Energy / Solar construction tech · Deep dive

Terabase Energy

A solar-construction software and robotics company betting that utility-scale solar farms should be built the way factories build cars — with a digital twin, an on-site assembly line and autonomous rovers snapping panels onto trackers 24/7, instead of thousands of workers hand-bolting glass in the desert heat.

emerging

The question that decides it: Terabase's bet is that Terafab — a field-deployed automated assembly line plus install rovers pre-mating PV modules to tracker torque tubes on-site — drives the fully-loaded cost of installing a megawatt below what a manual EPC crew charges, at a repeatable per-factory throughput of ~1 GW/year, and that the software arm (Construct/PlantPredict, 25+ GW under management as of 2026) lands the developer relationships that pull Terafab onto sites. Does Terafab's installed cost-per-watt actually undercut manual labor once you amortize a capital-intensive robotic factory across real projects — before AES's Maximo, Built Robotics, Gritt, and the tracker OEMs (Nextracker, Array) either automate the same step or bundle it into hardware — and does that math still hold if the US utility-solar buildout slows on tariffs and post-IRA policy?

My take ★★ Very interesting ★ Interesting − Maybe ✗ Not interesting

Snapshot

Terabase Energy is a Berkeley, California company trying to industrialize how utility-scale solar farms get built. It sells two things: a software suite (design, procurement and construction-management tools branded PlantPredict and Construct) that developers and EPCs use to plan and run projects, and Terafab — a field-deployed, semi-robotic “automated field factory” that pre-mates solar modules to tracker torque tubes on-site and uses autonomous rovers to lay them onto pre-installed mounts, running 24/7 to cut construction labor. Founded in 2019 by ex-SunPower executive Matt Campbell, it has raised roughly $207M across five rounds, anchored by a $130M Series C led by SoftBank Vision Fund 2 in March 2025. The software touches 25-plus GW of projects; Terafab has only just moved from field testing to full commercial availability (its next-generation V2 was declared deployment-ready in March 2026). The thesis is that a chronic solar-labor shortage plus an AI-power-demand boom makes robotic construction inevitable; the question is whether Terafab’s economics beat a manual crew once you pay for the robots.

Founding story

Terabase is a bet against a decision Matt Campbell watched his own employer make. Campbell had spent about 15 years at SunPower, arriving in 2004 out of a Berkeley MBA and rising to run global power plants and power-plant products — a career spanning, by his telling, more than 200 projects in 20 countries and over $10B of investment. In 2019 SunPower pivoted away from utility-scale solar to chase residential rooftops. Campbell thought that was, in his words, exactly the wrong moment to abandon utility — that large ground-mount solar was about to become the core growth engine of the energy transition. He left.

He started Terabase in 2019 out of his basement, and reportedly won an ~800 MW project in Qatar within three months. The founding conviction was specific: utility-scale solar had gotten cheap on modules but not on construction, still a manual, weather-exposed, labor-hungry process largely unchanged in decades. Campbell’s insight, from the manufacturing side of SunPower, was that a solar farm is a repetitive assembly problem — millions of near-identical parts installed the same way — and therefore a candidate for both software orchestration and physical automation. The founding team leaned on SunPower alumni; ex-president Howard Wenger is among the angel backers.

How it works

Terabase runs two arms that are meant to reinforce each other.

The digital arm is conventional software. PlantPredict (with pro modules like Design Pro, Terrain Pro and Voltage Pro) models energy yield and engineers the plant; Construct is a cloud, GIS-native construction-management and quality-control platform that developers and EPCs use to plan procurement, sequence the build, and track quality in the field. This is normal SaaS: it books recurring revenue, scales cheaply, and — crucially — puts Terabase inside the developer’s project before a single pile goes in the ground.

The automation arm, Terafab, is the hard part and the differentiator. On a traditional site, crews first install steel torque tubes across thousands of tracker rows, then manually bolt hundreds of thousands of glass modules to them one by one — heavy, hot, repetitive work. Terafab inverts this. Palleted modules and tracker components arrive on site; a robot unpacks them and feeds module-and-torque-tube pairs through an inspection and loading point on a field-deployed assembly line. Modules are pre-mated to torque tubes with in-line quality control catching defects on the spot, and AI-assisted install rovers carry the finished assemblies out and set them onto pre-positioned mounts. Wrapping all of it is a site digital twin, supply-chain software, and an on-site wireless “command center.” The pitch: a single line at two-minute cycle times, 24/7, doing 20-plus MW a week — roughly 1 GW per factory per year — while eliminating the manual lifting of steel and glass, improving safety and the ability to keep working in extreme heat.

Product and business overview

Terabase is software-plus-hardware, deliberately. The software (PlantPredict for design, Construct for build management) is the land-and-expand wedge and the source of the “25+ GW under management” credibility. Terafab is the monetization spike: a physical service Terabase brings to site — operating the line itself or, per its 2026 launch, selling factories — to install a chunk of a plant faster and with fewer people. The two are meant to compound: software wins the developer relationship and the project data; Terafab converts it into a much larger, per-project construction contract.

Business model and pricing

Terabase does not publish price points, so unit economics are inferred. The software arm is subscription SaaS — PlantPredict/Construct sold in tiers (Pro, Enterprise), priced presumably per-seat or per-portfolio, with the appeal that it manages projects measured in gigawatts. Terafab is the opposite: a capital-intensive, services-heavy deployment monetized per-MW or per-project, where Terabase charges for installed output or sells/leases the factory. The company frames the value as labor productivity and safety rather than a sticker price — its first commercial deployment claimed a 25% efficiency gain versus manual crews, and it targets roughly a doubling of productivity with the next-generation system. The financial catch is stark: SaaS margins are lovely, but the growth story SoftBank funded is Terafab, and a robotic field factory carries capex, deployment, maintenance and mobilization costs a manual crew does not. The economics only work if fully-loaded per-MW installed cost comes in under what an EPC pays humans — and stays there across varied terrain and project sizes.

Traction over time

The shape matters: the software has real, multi-gigawatt adoption, but Terafab’s commercial record is thin — one 17 MW deployment in 2023, then a two-plus-year gap to a “ready for market” V2 in 2026, with the company guiding to two deployable factories then, a third by year-end, and ten by Q2 2027. That is ambition, not a proven fleet. Headcount is ~114 (LinkedIn, 2025); revenue undisclosed.

Market analysis

The tailwind is genuine and large. Global solar additions are projected to average ~540 GW/year through 2035 (IEA World Energy Outlook 2025), and US utility-scale demand is being pulled hard by data-center and AI power growth. The bottleneck Terabase sells against is real: the US solar workforce (~280,000 in 2025) needs to reach ~355,000 by late 2026 to hit targets — a near-term gap around 53,000 workers, most acute in utility-scale, where many firms call installation hiring “very difficult” (pv magazine / IREC, 2026). If you must build gigawatts and can’t hire crews, automation stops being optional.

But the same market is exposed. US solar economics ride on policy — IRA-era incentives and the ITC — and on tariff-exposed module supply; a slowdown, or a squeeze on developer returns, directly thins Terabase’s addressable projects. And Terabase is levered to the hardest-to-automate, most capital-intensive slice of the chain: high value if it works, high burn if adoption is slow.

Competitive intel

The threat comes from three directions (full set in the sidebar). On robotics, AES’s Maximo is the sharpest — a rival module-placing robot that installed 100 MW at scale in California and, because AES is itself a developer, needs no customer to deploy. Built Robotics automates the pile/foundation step and Gritt ($34M) is one of a widening field of automation startups chasing the same labor thesis. From above, the tracker OEMs — Nextracker (150+ GW shipped) and Array — own the hardware Terafab installs and the EPC relationships, and can push install-ease into the product itself. On software, RatedPower (Enverus) and PVcase contest the design layer. And the incumbent EPCs (Rosendin, Blattner/Quanta, McCarthy) are both customers and the manual-labor status quo Terafab must out-price. Terabase’s edge: it is the only player pairing a multi-gigawatt software footprint with a full assembly-line-plus-rover install system and a digital twin tying them together. Its exposure: a ~114-person company trying to out-industrialize both listed developers with in-house robots and multi-billion-dollar OEMs.

History and evolution

• 2019 — Campbell exits SunPower over its utility retreat; founds Terabase; seed round (Nov).
• 2020-2022 — Builds the software business; raises Series A (2020) and a $44M Series B co-led by Breakthrough Energy Ventures and Prelude Ventures (Aug 2022), with Bill Gates’ Breakthrough making the automation bet public.
• 2023 — Unveils Terafab (May); $25M growth round led by Fifth Wall (Jul); first commercial Terafab installs 17 MW of the 225 MW White Wing Ranch (AZ) with Leeward and RES (Nov), claiming ~25% efficiency — real but modest versus the “doubling” ambition.
• Mar 2025 — $130M Series C led by SoftBank Vision Fund 2 pushes total past $200M.
• Mar-Apr 2026 — Next-gen Terafab V2 completes field testing and is declared ready for sale, with a fleet ramp guided into 2027. (Naming-collision footnote: Tesla/SpaceX separately announced an unrelated “Terafab” semiconductor project in 2026.)

The through-line is a company that built a credible software business quickly and has spent years, and most of its capital, trying to make the robotics half real.

What people say

The case for. Trade press (pv magazine, Electrek, PV Tech, Solar Power World) has framed Terafab as one of the more concrete answers to solar’s labor crunch, and the software adoption is hard to argue with — Construct/PlantPredict across 25+ GW, EDP Renewables alone past 2 GW, gives Terabase real developer relationships. The investor roster is a strong signal: Breakthrough Energy Ventures, Prelude and Fifth Wall early, then SoftBank Vision Fund 2 leading a $130M Series C in 2025. Founder-market fit is about as clean as it gets: a SunPower VP who ran power-plant products and left because he thought utility solar was underbuilt. And the macro — a 53,000-worker gap against AI-driven demand — is the kind of tailwind that makes automation buyers finally sign.

The complaints. The bear case is about proof and physics. First, Terafab’s commercial record is thin: one 17 MW deployment in 2023, a multi-year gap, and a 2026 “ready for market” claim with a fleet that mostly does not exist yet — the productivity number actually delivered (~25%) badly trailed the “doubling” pitch. Second, capital intensity: a robotic field factory is expensive to build, mobilize and maintain, and the whole thesis dies if fully-loaded per-MW cost doesn’t beat a manual crew — which broader commentary notes construction-automation startups have repeatedly struggled to achieve. Third, competition converges from all sides: AES’s Maximo already installed 100 MW at scale and needs no customer; the tracker OEMs own the hardware and could commoditize install-ease; a field of funded startups (Gritt, Built Robotics) is crowding in. Fourth, policy and macro exposure: US utility solar depends on the ITC/IRA regime and tariff-sensitive module supply, and a buildout slowdown hits Terabase’s addressable projects directly. On ~114 people and a physical product still ramping, that is a lot to prove before the market decides.

Outlook: the open question

Terabase works if Terafab’s fully-loaded installed cost-per-watt genuinely undercuts a manual EPC crew — not in a hand-picked demo but across varied terrain and project sizes — at a repeatable ~1 GW/year per factory, and if the software arm (Construct/PlantPredict, 25+ GW) keeps feeding developer relationships that pull Terafab onto sites, so a capital-intensive robot fleet fills up fast enough to earn its capex. It fails if the robots don’t beat human labor once you pay for the machines, if the fleet ramps slower than the 2027 guidance (leaving expensive assets idle), if AES’s Maximo or the tracker OEMs commoditize the same install step, or if a tariff- and policy-driven solar slowdown starves the pipeline before Terafab reaches scale. The two halves of the company are on very different footing.

The software business looks durable and is arguably already a real company on its own. Terafab is the moonshot SoftBank paid for, and it is unproven where it counts. The signal to watch is not press releases about factories built but installed cost-per-watt versus manual crews on live projects, and utilization — how many promised factories are actually working on paying gigawatts by 2027. What settles it for Terabase: multiple Terafab deployments showing total-cost savings that hold up outside a demo, a fleet booked against real backlog, and the software footprint converting into Terafab pull-through. What settles it against: a fleet that ships slower than guided, per-MW economics that need subsidy or ideal conditions to pencil, or AES/tracker OEMs proving you don’t need an independent automation layer at all. For a ~114-person company against listed developers and multi-billion-dollar OEMs, the next 18-24 months of deployment data decide whether Terafab is the future of solar construction or an expensive detour off a very good software business.

How a challenger would attack it

Attack the capital intensity, not the concept. Terabase’s exposed flank is the gap between its software credibility (25+ GW) and its robotics proof (one 17 MW deployment in 2023, then a two-year silence before V2). A challenger skips the field-deployed factory entirely and follows the Maximo template: a self-contained, truck-mobile module-placing robot that works alongside existing EPC crews with no assembly line to mobilize, no command center to erect, no two-minute-cycle-time choreography to keep fed. AES proved the economics at 100 MW real scale — roughly half the time and cost of manual install — before Terabase’s V2 was even declared market-ready. The go-to-market attack is sharper still: sell the robot to EPCs as owned equipment rather than showing up as a competing site operator, because Rosendin- and Blattner-class incumbents would rather buy a tool that protects their labor margin than hire a Berkeley startup to take over their scope. Terafab’s 1 GW/year-per-factory model needs booked backlog to earn its capex; a challenger with cheaper, dumber, incrementally-deployable machines wins every project too small, too sloped, or too uncertain for a factory mobilization — which, until the guided ten-factory fleet exists in Q2 2027, is nearly all of them.

Same playbook, new buyer

Terabase aimed factory-style construction at US utility-scale solar, the one market where AES, Nextracker, Array, and a widening startup field are all converging. The same pre-assembly-plus-rover playbook fits buyers nobody is automating. Geographically: the Gulf and MENA — where Campbell won his ~800 MW Qatar project in month three — plus India and Australia, where gigawatt-scale flat desert sites are ideal for assembly lines, extreme heat makes 24/7 non-human install a genuine safety sell, and the ITC/IRA policy risk that shadows Terabase’s US pipeline doesn’t apply. Terabase, at ~114 people with SoftBank money committed to a US fleet ramp, can’t staff a second continent. Vertically: battery storage buildouts and tracker-plus-foundation scope (Built Robotics’ step) share the same repetitive-assembly physics with less entrenched labor pricing. And the software arm suggests the cheapest shift of all — Construct-style GIS-native build management sold to wind, transmission, and storage EPCs, buyers RatedPower and PVcase ignore because they only think in PV layouts. The incumbent won’t chase any of it while its existence hangs on filling ten factories with American gigawatts.

Sources and further reading

• Terabase Secures $130 Million Series C From SoftBank Vision Fund 2 (BusinessWire, March 13 2025)
• SoftBank bets $130mn on robotics-assisted solar farm company (Financial Times via Terabase, March 2025)
• Bill Gates’ Breakthrough Energy joins $44M bet on Terabase (TechCrunch, August 4 2022)
• Terabase Raises $25M to Accelerate its Digital and Automation Solutions (BusinessWire, July 19 2023)
• Terabase Energy Completes First Commercial Deployment of Terafab (Solar Power World, November 2023)
• A 24/7 solar farm-building robot just hit the market (Electrek, March 20 2026)
• Terabase Energy advances automated PV construction with robotics, AI tools (pv magazine, April 3 2026)
• Meet Maximo, the AI-enabled solar installation robot (Renewable Energy World / AES, 2024)
• Workforce gap of 53,000 threatens 2026 solar deployment targets (pv magazine USA, April 2 2026)
• Watt It Takes: Terabase Energy CEO and Co-Founder Matt Campbell (Powerhouse, founder profile)