Subscribe to Our Digest
ABB unveils Infinitus, the world’s first source-to-rack DC portfolio for next-gen data centers and energy-intensive industries.
By: Filippo Di Giovanni, Data Centre Digest Contributing Editor
Artificial intelligence (AI) is reshaping the electrical architecture of data centers. As GPU power continues to scale, the industry is moving from conventional 48 V and AC-based distribution toward 800 VDC, with NVIDIA, the Open Compute Project (OCP), power-semiconductor companies and electrical-equipment suppliers focusing on different aspects of the transition.
ABB’s new Infinitus portfolio marks a major milestone in this shift. Rather than focusing only on the power conversion adjacent to data center racks, ABB is proposing an end-to-end architecture extending from medium- voltage AC at the facility entrance to 800 VDC distribution and protection at the rack. The approach builds on ABB’s collaboration with NVIDIA on 800 VDC architectures for future gigawatt-scale AI facilities and combines solid- state transformation, DC distribution, protection and cooling optimization.
To explore ABB’s strategy and vision behind Infinitus, Filippo Di Giovanni, Data Centre Digest Contributing Editor recently spoke with Giampiero Frisio, President of ABB’s Electrification Business Area.
The reason for increasing the distribution voltage lies essentially in a simple (P=I 2 R) equation, relating resistive power losses and current. Increasing distribution voltages reduce current, which in turn reduces conduction losses and the amount of copper for a given power level. This becomes increasingly important as AI racks approach the 200 kW range and systems are expected to require 1 MW per rack. At such power levels, distributing electricity at relatively low voltage would require exceptionally high currents, creating challenges for conductors, connectors, busbars, protection and cooling.
The industry is consequently heading toward much higher-voltage DC distribution. NVIDIA announced its 800 VDC architecture for future megawatt-class AI systems, while the OCP is developing specifications for high-voltage DC power delivery. Similarly, ABB’s own Infinitus portfolio identifies 800 VDC as a foundation for the next generation of AI data centers.
Today’s AI power architecture is mainly based on multiple voltage conversions. AC power from the grid is transformed and distributed across the facility before reaching a UPS and eventually the IT rack. Within the rack, power supplies convert AC to an intermediate DC voltage, typically around 48–50 V, which is then further converted to the much lower voltages required by CPUs, GPUs, TPU, and other chips.
A 200 kW rack operating from a 50 V bus would, in principle, require 4,000 A. At 800 V, the same power corresponds to only 250 A. At 1 MW, the corresponding figures become 20,000 A and 1,250 A, respectively. These are purely theoretical figures that do not account for conversion losses and other topology details, but they attest to the industry’s need to transition to higher-power distribution schemes.
Behind this transition is NVIDIA, the undisputed world leader in GPUs. In October 2025 ABB announced a collaboration with NVIDIA aimed at supporting the 800 VDC architecture for 1 MW server racks. This collaboration is important for understanding ABB’s expansion into this fast-growing market.
Infinitus is not simply a response to a new market trend. It elevates this collaboration to a broader portfolio level, bringing together the technologies needed to build the core electrical infrastructure for this new architecture. “ABB’s strongest differentiator is the breadth of the industry’s first integrated source-to-rack DC portfolio, which brings together the building blocks needed to support the power architecture from source to rack. These span medium-voltage powertrains, DC sources and power quality, power distribution, protection and cooling optimization,” said Frisio.
Frisio highlighted the company’s ability to cover the complete value chain: “The broader advantage is system-level engineering. Rather than optimizing individual products in isolation, ABB can help customers optimize the complete power architecture, simplifying integration, reducing compatibility risks, and supporting more efficient and reliable deployment. ABB drives and high-efficiency motors also optimize cooling equipment, helping improve energy productivity while maintaining the thermal conditions required by high-density AI workloads.”
The transition to 800 VDC is unlikely to materialize overnight. Data-center operators and hyperscalers have heavily invested in AC transformers, switchgear, backup systems, busways and distribution equipment. Replacing this infrastructure simply to gain a few percentage points of efficiency would rarely be economical. A more likely path is therefore evolutionary. One option is to retain much of the existing AC infrastructure and introduce an 800 VDC power rack, or “sidecar,” close to the IT racks. AC is delivered to the cabinet, converted to high-voltage DC and distributed over relatively short distances to the AI racks. This approach allows high-voltage DC to be introduced without redesigning the entire electrical infrastructure.
To this end, ABB supports wider initiatives to help operators optimize power distribution and ultimately improve productivity. “Infinitus does not prescribe a single migration path or necessarily require the immediate replacement of existing 48 V infrastructure. ABB is developing building blocks for new DC-native facilities, hybrid AC/DC environments and retrofit applications. As rack density and power requirements increase, operators can determine where introducing 800 VDC creates the greatest value,” commented Frisio.
ABB explicitly supports both DC-native and hybrid AC/DC architectures, as Frisio further explained: “ABB supports different paths to 800 VDC through technologies including the Infinitus Solid State Transformer, DC UPS and Transformer Rectifier Unit, together with DC distribution and protection solutions. This allows ABB to work with customers to identify the architecture best suited to their current infrastructure and future computing requirements.”
The company expects the main Infinitus components to enter production within the next 12 months and says the first DC-native facilities using the complete portfolio could be operational within two to three years.
The higher voltage is not without challenges. Interrupting AC current is generally easier than interrupting DC current because AC naturally passes through zeros during its operation, helping to extinguish the arc. At hundreds of kilowatts or megawatts, a fault on an 800 V bus is indeed a substantial amount of stored and delivered energy that must be detected, interrupted, and contained rapidly. Protection must consequently align with distribution.
“At the technology level, ABB is the only major electrification supplier with both the world’s first fully IEC-certified solid-state circuit breaker and an industry-first static medium-voltage UPS in production. The Infinitus Solid State Transformer is designed to convert medium-voltage AC directly to 800 VDC with fewer conversion stages,” remarked Frisio.
Regarding the economic benefits such as extra computing capacity achievable through efficiency improvements, Frisio gave a detailed explanation: “The benefits are interconnected, and there is no universal breakdown between conversion and distribution losses. The actual contribution of each depends on the baseline architecture, number of conversion stages, system configuration, and load profile. By reducing power-conversion and distribution losses, the Infinitus DC portfolio can deliver efficiency gains of more than 5%.”
He then gave an estimation for the expected gain: “For a 500 MW data center, ABB estimates that a 5% efficiency gain can make 25 MW more power available for revenue-generating IT loads. Reducing conductor requirements and the power infrastructure footprint can also free space for additional compute racks. Together, these benefits allow operators to increase the computing capacity delivered from the power and space already available. In the scenario assessed by ABB and BCG, this could represent more than $300 million in potential additional revenue,” added Frisio11.
Asked why ABB mentions common characteristics with other markets such as manufacturing, renewables and marine operations, as well as energy storage, Frisio responded: “These industries share several characteristics: significant power-conversion requirements, high or rapidly changing loads, and growing use of technologies that generate, store or use power internally as DC. Renewable energy sources such as solar generate DC power, batteries store it as DC, and EV charging relies on DC power conversion. Modern factories are also already partly DC internally, with variable-frequency drives controlling motors,conveyors and robots.”
Remarking the advantage of widespread DC deployment, he said: “Using DC distribution where it creates value can reduce unnecessary conversion stages and associated energy losses, improve the integration of renewable generation and battery storage, and reduce infrastructure complexity.”
ABB’s announcement goes beyond a new portfolio of electrical equipment. It reflects a broader shift in data-center power architecture. As AI rack power approaches the megawatt range—driven by accelerators that individually draw on the order of a kilowatt, aggregated by the dozens per rack—DC is becoming increasingly attractive as the primary distribution platform inside the facility, while AC remains the interface to the grid.
NVIDIA is driving this transition from the computing side, OCP is developing interoperable specifications, and power-semiconductor companies are providing solutions to implement it. ABB is extending the architecture all the way to the medium-voltage infrastructure of the data center. The final power hierarchy could be:
MV AC → 800 VDC → intermediate DC → processor voltage.
Implementing it will require new power-converter topologies, protection devices, connectors, cables, cooling systems and standards. But the ultimate goal is clear: as AI compute density increases, delivering power efficiently becomes part of the computing architecture itself.
1 Boston Consulting Group2 https://www.abb.com/global/en/news/138900/abbs-new-direct-current-portfolio-aims-to-rewire-ai-data-center-energy-infrastructure3 https://www.abb.com/global/en/areas/electrification/campaigns/infinitus-dc-portfolio