificial Analysis tested 11 providers running Moonshot AI’s Kimi K2.6 model. One benchmark cannot settle the whole market, but it shows that a neocloud can compete on delivered tokens rather than merely on the number of GPUs installed.
The company has also introduced dedicated inference capacity that lets customers choose specific chips and runtimes. Its agreements with Anthropic, Perplexity and other AI companies cover production use as well as model development.
AWS has a powerful answer through Bedrock. Amazon said customer spending on Bedrock rose 170% from the previous quarter, while the service processed more tokens during that quarter than during all previous years combined. Most companies using Bedrock can connect a model directly to data, security tools and applications already hosted on AWS.
Neoclouds have the clearest opening with large applications that can reserve a dedicated fleet. AWS is better placed to serve the much larger number of companies adding smaller amounts of AI to existing software.
Why would a normal company choose a neocloud?
Most companies will choose a neocloud for one expensive AI project rather than move their whole technology stack.
Jane Street is a useful example. It trains complex models on huge volumes of market data, giving it a workload large enough to justify a $6 billion CoreWeave commitment. It does not need CoreWeave to replace every database, website or internal business application used by the firm.
AI startups have an even clearer reason. Their main product may depend directly on model training or inference. Getting better cluster performance can determine how quickly they launch, how much they charge and whether the product works at all.
The case becomes weaker for a retailer, manufacturer or insurer running a few AI applications. Such companies already have cloud contracts, security rules, trained employees and data stored with AWS, Microsoft or Google. Adding an approved AI service inside the same environment is usually easier than introducing a new infrastructure provider.
Neoclouds can still enter through a demanding project and expand later. CoreWeave has added Weights & Biases, orchestration tools and production inference services partly to increase the amount of work customers can do on its platform.
That route will produce enterprise growth, although it looks more like gradual expansion from AI workloads than a mass migration away from AWS.
Can neoclouds match AWS outside a few giant data centers?
AWS still has an overwhelming advantage in global coverage, disaster recovery and regulatory support.
AWS currently operates 123 Availability Zones across 39 geographic regions. A multinational company can place systems near customers, keep data inside particular countries and duplicate critical applications across separate facilities.
CoreWeave remains far more concentrated. Its latest filing showed that around 91% of quarterly revenue came from the United States and 88% of long-lived assets were located there. It has expanded into Europe, but its international footprint is still young.
Nebius is pushing harder outside the United States. It recently committed approximately £1.7 billion to four UK deployments that should reach 65 megawatts when fully ramped. It also operates sites elsewhere in Europe and is inviting regional partners to deploy the Nebius platform inside their own data centers.
Large AI laboratories may accept geographic concentration when they need one enormous training cluster. A bank serving customers in 20 countries faces a different problem. It may need local data storage, several backup regions, specific certifications and an existing legal agreement covering every market.
Neoclouds can build denser AI facilities than AWS in selected locations. Matching AWS across countries and regulated industries will take much longer.
What happens if a big neocloud customer cuts its order?
One customer changing course could still hurt a neocloud far more than it would hurt AWS.
CoreWeave’s two largest customers produced approximately 65% of its latest quarterly revenue. The largest accounted for 45%, while the second represented 20%. One year earlier, a single customer generated 72%.
The concentration helps CoreWeave grow quickly. A long contract with a financially strong customer supports billions of dollars of debt and gives the company confidence that a new facility will be used.
It also gives a small group of buyers considerable influence. Microsoft, Meta and major AI laboratories can spread future work across several providers, build more infrastructure themselves or shift workloads toward custom chips.
Committed contracts reduce the risk of empty data centers, but they create delivery obligations. CoreWeave must provide the agreed capacity, meet performance requirements and complete facilities on schedule before all the booked revenue can be recognized.
Its latest filing showed that only 36% of remaining performance obligations were expected to become revenue during the following 24 months. Much of the promised business therefore depends on infrastructure that will operate further into the future.
The recent addition of customers such as Jane Street improves the mix. CoreWeave still needs many more large independent customers before one buyer’s strategy stops shaping the whole company.
Can neoclouds afford the AI infrastructure buildout?
Neoclouds can currently finance their expansion, although the funding is expensive and usually tied to specific customer contracts.
CoreWeave spent $536 million on net interest during its latest quarter, equal to roughly one-quarter of revenue. Its reported current and long-term debt was close to $25 billion, before counting around $10 billion of operating lease liabilities.
Some of its newer financing is cheaper and safer than its early borrowing. CoreWeave secured an $8.5 billion facility backed by GPUs and contracted cash flows, with parts priced at SOFR plus 2.25% and other parts at roughly 5.9%.
Other borrowing remains costly. The company issued $2.75 billion of senior unsecured notes carrying a 9.75% interest rate. That coupon shows how much investors still charge for capital that lacks the strongest project-level protections.
Nebius recently raised $775 million through its first senior secured debt facility. The loan is priced at SOFR plus 2.5% and is backed by deployed GPUs and contracted payments from an investment-grade customer. Nebius says the contract and loan together cover more than 100% of the related capital expenditure.
Nebius is also testing a lighter model. Infrastructure partners finance and own the buildings and hardware, while Nebius supplies the architecture, software, customers and cloud operations. That lowers Nebius’s spending but leaves it dependent on partners delivering consistent facilities.
AWS can fund investment from Amazon’s much larger cash engine. Amazon generated $148.5 billion of operating cash flow over the latest 12 months. Even so, its free cash flow fell to $1.2 billion after an additional $59.3 billion of property and equipment spending, mainly for AI.
| Provider | Recent financing evidence | What it tells us |
|---|---|---|
| CoreWeave | $8.5 billion contract-backed facility | Strong contracts can unlock cheaper project debt |
| CoreWeave | $2.75 billion of notes at 9.75% | General corporate borrowing remains expensive |
| Nebius | $775 million facility at SOFR plus 2.5% | Delivered capacity can be refinanced against customer payments |
| Nebius | Partner-owned infrastructure model | Growth can continue with less company capital |
| Amazon and AWS | $148.5 billion of operating cash flow | AWS can finance expansion from a much larger business |
Does NVIDIA give neoclouds an edge or own their edge?
NVIDIA gives neoclouds credibility, early hardware access and technical support while keeping a large part of their advantage under NVIDIA’s control.
NVIDIA has invested $2 billion in CoreWeave and another $2 billion in Nebius. Both companies are working with NVIDIA on plans that could exceed five gigawatts of capacity by 2030.
That adds up to at least $4 billion invested in two specialist cloud providers. NVIDIA is effectively helping finance extra distribution channels for its chips, reducing its reliance on AWS, Microsoft and Google.
The neoclouds gain more than money. Close relationships can bring earlier access to new systems, help with facility design and provide engineering support when new hardware is installed. NVIDIA’s involvement also reassures customers and lenders.
The catch is obvious. CoreWeave and Nebius do not control the chip roadmap, the wholesale hardware price or CUDA, the software ecosystem used by most AI developers. Competing clouds can buy the same generation of systems.
A new chip can also weaken the economics of equipment installed only a few years earlier. Long contracts help protect revenue, but customers will eventually want the latest hardware or a lower price.
Neoclouds need software, operations and customer service to carry more of the value over time. Hardware access gave them their opening; it cannot remain their whole defense.
If you want more recent data on this point, please see our latest GPU cloud market report.
Can AWS Trainium undercut NVIDIA-based neoclouds?
Trainium is AWS’s strongest weapon against neoclouds because Amazon controls the chip, the cloud and the services running around it.
Amazon says its custom-chip activities, including Trainium, Graviton and Nitro, have passed a $20 billion annual revenue run rate and are growing at a triple-digit pace. That business is already close in size to the recent annual revenue of the entire neocloud category.
AWS has also moved beyond small deployments. Project Rainier was built with nearly 500,000 Trainium2 chips for Anthropic. OpenAI has agreed to consume around two gigawatts of Trainium capacity for future workloads.
Amazon claims that customers can reduce some training and inference costs by as much as 50% with newer Trainium systems. Those are vendor figures, and results will vary by model, but the economic logic is solid. AWS can avoid paying NVIDIA’s full margin on every accelerator it rents to customers.
Neoclouds still benefit from NVIDIA’s flexibility. Most AI teams already use CUDA, and models often run on NVIDIA hardware before they are optimized for alternative chips. Moving a large workload to Trainium can require engineering time and new software.
AWS does not need every customer to switch. Moving a meaningful share of recurring training and inference onto its own chips would lower its costs and shrink the market where NVIDIA-focused providers have the clearest advantage.
Will most neoclouds survive?
A few neoclouds can become major infrastructure companies, while many smaller providers will be bought, pushed into narrow niches or disappear.
The market can support more than AWS, Azure and Google Cloud. Synergy Research Group already places five neocloud companies among the world’s 30 largest cloud providers.
It cannot support unlimited providers with identical offers. Each serious competitor needs advanced chips, scarce power, high-speed networking, experienced engineers and billions of dollars of financing. One delayed facility or lost customer can damage a smaller company quickly.
The larger neoclouds are already choosing different paths. CoreWeave is building a broad AI software and infrastructure platform. Nebius combines owned sites with partner-financed capacity. Crusoe focuses heavily on developing large AI campuses. Fluidstack sells custom infrastructure to major AI laboratories.
Some companies may end up operating the cloud software while infrastructure investors own the physical assets. Others may specialize in one country, industry, chip architecture or type of workload.
The category should grow even as individual names disappear. Three or four global neoclouds, several regional specialists and a long list of infrastructure partners looks more plausible than dozens of independent AWS challengers.
Is AI compute becoming its own cloud market?
AI compute is separating into its own cloud market, giving neoclouds room to grow without taking over everything AWS does.
An ordinary cloud server may run a website, database or internal business application. A modern AI cluster combines thousands of accelerators, liquid cooling, enormous electrical loads and networks designed to move data between machines almost instantly.
Those requirements change where the infrastructure is built and how it is sold. A company may reserve an entire cluster for several years instead of requesting a few virtual machines whenever traffic rises.
The software architecture is also becoming more modular. A company can train a model on CoreWeave, keep its business data on AWS, access commercial models through Bedrock and run selected inference workloads in its own facility.
That setup may look messy from the outside, but customers already combine several clouds, software vendors and private systems. AI adds another specialized layer.
AWS can remain the main home for data and business applications while losing some large training and inference jobs. Neoclouds can become important infrastructure providers without recreating every AWS product.
Can neoclouds really beat AWS?
Yes, neoclouds can beat AWS for dedicated AI compute. They cannot currently beat AWS as the world’s default cloud.
The narrow victory is already happening. Neoclouds have won contracts measured in billions of dollars, built clusters for leading AI companies and grown into a meaningful share of cloud infrastructure revenue. Their focus gives them an advantage when a customer wants a huge amount of one chip, in one place, on a specific schedule.
AWS remains stronger almost everywhere else. It has far more revenue, large operating profits, global coverage, existing enterprise customers and its own chips. It can sell AI computing together with the data, security and software that companies already use.
The financial difference is especially important. Neoclouds must borrow heavily before new facilities produce revenue. AWS can finance expansion from one of the world’s largest cash-generating companies. A weaker chip cycle or delayed project would therefore hit the specialists harder.
We expect several neoclouds to become large and lasting businesses. They can capture a substantial part of frontier-model training, dedicated inference and specialized AI infrastructure. Their presence will also push AWS to deploy capacity faster, improve performance and offer better prices.
The broader claim goes too far. Neoclouds are building a valuable new layer of the cloud market rather than inheriting AWS’s position. They will beat AWS contract by contract and workload by workload, while AWS remains the larger, broader and safer platform.
So, yes—but only at the AI-factory layer. Neoclouds can win that battle while AWS continues to win the wider cloud war.
If you want more recent data on this point, please see our latest GPU cloud market report.
OUR METHODOLOGY
This analysis tests whether neoclouds can meaningfully beat AWS based on the evidence available today. We separate winning a frontier-model training contract, delivering competitive inference and replacing AWS as a company’s primary cloud because those are fundamentally different achievements.
We assessed the competitive dimensions that most directly reveal strength in this market: commercial traction, recognized revenue, growth, revenue backlog, infrastructure capacity, workload performance, product depth, geographic reach, customer concentration, financing and control of the underlying technology stack.
For each dimension, we used the freshest meaningful evidence available, including reported financial results, signed customer commitments, deployed and contracted power, infrastructure financing, customer diversification, independent performance testing and newly launched services. We gave more weight to actual spending, deployment and delivered performance than to general ambitions or projected capacity.
We kept different kinds of evidence in their proper place. Recognized revenue shows the business operating today; backlog and long-term contracts show demand already reserved; announced infrastructure indicates where future capacity may emerge; and benchmarks measure particular workloads under particular conditions. No single contract, forecast or benchmark settles the broader question.
We also compared neocloud economics with AWS’s broader position. This includes AWS revenue and operating income, Amazon’s cash generation, AWS’s international footprint, its enterprise software ecosystem and its ability to move more workloads onto proprietary Trainium hardware.
Our conclusion comes from signals that line up across several dimensions rather than from one eye-catching deal. This lets us distinguish where neoclouds already have an advantage, where they are becoming credible challengers and where AWS retains strengths that would take many years to reproduce.
Key sources used for this analysis include Synergy Research Group’s neocloud market analysis, Amazon’s full-year 2025 results, Amazon’s first-quarter 2026 results, CoreWeave’s first-quarter 2026 results, CoreWeave’s expanded Meta agreement, Nebius’s Meta agreement, CoreWeave’s Jane Street agreement, OpenAI and Amazon’s AWS and Trainium partnership, Artificial Analysis’s Kimi K2.6 provider benchmark, AWS’s global infrastructure data, NVIDIA’s CoreWeave investment and expansion announcement, NVIDIA and Nebius’s full-stack cloud partnership, and CoreWeave’s $8.5 billion infrastructure financing announcement.
