Nvidia's Local AI Bet Arrives as Investors Question Who Pays for the Cloud

Nvidia's push to put artificial intelligence on the desktop reached its widest audience yet on Friday, when RTX Spark notebooks went on sale worldwide through ASUS, Dell, HP, Lenovo, Microsoft and MSI. The launch gives the chipmaker a consumer-facing counterweight to the data-centre buildout that has powered its earnings — and that the stock market is now sizing up with growing unease.

The machines pair Blackwell RTX graphics with Nvidia's AI platform and carry up to 128 GB of shared memory. Nvidia quotes AI throughput of up to 1 petaflop in FP4 format and an entry price of $2,599. Those are manufacturer figures describing the hardware, not a guarantee that every workload will run at that speed.

Benchmarks Back the Silicon, Not the Whole Package

Early press testing published on Thursday gave the Blackwell GPUs strong marks and highlighted the appeal of running AI locally rather than in the cloud. The same reports flagged weaker results from the main processor and gaps in software compatibility. In one documented case, OpenClaw failed to establish a connection to its gateway.

That distinction matters for anyone weighing the launch. Raw AI compute and smooth day-to-day software operation are separate questions, and the mixed early reviews leave the practical payoff of the platform only partly established. Local execution is the core sales pitch, so how reliably third-party programs work with the new hardware will shape whether the range becomes a meaningful revenue line or stays a technical showcase.

Software and Vehicles Broaden the Footprint

Nvidia spent the same week widening where its technology can be used. On Wednesday it introduced cuPotential, a CUDA-X library aimed at AI agents that simulate atomic and molecular interactions. Semiconductor companies have also begun adopting its libraries for chip design and manufacturing, the company said.

Nvidia reported that early tests showed workflows already optimised for GPUs running two to three times faster, and cited Lam Research achieving a 30-fold speed-up in a simulation using NanoVDB. These are performance metrics, not sales figures — but they explain why software libraries deserve as much attention as new machines when assessing the company's recent announcements.

A day later, Volvo Cars expanded its collaboration with Nvidia on automated driving, with the technology slated for a next-generation vehicle. That, too, is a future application rather than revenue already booked.

Rate Fears and CoreWeave Stress Cap the Upside

The broadening product story ran into a harsher market backdrop. Reuters reported on Wednesday that US consumer inflation came in above expectations, strengthening the case for an interest-rate increase and sharpening concern about borrowing costs for debt-financed AI infrastructure. Doubts over whether those investments can generate adequate returns weighed on the technology sector, with financing stress at CoreWeave drawing particular scrutiny.

Nvidia closed Friday at €190.52, caught between an expanding set of use cases and the market's questions about how the AI boom gets paid for. The two threads run on different levels: product launches and partnerships demonstrate technological progress, while whether the capital behind them earns a sufficient return is a separate economic question — one the notebook rollout does little to settle.