By Javier Surasky
.png)
Anyone who follows artificial intelligence with some attention has probably heard of Nvidia, OpenAI, Amazon, Microsoft, or Google, but is less likely to know CoreWeave. And yet, some of the most important actors in the sector buy from this company a resource without which their models cannot be trained or operate at scale: compute capacity.
CoreWeave
began operating in 2017 in the cryptocurrency mining business, but later
converted its infrastructure to high-performance computing, and the rise of
generative AI ultimately consolidated this new line of business.
The company
found an opportunity in a problem that was becoming increasingly serious for
artificial intelligence developers: obtaining thousands of specialized
processors, installing them in data centers capable of powering and cooling
them, connecting them to one another, and having them available for as long as
necessary. CoreWeave specialized in providing that service.
Its growth
was very rapid, and in the second quarter of 2026 it reported revenues of USD
2.575 billion, 112% more than a year earlier, along with contractual
commitments that could generate approximately USD 104 billion in future
revenues (CoreWeave, 2026a).
A New Actor Between Chips and Models
CoreWeave
is usually classified as a neocloud, a term used for cloud infrastructure providers specializing mainly in GPUs and other resources
intended for artificial intelligence workloads. These companies grew in
response to the sharp increase in demand and chipmakers’ interest in
diversifying the channels through which their products reached the market,
opening up a space between hardware manufacturers, major cloud providers, and
those who needed to run AI models (Mazza et al., 2025).
CoreWeave’s
position is easier to understand if we turn to Hawkins, Lehdonvirta, and Wu
(2025) and their distinction among “at least three” dimensions: the physical
location of compute, control over data centers, and control over the
accelerators operating inside them.
CoreWeave
operates in the spaces connecting these layers by obtaining processors,
securing physical infrastructure and energy, organizing those resources in data
centers, and offering compute capacity as its product.
That
position allows it to be an influential actor, but it also creates strong
external dependencies for the company, with access to hardware being the most
important.
CoreWeave
relies primarily on Nvidia accelerators and needs access to new generations of
its GPUs to remain competitive. But the relationship between the two companies
is not simply one of buying and selling processors, because the growth of neoclouds
benefits accelerator manufacturers by developing a new market of users for
their products: Mazza et al. (2025) note that manufacturers have contributed to
the sector’s development through processor allocations, financing, and
commercial commitments.
The
relationship between CoreWeave and Nvidia is particularly close, to the point
that Nvidia is CoreWeave’s technology supplier, investor, and strategic partner
(CoreWeave & NVIDIA, 2026), within an asymmetric relationship. CoreWeave
could not maintain its current position without continued access to Nvidia
products, while for Nvidia, CoreWeave is one among several major buyers of its
accelerators.
That
asymmetry has not prevented each company from benefiting from the growth of the
other.
When Customers Also Constrain
If the
relationship with Nvidia can be understood as an “upstream dependence,” there
is another that we should consider a “downstream dependence”: CoreWeave’s
dependence on customers seeking to increase their compute capacity.
CoreWeave
has secured large-scale contracts with model developers and technology
companies, including OpenAI and Microsoft, but to develop and grow it needs
long-term contracts, because infrastructure using thousands of GPUs requires
substantial investment before it begins generating revenue, and the most direct
way to address investment risk is to have customers commit to future capacity.
Along with
that need comes another form of dependence and concentration, because some neoclouds
receive a very high proportion of their revenues from one or two buyers (Mazza
et al., 2025), a situation that becomes more complex when those same buyers
have their own cloud infrastructure.
Microsoft,
for example, can purchase capacity from CoreWeave while developing Azure, but
what will happen later? The fact is that the relationship between companies
that support one another temporarily while competing within the same sector
does not fit neatly into the traditional division between competitors and
partners.
And one of
CoreWeave’s most important relationships requires a prior clarification to
avoid confusion: Core Scientific is independent from CoreWeave, and the two
companies do not belong to the same corporate group. Their relationship,
however, is crucial for both: Core Scientific develops and operates
infrastructure that CoreWeave uses through long-term contracts, and as a
result, part of the capacity CoreWeave offers its customers depends on physical
facilities controlled by Core Scientific. The relationship between the two is
so close that, in July 2025, CoreWeave agreed to acquire Core Scientific
through an all-stock transaction that ultimately failed to receive approval
from Core Scientific’s shareholders and was abandoned (Core Scientific, 2025).
This
episode helps us understand that CoreWeave can control the provision of compute
services without owning the buildings, land, and electrical connections that
support them, as long as it has contracts securing long-term access to
infrastructure managed by third parties, which constitutes another “exposed
side” of the business.
The Cloud Has a Street Address
Talking
about cloud computing should not lead us away from the territorial
dimension of compute: as Sachin Jain, CoreWeave’s Chief Operating Officer, put
it in July 2026, data centers depend on land, energy, water, roads, public
institutions, and the trust of the communities where they operate (Jain, 2026).
This brings into focus the increasingly sensitive issue of local acceptance of
data centers, whose expansion is indispensable if CoreWeave is to have access
to the assets its business model requires.
The
political dimension of this infrastructure has begun to be explicitly
recognized following several protests by residents over the location of data
centers near residential areas, and the European Commission (2026), for
example, already links access to compute infrastructure with issues of economic
security, sovereignty, resilience, and competitiveness.
Growing with Borrowed Money
No less
important is capital: buying processors, contracting data centers, and securing
energy supply require major investments, which helps explain why CoreWeave’s
rapid expansion has been sustained to a large extent through debt.
It is one
case within a broader transformation that has been taking place for some time
in the financing of AI infrastructure: Aldasoro, Doerr, and Rees (2026) argue
that the expected volume of investment is moving the sector away from a model
financed mainly through operating cash flows toward growing dependence on debt
and private credit.
In a
company such as CoreWeave, contracts committing customers to purchase capacity
create expectations of future revenue, which in turn facilitate access to
capital to acquire processors and reserve new infrastructure on the assumption
that the returns generated by those facilities will be sufficient to cover the
debt. This mechanism introduces risks, and its sustainability depends on
continued demand as much as on predictable financial conditions and the
economic performance of technological assets that, given the pace of technological
change in AI, can depreciate rapidly (Aldasoro et al., 2026).
CoreWeave
also faces another problem as the market matures: providing GPU-based capacity
may offer few opportunities to differentiate itself from competitors as supply
increases, and here a paradox emerges, because while large contracts guarantee
utilization, they also increase dependence on a small number of customers that
can replace one provider with another.
To respond
to this problem, some neoclouds are incorporating software services and
tools designed to support other stages in the development and operation of AI
systems (Mazza et al., 2025), but this only adds further complications, because
the more they expand their services, the closer they move to activities also
offered by their own customers, who begin to see them as direct competitors.
Where Does CoreWeave’s Power Lie?
The path we
have followed is enough to identify that the power of this actor in the AI
ecosystem lies in its ability to bring together resources distributed among
different actors and turn them into available compute.
This
creates a situation in which the company depends on actors positioned on both
sides of its activity and can, at the same time, become a point on which others
depend to obtain compute capacity.
Infrastructure
studies offer a useful way to think about this situation: infrastructures gain
importance when other activities begin to organize themselves around them and
when their failures or limitations produce effects that extend beyond the
organization that manages them (Plantin et al., 2018). CoreWeave is building
such a position, but it is not certain that it will be able to preserve it in
the face of much larger competitors, rapid technological change, and growing
financial needs.
Still, its
trajectory makes visible a part of AI power that often remains hidden behind
the models: before a system can be trained or used, someone has to bring
together processors, buildings, electrical grids, capital, and contracts in a
coherent way and adapt them to the needs of major developers. CoreWeave has
turned that condition into a business.
Key Facts
- Its headquarters are in New Jersey, United States.
- Its main activity is providing cloud infrastructure specialized in AI workloads, within the neocloud sector.
- CoreWeave’s founders are Michael Intrator, currently the company’s CEO, Brian Venturo, Brannin McBee, and Peter Salanki. The first three had professional backgrounds in energy, emissions, and commodity markets. Peter Salanki, by contrast, developed his career in the technology field (CoreWeave, n.d.).
- The company’s key business relationships involve Nvidia, Core Scientific, OpenAI, and Microsoft.
- In the second quarter of 2026, its three largest customers accounted for approximately 72% of quarterly revenue.
- OpenAI committed to paying up to approximately USD 6.5 billion for cloud computing capacity through May 31, 2031, while Meta made a commitment of up to approximately USD 21 billion for capacity through 2032 (CoreWeave, 2026b).
- Nvidia, from which CoreWeave purchases advanced chips, appears in an infrastructure financing transaction disclosed in 2026 as being obligated to purchase up to USD 6.3 billion in residual CoreWeave cloud capacity through April 2032 (Galaxy Digital Inc. et al., 2026).
- The company is listed on Nasdaq under the ticker CRWV, where it had a market capitalization of approximately USD 49.4 billion as of October 2, 2026 (CompaniesMarketCap, 2026).
References
Aldasoro,
I., Doerr, S., & Rees, D. (2026, 7 de enero). Financing the AI boom:
From cash flows to debt (BIS Bulletin No. 120). Bank for International
Settlements. https://www.bis.org/publ/bisbull120.pdf
CompaniesMarketCap.
(2026). CoreWeave (CRWV) - Market capitalization. Recuperado el 4
de octubre de 2026 de https://companiesmarketcap.com/coreweave/marketcap/
Core Scientific,
Inc. (2025, 30 de octubre). Core Scientific announces termination of merger agreement with CoreWeave [Comunicado de prensa presentado
como Exhibit 99.1]. U.S. Securities and Exchange Commission. https://www.sec.gov/Archives/edgar/data/1839341/000114036125039808/ef20057995_ex99-1.htm
CoreWeave,
Inc. (s. f.). Executive management. https://investors.coreweave.com/governance/executive-management/default.aspx
CoreWeave, Inc. (2026a, 11 de agosto). CoreWeave
reports strong second quarter 2026 results. https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Second-Quarter-2026-Results/default.aspx
CoreWeave,
Inc. (2026b). Quarterly report for the quarter ended June 30, 2026 (Form
10-Q). U.S. Securities and Exchange Commission. https://www.sec.gov/Archives/edgar/data/1769628/000176962826000366/crwv-20260630.htm
CoreWeave,
Inc., & NVIDIA Corporation. (2026, 26 de enero). NVIDIA and CoreWeave
strengthen collaboration to accelerate buildout of AI factories. https://investors.coreweave.com/news/news-details/2026/NVIDIA-and-CoreWeave-Strengthen-Collaboration-to-Accelerate-Buildout-of-AI-Factories/default.aspx
European
Commission. (2026, 3 de junio). Proposal for the Cloud and AI Development
Act (CADA) (COM(2026) 502 final). https://digital-strategy.ec.europa.eu/en/library/proposal-cloud-and-ai-development-act-cada
Galaxy
Digital Inc., Galaxy Helios Data Centers II LLC, & Galaxy Helios II LLC.
(2026, julio). Investor presentation [Exhibit 99.1]. U.S. Securities and
Exchange Commission. https://www.sec.gov/Archives/edgar/data/1859392/000185939226000068/launch8-kexhibit99172226.htm
Hawkins, Z. J., Lehdonvirta, V., & Wu, B. (2025, 20 de
junio). AI compute
sovereignty: Infrastructure control across territories, cloud providers, and
accelerators. SSRN.
https://doi.org/10.2139/ssrn.5312977
Jain, S.
(2026, 23 de julio). A data center should make its community stronger.
CoreWeave. https://www.coreweave.com/blog/a-data-center-should-make-its-community-stronger
Mazza, M., Sachdeva, P., Arutyunyan, S., & Alatovic, T.
(2025, 19 de noviembre). The
evolution of neoclouds and their next moves. McKinsey & Company. https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/the-evolution-of-neoclouds-and-their-next-moves
Plantin,
J.-C., Lagoze, C., Edwards, P. N., & Sandvig, C. (2018). Infrastructure
studies meet platform studies in the age of Google and Facebook. New
Media & Society, 20(1), 293–310. https://doi.org/10.1177/1461444816661553