The Morning Download: Nvidia-CoreWeave AI Deal Reflects Surging Demand

 Here's what's up for Jan. 27, 2026: Microsoft Unveils New AI Chip • TikTok's Rocky U.S. Start • Social Media Under Fire in Landmark Trial.

Good morning. Nvidia’s decision to invest an additional $2 billion in AI cloud upstart CoreWeave reflects massive business demand for computing capacity.

CoreWeave co-founder and CEO Michael Intrator said on Monday that “this expanded collaboration underscores the strength of demand we are seeing across our customer base and the broader market signals as AI systems move into large-scale production.”

From the WSJ story:

• The companies said Monday that they would build new AI factories on Nvidia’s computing-platform technology, operated by CoreWeave, with a goal of 5 gigawatts of capacity by 2030. Nvidia said it would use its financial strength to help procure land and power for the projects …
• The new factories will test and validate CoreWeave’s software and reference architecture, including CoreWeave Mission Control, the companies said in their announcement. The project would also work toward deeper interoperability between Nvidia’s reference architectures and CoreWeave technology, they said…
• The factories will use Nvidia’s Rubin platform, Vera central processing units and Bluefield storage systems, the companies said.

I asked CoreWeave’s vice president of strategy, Nick Robbins, to explain that demand, which is critical to understanding the broader state of AI and why so much equity and debt capital is pouring into construction of the global compute infrastructure to support it. Those investments are driven by what Robbins describes as mind-bending demand.

Highlights of our conversation are below.

WSJ Leadership Institute: CoreWeave and Nvidia say their expanded relationship reflects the strength of demand for AI. Can that demand be measured?

…. I think the foundation for a lot of the agentic based workload is reasoning, which means they're going to have more agents using more resources to think for more people, which almost exponentially increases the need for computing [on] the inference side, not just on the training side.

WSJLI: But you did say that demand is several multiples of supply?

Robbins: Let’s say multiples of supply. Yeah, think that’s reasonable. 

We hired our first chief revenue officer late last year, Jon Jones, a senior guy within go-to-market AWS for a number of years. To be clear, we did not bring Jon on because we needed help selling capacity to the top 15 largest consumers of computing [in] the world, we're really good at that. We work with most of them already... We hired Jon to sell to the next 20,000 customers. We expect that the enterprise arm, which has already been growing... to really accelerate.

As of the end of the third quarter, CoreWeave had 2.9 gigawatts of capacity contracted, virtually all of which is scheduled to come online by the end of 2027. Consequently, any new capacity added to the pipeline since Q3 is likely scheduled for delivery at the very end of 2026, throughout 2027, and now stretching into 2028. As we enter ‘26, we're beginning to have those ‘27 conversations.

WSJLI: CoreWeave is raising considerable debt. How are those deals structured?

Robbins: The project financing model is distinct from how hyperscalers operate. CoreWeave does not build speculatively. We only deploy capital when a customer contract is signed. This contract acts as collateral, allowing us to secure debt financing, typically 90% loan-to-cost,  against the future revenue of that specific deployment. This minimizes the equity CoreWeave must put up, roughly 10%. Historically that has come in the form of a customer prepayment, or it could come from operating free cash flow.

WSJLI: What is the role of software in the newest deal with Nvidia?

Robbins: We're talking about everything from them testing and validating our software reference architecture, as we move towards jointly making our software and our operating standards... available to other Nvidia cloud and enterprise customers, which probably includes essential sovereigns as well.

WSJLI: Where does your software sit in the stack?

Robbins: It comes in at the orchestration layer … maintaining flexibility of GPUs, maintaining [the ability] to work across inference and training workloads, to run a cluster as a single 50,000 GPU cluster or 500 clusters of 100 GPUs.

Microsoft on Monday unveiled its second-generation AI chip, the Maia 200, positioning it as a direct alternative to hardware from Nvidia, Amazon and Google.

“Maia 200 is also the most efficient inference system Microsoft has ever deployed,” Scott Guthrie, Microsoft’s executive vice president for cloud and AI, in a blog post.

Built by TSMC, the AI chip is already deployed at Microsoft's U.S. Central region of data centers. Early users include the company’s superintelligence team, CNBC reports, and the chip will power Copilot and AI models such as OpenAI’s.

China’s AI labs are matching the pace with major releases of their own.

Moonshot AI released Kimi K2.5, an upgraded model that can generate text, images, and videos from a single prompt, Bloomberg reports. It can also generate code based on images and video and "direct an agent swarm with up to 100 sub-agents, executing parallel workflows across up to 1,500 tool calls," according to the Alibaba-backed company.

DeepSeek, the Chinese company that last year handed U.S.-based AI firms their existential moment, is set to unveil its latest model.

Meta Platforms, TikTok, Snap and YouTube face thousands of lawsuits alleging their technology fosters addiction and harms adolescents.

Social media companies deny the claims, which cut to the heart of their business models, saying they have invested in safety features even at the cost of growth.

This week that defense will get tested as the first such trial goes to court in Los Angeles: a personal-injury suit in which a young woman says the platforms designed addictive products that damaged her mental health.

And the pressure isn’t confined to U.S. courtrooms.

The European Commission said Monday that Elon Musk’s X may have violated the bloc’s landmark digital-content law, WSJ reports.
A new investigation will focus on how well the social network platform manages systemic risks tied to its Grok AI chatbot and expand a 2023 probe into the company’s recommender system. The move follows a public outcry over sexualized deepfake images created by Grok and shared on X.

And just days after the $14 billion deal officially created a TikTok U.S. subsidiary, California Gov. Gavin Newsom announced a review into whether the app is suppressing criticism of President Trump and ICE, Politico report.

His office said it confirmed instances of flagged or blocked content—including a direct message containing “Epstein.” Other users reported that TikTok appeared to be stifling content about Minneapolis ICE protests

TikTok in a post on X linked the issue to a “major infrastructure issue” caused by a power outage at one of its data centers.

Silence by Big Tech chiefs on the Trump administration’s actions on the streets of Minneapolis is prompting frustration among some rank-and-file workers.

Axios reports that hundreds of employees at firms including Google, OpenAI and Amazon signed a letter urging tech leaders to "speak out publicly against ICE’s violence" and cancel contracts with Immigration and Customs Enforcement.

CNBC notes the contrast with 2020, when George Floyd’s killing in Minneapolis drew swift public condemnation from tech companies.

On Saturday, the same day a Border Patrol agent killed ICU nurse Alex Pretti in Minneapolis, the CEOs of Apple, AMD and Amazon were among those attending a White House screening of a Melania Trump documentary, the Hollywood Reporter said.

The shared-workspace business, once leveled by the pandemic, is surging back, the WSJ reports.

And this time it’s major corporations like Pfizer, Amazon.com, JPMorgan Chase, Lyft and Anthropic using shared offices to avoid long-term commitments and create satellite spaces closer to workers’ homes. The sector has  grown to 158 million square feet from 115 million three years ago. And yes, a post-bankruptcy WeWork is still in the game, but not nearly as dominant as during the kombucha-on-tap glory days.

The U.S. Army signed a $5.6 billion, 10-year contract with Salesforce, the latest example of the Pentagon turning to commercial software to speed up dated processes and technology. It is by a wide margin the largest Defense Department award the  software maker has received in its recent foray into defense work, WSJ reports.