Software digitized the office. Filing cabinets became databases; human queues became APIs. The firms that embraced code sprinted ahead, and those that didn’t fell behind.
But the revolution was uneven. In the physical economy, software digitized planning and design, yet “last‑mile” actions remained bottlenecked by real‑world complexity and human‑in‑the‑loop operations. Manufacturing productivity boomed in the 1990s with IT and automation, then plateaued in the mid-2000s as easy gains were absorbed and complementary investments needed to automate further hit institutional and physical constraints.
By contrast, digital sectors, like e-commerce and media, rapidly compounded, iterating in code, scaling on cloud and mobile, and sidestepping the frictions of steel, concrete, and regulation. For physical sectors to match that trajectory, we first need a bridge that truly connects bits to atoms.
That bridge is the electro‑industrial stack — the technologies that enable machines to behave like software: minerals and metals processed into advanced components, energy stored in batteries, electrons channeled by power electronics, force delivered by motors and actuators, all orchestrated by software running on high-performance compute.
This is the shift from software that merely summons a taxi to software that takes the wheel. Software ate the world. Now it will move it.
The Electro-Industrial Stack Is Inevitable
The electro-industrial stack is changing how we build and run machines. Put simply, the 2010s wired physical workflows with APIs, but the 2020s will put autonomous systems and agents in control of them.
Industrial machines, in particular, might’ve come with interfaces and endpoints, but using them still depended on tacit knowledge, custom code, and operators skilled enough to coax complex equipment into motion. Programmable logic controllers, it turned out, are hardly programmable. These systems were built to leverage software, but with software never fully in control, industrial productivity ultimately remained constrained.
But that’s beginning to change. Advanced reasoning models and machine learning will continue to collapse the distance between human intent and machine execution: language becomes code, code becomes control, and control drives the machine. We are headed toward a world where a task that once demanded years of expertise and skilled technicians is now mediated by an interface as natural as a conversation.
The machines themselves are changing, too. Electrified systems, built on batteries, power electronics, and high-torque motors, are more efficient, more precise, and more responsive to software. They can be tested in simulation, updated over the air, and improved continuously as telemetry feeds back into design. In practice, this means closed-loop systems where perception, planning, and control run in real-time, and hardware is designed to support deterministic control. In other words, with the electro-industrial stack, physical machines are beginning to behave like software.
Importantly, the electro-industrial stack isn’t new, but its presence is easy to miss. Today, core components already underpin energy storage systems, electric vehicles, drones, and industrial robots. Past enabling breakthroughs span rare-earth production at Mountain Pass, lithium-ion chemistry pioneered at Argonne National Lab, SiC power devices from DARPA-backed research, GM’s early permanent-magnet motors, robotics from NASA, autonomy advances from the DARPA Grand Challenges, and, of course, the modern logic chips born in Silicon Valley.
However, China now increasingly dominates both research and production of these technologies. This is particularly true for upstream metals/chemicals, batteries, motors/actuators, and power electronics. And where it doesn’t lead, like in high-performance compute, it is either investing aggressively to catch up or seeking to collect as a spoil of war. These technologies may be American, but the capacity to build, scale, and upgrade them increasingly isn’t.
More broadly, the collision of AI with a digitally integrated industrial base is the real unlock for step-change productivity in the physical world. Tesla has set the pace, pulling this future forward by nearly a decade through fusing software, electrified systems, and networked sensors across its products and factories. But we must spread and accelerate that model further. Early signs are emerging: Diode is layering frontier language models on top of EDA tools to design PCBs, while others are deploying reasoning models to supervise autonomous systems. This is how the Dow Jones begins to look more like the Nasdaq. The substrate that makes this future possible is the electro-industrial stack.
If we want to lead the next industrial era and remain the world’s technology and military superpower, we must always skate where the puck is headed. As China races forward, moving goods, people, and information at machine speed, we risk being stuck in the past. The following is a discussion of each segment of the electro-industrial stack, and what it might take to secure leadership.
Minerals & Metals
Lithium, nickel, copper, and rare earths remain the headline dependencies. But as compute, autonomy, and electrification scale, new chokepoints are surfacing: uranium for nuclear reactors, graphite for batteries, gallium for semiconductors, and magnesium for lightweight alloys, among many others. These minerals are becoming as strategically contested as oil was in the 20th century.
Ore deposits are scattered globally, but the real leverage sits in the midstream where China has built a state-backed, vertically integrated machine. That midstream controls the chemistry, metallurgy, and finishing that turn raw concentrates into battery-grade chemicals, alloys, foils, laminations, and powders. It’s also where margins expand and supply chains harden, locking in customers through long-term offtakes and driving upstream integration to secure feedstock — exactly the strategy China has executed worldwide.
But today’s mining majors and juniors are not structured to move at the speed this moment demands. Majors are optimized for decades-long megaprojects, high compliance loads, and predictable cash flows, not rapid adoption of new technologies or even decades-old software. Juniors may be faster and more risk-tolerant, but they lack the capital depth, permitting muscle, and integrated downstream relationships to effectively develop and operate a mineral project on their own. As a result, even breakthrough extraction, refining, and software innovations struggle to scale; closing the gap will require a new class of mining company.
The path forward, as we’ve outlined in our thesis, is to build vertically integrated, software-native mineral and metal companies that operate at machine speed — using AI, real-time telemetry, and closed-loop operations to compress development timelines and operate more efficiently. In parallel, friend-shore the midstream with allies and finance domestic capacity in America where it makes sense. Pair this with long-term offtakes, floor-price insurance, targeted subsidies, export-finance support, and permitting reform so projects reach scale, fast. Companies like KoBold Metals and Mariana Minerals are already showing how it can be done.
Batteries
Before lithium-ion, electricity was largely just-in-time — generated and consumed in the same moment. Solar and other renewables could produce abundant power during the day and close to where it was needed, but without cost-effective storage, its value quickly declined. Low-cost, high-energy density batteries changed that. We can now store, move, modulate, and deliver electricity efficiently, powering everything from data centers to drones.
Put bluntly, the electro-industrial stack cannot scale without a major Western battery cell manufacturer able to supply energy storage units for systems beyond the tether of the grid. A true “energy dominance” strategy must include batteries. The Inflation Reduction Act unleashed announcements and domestic-content rules, but many projects are stuck in permitting, capital, or policy limbo. A few players — Tesla, ONE, Ultium, and others — are still building, but America will ultimately need hundreds of GWh of annual capacity to meet targets.
As the collapse of Northvolt recently illustrated, building a Western battery base is brutally hard. You must secure precursors, master cell chemistries and coatings, and stand up module/pack lines with meaningful yield — all capital-intensive and earning thin margins. Increasingly, Chinese champions lead on both scale and process knowledge: CATL supplies ~40% of global EV capacity and BYD ~20%. Both supply Tesla with cells and manufacturing equipment. As high-nickel blends cede market share to lithium-iron phosphate and newer chemistries, the center of gravity in manufacturing and R&D expertise shifts more to China.
Critically, tariffs and bans alone haven’t reduced dependence; they often just shift Chinese production to other compliant jurisdictions. Real independence means constructing entirely new, allied industrial ecosystems with countries like Korea, Japan, Mexico, and Vietnam.
Coming from the demand side, our portfolio company Base Power shows a potentially winning model: capture higher-margin energy storage services with proprietary software and strong customer relationships, then integrate upstream into critical components to stabilize volatile supply chains and capture margin. But if America is to truly lead in batteries — for the grid, drones, or robotics — it must go further to radically accelerate production capacity.
That means a comprehensive industrial strategy with long-term offtakes, targeted subsidies, strategic partnerships, and aggressive permitting reform, paired with rapid manufacturing scale-up through the transfer of process expertise. The Panasonic–Tesla partnership is a case in point: Japan brought decades of cell manufacturing know-how, enabling the Gigafactory to reach competitive yields far faster than a greenfield effort could have achieved alone. Indeed, reshoring at scale will require the full weight of policy, procurement, and internationally-coordinated industrial execution.
Power Electronics
Power electronics are the hidden nervous system of modern machines. At their core are power semiconductors which, unlike logic chips that process information, manage energy itself — converting, inverting, and regulating flows between sources and loads.
Historically, power systems relied on slow silicon switches, steel-core transformers, and bulky analog controls. At high voltage and frequency those approaches are at their limits. Wide-bandgap devices — like silicon carbide (SiC) and gallium nitride (GaN) — switch faster, withstand higher temperatures, and enable precise digital control. This software-driven, solid-state (no moving parts) foundation stitches together the electro-industrial stack.
Already, SiC anchors EV inverters, DC fast chargers, and utility-scale solar/wind, while GaN shines in compact chargers, telecom, and high-powered radar systems. Both are spreading into data-center power infrastructure as voltages increase, and adoption across aerospace, defense, and the grid is moving from pilots to real programs. For example, Astro Mechanica uses SiC converters to drive high-power electric motors in a jet-engine compressor, unlocking new flight regimes and electricity-hungry military applications.
Notably, the United States still holds a technical edge in WBG design and manufacturing knowledge, but upstream risk is real: China dominates gallium and other key inputs, and is scaling wafers and devices rapidly. YOFC’s ~$2.8B SiC wafer fab in Wuhan reportedly ramped in about 18 months and targets a large share of domestic demand, BYD is standing up production for its 10C(!) fast-chargers, and Innoscience has been selected as a key GaN supplier for Nvidia within emerging 800V DC data-center power ecosystems. This rise pressures Western suppliers, made sharper as Wolfspeed, the American pioneers of SiC manufacturing, navigates bankruptcy.
Scaling WBG power electronics is also critical to easing the grid’s growing infrastructure bottlenecks. Since 2020, demand for some power transformer classes has tripled, while over half of the nation’s 40 million distribution transformers are past their 33-year service life. The United States now imports 80% of our power transformers and 50% of our distribution units, with tariffs on electric steel and copper pushing costs higher. Lead times have surged and some unit prices are up 95%. This system is not set up to meet surging electricity demand; the way forward is solid-state transformers built with domestically produced WBG power electronics.
To avoid bottlenecks, America must scale 200 mm wafer and epitaxy production with higher yields, and drive SiC/GaN adoption into the highest-leverage use cases like power transformers and spectrum dominance. Policy should create demand pull and fund wide-bandgap packaging and R&D. Future systems will require vast amounts of precisely managed power; delivering it will depend on solid-state electronics under digital control.
Motors & Actuators
Motors and actuators convert electrical energy into mechanical motion, like in a drone motor or an industrial robot arm. Today’s performance leader is the brushless permanent-magnet synchronous motor (PMSM) using NdFeB magnets, prized for torque density, efficiency, and compactness. But that advantage comes with a strategic cost: dependence on rare earths.
Alternatives for motion systems span both motor design and actuation type, each with trade-offs. On the motor side, induction designs are rugged and rare-earth-free but less efficient; wound-field machines eliminate permanent magnets but add complexity and thermal overhead; switched-reluctance motors avoid magnets entirely and tolerate heat yet require sophisticated controls; and ferrite-based PMSMs use low-cost magnets but deliver less performance. The market is testing these across applications — Tesla moved from induction to PMSM for efficiency, later signaling a rare-earth-free next-gen design, while BMW and Renault have deployed wound-field units.
Actuation choices are also evolving. Flight surfaces, reclining seats, missile fins, landing gear, and industrial end-effectors are shifting from old-school hydraulics to electromechanical systems for lower weight, higher reliability, and precise digital control. Linear and direct-drive actuators dominate high-performance use cases today, while piezoelectric designs are emerging in “soft” robotics and other niche applications where fine precision and compliance matter most. Physics still imposes limits, but incremental advances in materials, controls, and integration continue to narrow performance gaps across the spectrum. Open-source designs and broader commodification of components further expand commercial possibilities
Notably, this same stack underpins small “Group 1” drones that have changed the modern battlefield, and which depend heavily on foreign-sourced motors. Though Western firms still hold the lead at the highest performance tier for specific motors, like servos, the most acute near-term production gap lies in drone motors, still sourced overwhelmingly from China.
To fix this, America should continue securing magnet supply, encourage rare-earth-free motor designs where performance allows, reshore PMSM production for mission-critical uses, and rapidly stand up a domestic drone-propulsion ecosystem. In parallel, invest in precision components, like gearboxes, bearings, and encoders. Small gains in motor and actuator efficiency compound across the stack, and as general-purpose robotics scale, this industrial “muscle” will move larger fractions of GDP.
Compute
The compute layer converts electrical energy into intelligence, controlling everything from autonomous vehicles to advanced weapon systems and industrial robots. Today’s performance leader is 2 to 4 nm-class logic — most often, GPUs designed by Nvidia and fabricated by TSMC. That lead presents a geopolitical challenge; America depends on offshore fabs to produce its most advanced chips.
Breaking the Nvidia/TSMC dominance is quite challenging. Custom AI ASICs might be able to beat GPUs on latency and cost at scale but face the same foundry and packaging limitations, plus have weaker software. Ultimately, ecosystems drive performance and CUDA is a deep moat. Export controls cap Chinese import of top GPUs, but gray markets and a new 15% levy on older chips like Nvidia’s H20 keep clusters growing on legacy nodes (and fueled by China’s 2x larger electric grid.)
Compute extends beyond the GPU alone; advanced packaging now sets performance limits as much as transistor scaling. System design has expanded from single chips to whole machines, like co-optimizing die, memory, and interconnects alongside rack-level power delivery and cooling. Just as important, software frameworks, compilers, kernels, and drivers map models onto this topology, manage communication and memory, and orchestrate performance across infrastructure.
Chinese firms lead in scale for mature nodes and low-cost packaging; the West leads in EDA, lithography, and software ecosystems, but still lacks true leading-edge manufacturing capacity outside Taiwan. To win, Americans and allies must build domestic foundry and advanced packaging hubs, developing deeper ecosystems for substrates, HBM, and assembly & test. Export controls on advanced hardware and software, including EUV lithography, EDA tools, and ISAs, will also remain important levers.
Critically, America needs an “American TSMC”, as TSMC remains reluctant to offshore leading-edge fabs — known as the “Silicon Shield” strategy. Whether Intel can rebound to fill that role is an open question. In the meantime, firms are hedging; Tesla is leaning on Samsung to stand up capacity in Texas. Until America controls fabs and packaging at home, the entire stack remains strategically exposed.
The Critical Path to Powering Everything
The electro-industrial stack is the bridge between software and the physical world, the foundation animating the machines that will ultimately shape the future.
No company illustrates this power more clearly than BYD. What began as a Chinese battery maker now dominates the global EV market and extends into cargo ships, trains, buses, and industrial equipment. BYD even supplies more than half of DJI’s drones, by some accounts. This breadth is possible because its products share the same core technologies in which BYD has built deep expertise — mineral sourcing and refining, batteries, motors/actuators, power electronics, compute, and final assembly.
We should not aim to build an “American BYD”, but we must create its equivalent through an ecosystem of integrated suppliers and OEMs capable of the same system engineering discipline. Without this foundation, companies are forced to vertically-integrate by necessity, not strategy, and risk bottlenecks when scaling.
In such scenarios, Anduril can’t field drones; Saronic can’t control unmanned vessels; Base Power can’t deliver battery storage; Astro Mechanica can’t power their engines; Apex can’t keep spacecraft online; Castelion can’t guide hypersonic missiles; Radiant can’t turn nuclear heat into electricity; Northwood and CX2 can’t build RF systems; SpaceX can’t land rockets; and frontier AI labs can’t train or serve models. Without securing the electro-industrial stack, America is not truly in control of its future.
Critically, winning will require treating the stack as a single, integrated system, not a set of disconnected parts. Siloed thinking only shifts bottlenecks. And while it’s not the focus here, powering these technologies will also demand far more generation capacity. Meeting this challenge will take world-class talent, deep operational expertise, and coordinated policy across the value chain — anchored in the United States, and aligned with trusted allies.
We can’t wait for this to self-assemble. At Andreessen Horowitz, we’ve been backing this future for years; now it has a name. Driven by software and artificial intelligence, the electro-industrial stack will move the world, unleashing a productivity boom and industrial renaissance. It’s up to us to ensure it moves on our terms.
