By Kyle Proffitt
March 25, 2026 | The 2026 International Battery Seminar & Exhibit held its fourth Battery Venture, Innovation & Partnering event focused on the financials driving the industry. Day one this year held a renewed focus on how batteries—briefly rattled by a cooling EV industry and expiring U.S. tax credits—are finding a second wind, sails filled by the AI boom and the power needs of data centers.
Michael Liu began with a recap of the 2025 Volta report, highlighting some of the biggest stories and starting with a reassuring statistic. “In 2025, battery production grew 40%, from 1.6 TWh to 2.3 TWh,” he said. Sixty-five percent of that was LFP, and one fifth of it was for stationary storage systems of one sort or another. However, 86% of those batteries were produced in just one country—China. Accordingly, Liu reported that “half of all new vehicles in China were EVs last year, whereas that number was hovering around 9% in North America.”
Stationary storage is the new winner. Liu described 2025 as continuing the “decade of storage,” with “over 100 gigawatts of power deployed [globally] and over 250 gigawatt hours of capacity… 10% of the total power generation currently deployed in the world.” China has over 50% of the total deployments and the United States ranks second, with 20 gigawatts of power and 50 gigawatt hours of capacity. However, China is over-producing; “Chinese production alone outstrips the global demand for batteries for ESS [stationary energy storage],” Liu said. That in turn is creating strong competition and downward price pressure, pushing the average pack price down from $118/kWh in 2024 to $108/kWh in 2025. Accordingly, Liu showed that the margins for battery manufacturers are often razor-thin.
Katherine He of TDK Ventures provided important context for the renewed vitality of batteries. She had just returned from NVIDIA’s GTC conference and relayed Jensen Huang’s view that “every data center is a token factory” and his description of AI as a five-layer cake. Katherine He described the base layer, energy, saying that “the very bottom of the stack is actually the engine, which is always overlooked…right now, who owns the energy owns the AI scalability, and this is where batteries come in.”
He and Liu both accentuated a pain point for AI scaling, but one that comes with opportunity for the battery industry. “Right now, for new data centers to be connected to the grid, it will take them up to 10 or 11 years for the new interconnect,” He said. Meanwhile, AI compute demand is scaling in months. The result is that hyperscalers are acquiring energy assets and pursuing their own power, “from making investments to making acquisitions … purchasing more and more assets and even becoming energy companies themselves.” Among all the available energy sources, she argued, only batteries and solar can match the timelines these projects require, reducing the interconnect time to “only 9 months instead of waiting for years.” She added that, “using BESS is no longer a secret,” it’s widely adopted and explored by many customers.
She mentioned a recent visit to a Switch campus, a giant, eventual 495-MW data center in Las Vegas, calling it a national mission-critical infrastructure project that is outmatched only by the Hoover dam. Her stance was that it just doesn’t make sense to become a mission-critical national data center using Chinese CATL batteries as core assets.
The second big opportunity for batteries, He said, sits inside the data center with a coming power architecture shift. Traditional data centers were designed for simple, steady compute loads, but “AI compute is very different … the AI data center is very GPU-heavy; they have very bursty modes … you will see this kind of voltage spike every second,” she said. Current AC power architecture loses “10 to 20% of energy efficiency” through cascading voltage conversion steps. The solution, she argued, is a shift to a high-voltage 800 V DC power distribution architecture, a direction NVIDIA has outlined in a white paper on the future of data center infrastructure to free space for more computation.
In this new architecture, rack-level battery units would no longer sit idle as passive UPS backups. Instead, they will “very actively participate in a high-frequency, micro-charge/discharge cycles … a much faster response coordinated to synchronize the compute load.” She called this “compute-coupled storage” and noted that it demands a fundamentally different battery chemistry than what the industry currently deploys. If you use LFP in this application, she said, “you’ll be dead in two to three weeks.” She proposed an alternative with Peak Energy’s NFPP sodium-ion solution, which could last in these harsh cycling conditions for “ten to twelve years.”
Alex Manoloff, investment banker at Barclays, leaned in to the view that data centers are changing our battery needs. “Five-plus different kinds of batteries and use cases exist in data centers,” she said, including cells for UPS, providing a bridge to backup, load balancing, pulse smoothing for GPU workloads (“a huge, huge problem that people are just now waking up to”), and on-site backup energy storage, potentially replacing the backup currently provided by diesel generators. She provided an example of potential diesel replacement. “We saw a great announcement: Form Energy with Xcel and Google put 30 GWh into play … numbers like that are staggering,” she said.
With regard to financing, Manoloff explained a unique challenge around investing in this space. “On one hand, we have hyperscalers, which are the highest credit quality you could possibly have, pretty much … on the other hand, they’re moving at the speed of light, and they’re doing something completely different, in some cases, that they’ve never done before. How do you assess risk? How do you do due diligence? How do you price this? How do you structure all of these things? We’re just trying to figure it out in real time.”
Manoloff ended her time by citing a surprising statistic that reveals the speed of change. “The fourth largest end market for lithium is now a robot. It’s right next to consumer electronics … It’s not some niche end case anymore. It’s here. It’s everywhere,” she said.
Jun Sup Shin of Blue Nova agreed with the assessment of batteries as critical in the AI era. However, he was focused on the idiosyncrasies of individual batteries, indicating that each one lives a unique life. The same cell from the same factory, he said, can perform very differently a decade into operation depending on how it has been used and managed. And we have to plan for that with “a system designed to survive its worst day.” “Are we doing that?” he asked. Shin argued that lab-collected data cannot replace actual data from the field.
He called for a rethinking of battery certification regimes, not just as a one-time event at deployment, but as an ongoing, dynamic assessment. “Do we ever go back two … ten years down the road, and re-certify that these batteries are working as designed, and are still safe and reliable?” He proposed industry-level data sharing frameworks, including the use of blockchain to protect proprietary information while enabling collective learning from field deployments, a way for manufacturers, integrators, and operators to share data “in a secure way so that they are not giving up their trade secrets.”
