Revenue Sharing with Cloud Partners Reshapes AI Infrastructure

The setup — AI infrastructure is hitting its physical and financial constraints, forcing the market to find new ways to build and fund raw compute. That tension defines the tape today, culminating in a structural shift for an emerging class of specialized data center operators.

What's moving

Nvidia's roadmaps hit reality. Reports indicate that Nvidia ($NVDA) will delay its next-generation AI rack systems to 2028 due to manufacturing friction (CNBC). The pause underscores how aggressive release cadences are colliding with fabrication limits, forcing investors to re-evaluate near-term hardware cycles and the physical constraints of advanced packaging.

Memory chips catch a bid. Optimism is returning to the memory sector ahead of a multibillion-dollar U.S. IPO for SK Hynix on Friday, framing the company as a pure-play proxy for AI hardware demand (TechCrunch). The halo effect lifted peers like Micron ($MU) in early trading as the market prices in sustained pricing power for high-bandwidth memory (MarketWatch).

Anthropic secures physical capacity. The AI model builder is aggressively leasing data center space in Kentucky from TeraWulf ($WULF), a crypto miner pivoting to compute infrastructure (CNBC). Simultaneously, Alibaba ($BABA) banned Anthropic’s Claude code for employees following accusations of model distillation — the practice of using one system's outputs to train a competing model — highlighting the growing friction in global intellectual property protection (CNBC).

Featured: SharonAI Holdings, Inc. ($SHAZ)

The move
The stock climbed 17.74% to close at $79.96, stepping up sharply from a $67.91 prior close. Volume registered at zero on the tape, pointing to a data feed artifact or private block settlement, but the underlying valuation reset caps an aggressive repositioning of the company's entire infrastructure model.

What drove it
The catalyst is a fundamental change in how AI compute is financed and sold. The headlines report that Nvidia introduced a new framework for cloud providers that shifts away from pure upfront capital expenditure, pairing revenue sharing with credit support to create an ongoing, usage-based model tied directly to AI tokens.

What this actually means is that Nvidia is effectively financing its own customers. SharonAI is among the first to adopt this structure, planning to deploy up to 40,000 Grace Blackwell GB300 GPUs in Australia. To fund this physical buildout, the company just closed an oversubscribed $1.6 billion financing round of equity and convertible notes backed by major institutional players. To ensure the storage layer can keep pace with the silicon, they locked in a 600-petabyte data backbone agreement with VAST Data.

The bigger picture
We are watching the rise of the neoclouds. These are specialized infrastructure providers that lease bare-metal physical GPUs, networking, and power capacity, leaving the complex upper-layer software stack for the customer to build. Traditional hyperscalers are vast and slow to adapt; neoclouds are singular and fast.

But building these AI factories requires towering capital. By sharing revenue and offering credit support, Nvidia is actively clearing the financial bottlenecks that threatened to slow down data center expansion. The cycle constraint right now is not chip design. The constraint is dollars, copper, and electricity. When the primary supplier of the industry's most valuable asset steps in to underwrite the infrastructure required to run it, the entire supply chain accelerates. The risk shifts from the hardware buyer's balance sheet to the end-user's actual consumption of AI tokens.

Across the tape

The Nasdaq 100 ($QQQ) gained 1.43%, easily pulling ahead of the S&P 500 ($SPY) which closed up 0.87%. That index volatility spread is expected to widen further when SpaceX joins the Nasdaq-100 on Tuesday while remaining absent from the broader index benchmark (MarketWatch).

Microsoft ($MSFT) is cutting 4,800 jobs across its commercial business and a shrinking Xbox gaming unit, an ongoing optimization of legacy divisions to protect margins (CNBC). Elsewhere, Solstice shares dropped 15% as the market heavily penalized the complexity and cost of its $14.5 billion acquisition of materials supplier Element Solutions (CNBC). Texas continues to direct immediate capital at its vulnerable grid, awarding a $113 million reliability grant to Southwestern Public Service for drone-based pole inspections and live capacity monitoring (Utility Dive).

What to watch

• SK Hynix’s U.S. IPO on Friday: A critical pricing test that will reveal the exact threshold of institutional demand for pure-play memory tied to the AI cycle.
• Index mechanics on Tuesday: Watch for immediate liquidity distortions as SpaceX officially enters the Nasdaq-100 without corresponding S&P 500 inclusion.
• Nvidia's rack delay fallout: Monitor the supply chain around Server ODMs (original design manufacturers) to see how deferred 2028 deployments impact near-term revenue guidance.
• Nuclear regulatory semantics: The Nuclear Regulatory Commission is considering adjustments to its safety standard terminology, a subtle bureaucratic shift that could ease the permitting of on-site data center energy builds.
• Neocloud deployment speeds: Watch SharonAI and Firmus over the next two quarters to see how quickly Nvidia's new capital-efficient financing frameworks actually convert into operational, revenue-generating compute power.