THE APEX TIMES
Navitas teams with NVIDIA on 800V-to-12V power delivery board aimed at next-gen AI servers
The companies say a new 24 kW board can convert 800 VDC directly to 12 VDC in one stage, cutting a traditional intermediate step and potentially saving space and efficiency in data-center power architecture.
Navitas, a power-semiconductor supplier, said it has collaborated with NVIDIA on a next-generation power delivery board designed for high-performance AI infrastructure. The announcement centers on an approach to converting high-voltage power supplied to servers into the low-voltage rails used by computing hardware.
According to the reported details, the solution is a 24 kW power delivery board that converts 800 VDC directly to 12 VDC in a single stage. The concept, as described, removes a separate 50 V intermediate conversion stage that is otherwise used in some power architectures.
In practical terms, the company claims the one-stage conversion can improve system efficiency and reduce board space. That matters because AI systems increasingly face packaging constraints, where shrinking the physical footprint of power conversion hardware can help designers fit more compute while maintaining power and thermal requirements.
The collaboration is framed around next-generation data-center designs, where power delivery is a major determinant of both efficiency and how much infrastructure can be packed into a given rack or enclosure. By altering the number of conversion steps and consolidating the conversion process into a single stage, the companies are aiming to simplify the power path for workloads that draw substantial power.
For NVIDIA, the relevance is tied to the company’s broader push to accelerate AI training and inference at scale. While NVIDIA’s statement in the reported post is not expanded here beyond the existence of the collaboration, the topic aligns with its focus on building and supporting platform-level efficiency across the full stack, from compute to data-center systems.
For Navitas, which provides gallium nitride and silicon-based power solutions (as typically associated with its product focus), the move underscores how semiconductor companies are increasingly partnering with platform integrators to target power architecture bottlenecks rather than only selling individual components.
The companies did not disclose, in the reported account, which specific NVIDIA platform, server design, or AI workload the board is intended for, nor did they provide performance test results, efficiency figures, thermal measurements, or qualification timelines. It is also not stated whether the board will be offered broadly as a product, or whether it is an engineering prototype for evaluation by NVIDIA and data-center customers.
What to watch next is whether NVIDIA and Navitas provide additional technical detail, such as measured efficiency under realistic load profiles, integration requirements at the system level, and any manufacturing or deployment schedule. Those details will determine whether the claimed benefits translate into faster adoption in AI server builds.
Why It Matters
- AI data centers are increasingly constrained by power delivery efficiency and physical packaging, not only by compute performance.
- A one-stage 800 V to 12 V conversion approach could reduce conversion losses and simplify the power path, if it performs as claimed in real systems.
- Removing intermediate power stages may help system designers fit more infrastructure within the same rack or chassis footprint.
- The lack of disclosed test results and deployment timelines means it is still unclear how quickly this architecture could reach production AI servers.
Key Facts
- Navitas said it is collaborating with NVIDIA on a next-generation 24 kW power delivery board for AI infrastructure.
- The board is described as converting 800 VDC directly to 12 VDC in a single stage.
- The approach is said to eliminate a separate 50 V IBC power stage used in some existing architectures.
- The companies claim the change can boost system efficiency and reduce board space.
Technology Related
Is Salesforce Still the Business You Bought? Analysts point to ongoing subscription strength, but investors want more detail
A new market commentary reviewing Salesforce’s fundamentals argues the company remains a steady subscription software business with dependable cash generation, while pointing readers back to management’s remarks from its fiscal 2027 second-quarter earnings call.
AI chip boom may outlast 2028, ASML CEO says, as Broadcom flags potential financing bottleneck
Christophe Fouquet of ASML projected that demand for advanced chipmaking capacity driven by artificial intelligence could extend past 2028. Broadcom’s Hock Tan agreed on the strength of the theme, but warned that the availability of financing could become the limiting factor for the broader industry’s expansion.
Report: AMD weighs an $8.2 billion World Labs deal aimed at bolstering its AI push
A new report says AMD is linked to an $8.2 billion acquisition of World Labs, a move that, if real and executed, could reinforce AMD’s efforts to compete in artificial intelligence hardware and platforms dominated by Nvidia.
Palantir narrative centers on a Pentagon path that could favor whichever AI wins
A recent market commentary argues Palantir’s defense positioning may translate into advantage across shifting AI ecosystems, even as the Pentagon’s choices evolve.
Dow Jones Futures Steer Lower as Micron and Nvidia Drag AI-Linked Trade; Apple and Meta Gain on Near-Term Buying Interest
Equity futures pointed down as the AI complex wobbled, with Micron and Nvidia acting as notable laggards while Apple and Meta pushed higher toward widely watched technical “buy” levels.
Navitas jumps after-hours on Nvidia-backed push for advanced AI power delivery board
Navitas Semiconductor shares rose about 5% after hours following news of a strategic collaboration with Nvidia aimed at developing an advanced power delivery board for AI infrastructure.
Microsoft leans further into custom AI chips with the Maia 300, pushing costs down as it warns about AI control
The Maia 300, Microsoft’s latest purpose-built silicon for running artificial intelligence workloads, outlines a deeper push to reduce the expense of running AI systems shortly after CEO Satya Nadella said companies cannot control the AI they already have.
Broadcom CEO Hock Tan warns many AI investors “have no clue,” comparing the AI boom to a casino
In a cautionary tone about investor behavior and market hype, Broadcom CEO Hock Tan said artificial intelligence buyers and financiers often do not understand what they are purchasing, likening the rush to gambling even as he points to the scale of the AI buildout.
Apple shares edge toward record highs as iPhone market share climbs to 22% in Q3 despite a wider smartphone downturn
A report cited by Yahoo Finance said Apple’s global smartphone share rose to 22% in the third quarter of 2026, pushing above 20% for the first time in any third quarter, even as its shipments fell year over year.
Nvidia vs. Broadcom: AI stocks weigh an OpenAI growth scare, spotlighting different paths to the same market
After a report questioned OpenAI’s recent revenue trajectory and triggered a selloff in AI-related shares, investors refocused on which semiconductor companies have the strongest link to ongoing data-center buildouts. Nvidia and Broadcom are both widely viewed as beneficiaries of AI infrastructure spending, but they sell into that buildout in different ways.