Nvidia

Nvidia continues to solidify its position in AI data center infrastructure, with its hardware remaining essential for advanced computing. The company is exploring innovative solutions like orbital AI data centers and advanced liquid-cooled systems to meet escalating demand. However, supply chain pressures and operational risks are significant, prompting strategic investments and partnerships to manage infrastructure control and grid capacity challenges.

The company's focus is increasingly on AI inference economics and system-level design, prioritizing compute productivity and specialized applications. This includes expanding its software ecosystem and developing specialized CPUs for agentic AI workloads. Addressing thermal constraints with advanced cooling and exploring AI tools for industrial applications are key priorities as AI agents evolve.

Nvidia is proposing 'cost per token' as a critical metric for evaluating AI infrastructure, potentially influencing enterprise adoption and hardware assessment. Strategic investments in third-party infrastructure providers, alongside in-house data centers, indicate a pivot driven by GPU availability, grid constraints, and inference workload economics. The company is also building new manufacturing plants to bolster data center support and exploring data center integration in new home constructions.

Last updated May 10, 2026

Coverage

Key vendors in the AI server market, including Dell, HPE, Lenovo, and Supermicro, are experiencing substantial demand, though securing enterprise clients necessitates a strategic approach beyond merely supplying Nvidia chips.
NVIDIA's $2.1 billion warrant in IREN, purchased at a 23 percent strike premium, signifies a strategic move towards equity verticalization and a potential new operating standard for the neocould channel, bypassing hyperscalers for capital reallocation.
NVIDIA has secured a five-year cloud services contract with IREN, valued at $3.4 billion, and also obtained a $2.1 billion equity warrant.
The primary constraint on AI infrastructure expansion is no longer funding but the rapid acquisition and power enablement of sites to convert them into operational AI capacity.
The deepening partnership between Corning and Nvidia signifies Nvidia's increased involvement in the physical infrastructure supporting artificial intelligence, optical networking, and hyperscale deployments.
TotalEnergies has partnered with Nvidia and Dell to construct a supercomputer in Pau, France, which will be utilized to enhance seismic imaging for hydrocarbon exploration.
Nvidia and IREN are collaborating on a significant AI infrastructure deployment, targeting up to 5 gigawatts, with their Sweetwater, Texas campus serving as a key site for Nvidia's DSX AI factory architecture.
Nvidia is set to construct three new manufacturing plants across the U.S. to bolster support for data center infrastructure, signaling a strategic investment to meet growing demand in the digital infrastructure sector.
Nvidia and PulteGroup are collaborating with a startup to pilot data centers that will be integrated into new home constructions.
An attendee of the Microsoft AI Tour in Zurich observes a rebranding of LLMs to 'Agents' with minimal practical advancement, highlighting a significant disconnect between industry hype and enterprise reality, particularly concerning the 'editing tax' of AI-generated content and user adoption drop-off.
The shift in artificial intelligence adoption from model training to serving inferences presents AI chip startups with a critical opportunity to establish themselves in a market where Nvidia acts as both a potential collaborator and competitor.
Invenergy, Nvidia, and Emerald AI are collaborating to develop flexible artificial intelligence factories, capable of operating from the edge to multi-gigawatt campuses.
Celestica has launched new network switches featuring 64 ports of 1.6 terabits per second Ethernet, designed to meet the demands of high-bandwidth applications and coinciding with the release of Nvidia's 1.6 Tbps ConnectX-9 NICs.
The infrastructure shift driven by dual graphical processing units has priced out many investors, signaling a critical constraint in GPU density and highlighting how early movers have benefited from five years of misallocated capital.
Anthropic's $100 billion AWS commitment, coupled with a 10-year spend tenor and a 5GW ceiling, signals significant anchor tenant economics and competition in emerging market inference corridors, validating Trainium and Project Rainier expansion.
This week in data centers highlights that capital investment is outpacing grid capacity and regulatory approvals, making coordinated delivery across energy, permitting, and tenant demand essential for scaling AI infrastructure.
Nvidia suggests that the evolving nature of AI agents, which operate across extended periods and multiple systems, is making infrastructure, rather than just the models themselves, the primary limiting factor for data center throughput.
Intel is strategically focusing on AI inference to boost its CPU relevance, aiming to integrate AI into agents, robots, and edge devices, despite facing persistent chip manufacturing challenges.
Nvidia proposes evaluating AI infrastructure based on cost per token rather than traditional compute metrics, though analysts suggest this approach may primarily benefit hyperscale environments and might be premature for enterprise IT adoption.
Nvidia has released open-source Ising artificial intelligence models designed to aid in quantum chip development by providing tools for calibration and decoding to address quantum error correction.
Meta's substantial spending on CoreWeave, including take-or-pay GPU supply agreements and priority access to NVIDIA hardware, indicates a strategic shift away from its own US data centers due to grid constraints, hyperscaler self-build timelines, and inference workload economics.
RISC-V chip designer SiFive has successfully closed an oversubscribed Series G funding round, raising $400 million with participation from Nvidia, valuing the startup at $3.65 billion.
Industry watchers warn that Nvidia's next-generation Rubin GPUs and China-bound Hopper accelerators may face delays and reduced shipment volumes due to supply chain issues and technical challenges.
CoreWeave's announcements at NVIDIA GTC 2026 and Bell Canada's 300 MW development in Saskatchewan signify a shift towards integrated AI infrastructure, where power, platforms, and sovereign capacity are key determinants of future scale, moving beyond simple GPU access.
Supermicro has initiated an independent investigation after three individuals associated with the company were charged with violating US export restrictions on China, specifically concerning the diversion of Nvidia GPU servers.
CoreWeave's potential $8.5 billion deal signifies a pivotal moment for GPU assets, as investment-grade debt, hyperscaler contracts, and power constraints are recasting AI infrastructure into institutional-grade assets.
Nscale is pursuing a power-centric artificial intelligence infrastructure strategy, integrating data centers, GPU clusters, and energy development following its AIPCorp acquisition and Microsoft collaboration, with plans to deploy NVIDIA Vera Rubin systems across its expanding campus portfolio in the U.S. and Europe.
Oracle is reportedly close to securing $16 billion in financing for its Stargate development in Michigan, while Microsoft is committing $5.5 billion to its AI and cloud infrastructure in Singapore by 2029.
Microsoft plans a significant $5.5 billion investment in artificial intelligence and cloud services in Singapore by 2029, building upon its existing presence in the region since 2010.
Schneider Electric executives Marc Garner and Jim Simonelli discussed at NVIDIA GTC 2026 how artificial intelligence infrastructure is necessitating a comprehensive redesign of data centers, encompassing areas like digital twins, liquid cooling, grid integration, onsite power, storage, and the anticipated shift towards higher-density electrical architectures.
Mistral has secured $830 million in debt financing to establish an artificial intelligence hub in Europe powered by Nvidia technology, enhancing the region's efforts to develop independent AI infrastructure and decrease reliance on external cloud services.
Nebius's ambitious 310-megawatt data center project, alongside agreements with Meta and Nvidia, highlights a significant shift in artificial intelligence compute from on-demand cloud services to reserved capacity.
South Korean AI chip startup Rebellions is preparing for an IPO and expanding internationally, positioning itself as a challenger to dominant GPU manufacturers like Nvidia and AMD in the AI infrastructure market.
NVIDIA's hardware is being distributed through strategic partnerships with Akamai, Comcast, AT&T, and T-Mobile, expanding the footprint for edge inference capabilities within the new artificial intelligence economy.
Crusoe is expanding its 900-megawatt data center campus in Abilene, Texas, to support Microsoft's artificial intelligence ambitions, with the new campus including on-site power generation.
US authorities have charged three additional individuals suspected of involvement in schemes using Thai front companies to illegally reroute high-end Nvidia graphics processing units intended for artificial intelligence servers to China.
Microsoft and Nvidia are collaborating to leverage artificial intelligence tools to streamline the permitting, planning, and design processes, aiming to accelerate the approval timeline for new nuclear power plants.
opinion
Raz Elad, the founder and chief executive officer of Israeli startup NextSilicon, offers commentary on the potential for his firm to compete against established industry leader Nvidia in the next generation of silicon development.
Akash Systems, in collaboration with AMD and Nvidia, is pioneering diamond-based cooling solutions to address the thermal challenges hindering the scalability of artificial intelligence in data centers.
Vertiv is enhancing its thermal management offerings for AI infrastructure through the acquisition of ThermoKey, aiming to address critical cooling bottlenecks and strengthen its position in the AI hardware market.
Jensen Huang's GTC 2026 keynote outlined how AI factories, inference economics, and system-level design are reshaping data center infrastructure, shifting value towards compute productivity rather than just AI models.
NVIDIA and Emerald AI are collaborating on flexible AI factories, with support from six major utilities to integrate AI software for managing power during grid peaks.
An indictment against Super Micro concerning the smuggling of Nvidia chips exposes the mounting supply chain risks within the artificial intelligence infrastructure sector, stemming from tensions between high demand, export controls, and vendor trust.
A Super Micro indictment related to the smuggling of Nvidia chips underscores escalating risks within the artificial intelligence infrastructure supply chain, driven by high demand and evolving export control regulations.
A co-founder of Supermicro has been indicted along with two others for allegedly evading United States export controls by illicitly shipping servers equipped with Nvidia graphics processing units, valued at $2.5 billion, to customers in China using fraudulent documentation.
During the recent GTC conference, Nvidia CEO Jensen Huang finally explained the strategic rationale for licensing technology from artificial intelligence chip startup Groq and hiring its engineering talent rather than developing similar capabilities internally.
Following his keynote at GTC 2026, Jensen Huang described artificial intelligence infrastructure as a comprehensive industrial system where inference, token economics, and synchronized data center construction will dictate future expansion.
Backed by Nvidia, the startup Reflection AI is planning a multi-billion dollar data center in South Korea as part of a broader US initiative to promote open artificial intelligence infrastructure to counter rivals in China.
NVIDIA is positioning itself for an agent-driven future with new products like the Groq 3 LPX rack and NemoClaw, focusing on the inference inflection point in AI.
Nvidia CEO Jensen Huang announced that the company is resuming the manufacturing of its older H200 graphics processing units to fulfill sustained demand from China, suggesting Beijing may have temporarily relaxed its previous directives favoring locally produced chips.