A "gigafactory" is a factory so big its output is measured in billions — gigawatt-hours of battery, gigawatts of solar, or gigawatts of AI compute. Tesla coined the word for a single plant meant to out-build the entire battery industry. A decade later the gigawatt is the unit of the whole frontier: the batteries, panels and data centres being poured at a scale never attempted before.
The prefix "giga" means a billion. A gigawatt is a billion watts — the output of a large nuclear reactor, or roughly the power draw of a mid-sized city. A gigafactory is simply a factory whose capacity is measured in those billion-scale units rather than the millions a normal plant deals in.
Tesla popularised the term in September 2014, when it announced Gigafactory Nevada with an audacious promise: a single building that would produce more lithium-ion battery capacity than the entire world did at the time. It worked. The plant now turns out over 35 gigawatt-hours of cells a year — and the word escaped the battery world to describe any mega-plant built at gigawatt scale.
Today "gigafactory" is used three different ways, one for each frontier it powers: gigawatt-hours (GWh) of battery energy, gigawatts (GW) of solar or nuclear output, and — newest of all — gigawatts of electrical draw for the AI data centres now being built like industrial plants. Same prefix, three different frontiers.
Battery plants are rated in gigawatt-hours per year — how much stored energy their cells can hold, added up. The world now has 240+ of them and about 2,500 GWh of capacity, roughly three-quarters of it in China. This is the meaning Tesla invented, and still the most common one.
Solar factories are rated in gigawatts per year — the peak power of all the panels they can build. China dominates: Tongwei alone can make over 150 GW of cells a year. In the US, First Solar is scaling toward 6 GW a year to decouple the supply chain from China.
The newest gigafactory doesn't make anything physical — it consumes power. AI training campuses are now measured by the gigawatts of electricity they draw. Five separate 1-GW-plus sites come online in 2026, each the size of a small city's entire grid demand.
Feeding all of the above takes gigawatts of firm power. A single large reactor is about 1 GW; the AI build-out is now reviving nuclear, from restarted plants to factory-built small modular reactors, covered on the Energy page.
China's CATL has led the world in EV batteries for nine straight years, with over 40% of the global market in 2026. Its Liyang plant alone makes 126 GWh a year, and it is exporting the model to Europe — a 100 GWh plant in Hungary and a 50 GWh LFP plant in Spain with Stellantis.
BYD started as a battery company and never stopped. It installed nearly 195 GWh of its own Blade LFP cells in 2025 — the vertically-integrated model that lets it undercut rivals on price and now outsell Tesla on EV volume.
Giga Nevada (with Panasonic) proved the concept and passed 35 GWh a year; Giga Texas is one of the largest buildings on Earth by footprint; and Berlin and Shanghai round out a network now approaching 200 GWh of theoretical output.
Europe's great battery hope collapsed in 2025 — the largest bankruptcy in Swedish history. Its flagship plant was meant to reach 16 GWh but managed just 1 GWh, a reminder that a gigafactory is one of the hardest things in industry to actually ramp.
The fastest-growing kind of gigafactory makes no product at all — it turns electricity into intelligence. Training a frontier AI model now takes so much compute that the data centres housing the chips are planned like heavy industry, and rated the same way: by the gigawatts of power they pull from the grid. 2026 is the year the first ones cross a gigawatt each.
Elon Musk's xAI expanded its Memphis site to a full 2 gigawatts, housing about 555,000 NVIDIA GPUs bought for roughly $18 billion — the largest single AI training site yet, built in about a year.
OpenAI and Oracle's Stargate campus in Abilene, Texas scales toward 1.2 GW and around 400,000 NVIDIA B200s by the end of 2026 — the anchor of a $100B-plus build-out.
Meta's Prometheus in New Albany, Ohio is engineered as a 1-gigawatt supercluster with roughly 500,000 GPUs — and its Hyperion site in Louisiana is designed to reach 5 GW.
Five hyperscalers — Amazon/Anthropic, xAI, Microsoft, Meta and OpenAI — each bring a 1-GW-plus site online in 2026. The binding constraint is no longer chips but power, which is why these campuses are now sited next to their own generation.
A cross-section of the world's gigawatt-scale plants — batteries, solar and AI compute — by scale and status.
| Facility | Operator | Type | Scale | Status |
|---|---|---|---|---|
| Liyang | CATL (China) | Battery | 126 GWh/yr | Operating |
| Debrecen | CATL (Hungary) | Battery | 100 GWh/yr | Ramping 2026 |
| Zaragoza | CATL + Stellantis (Spain) | Battery · LFP | 50 GWh/yr | Building · 2026 |
| Giga Nevada | Tesla + Panasonic (USA) | Battery | 35 → 150 GWh/yr | Expanding |
| Giga Texas | Tesla (USA) | Battery + EV | 10M+ ft² floor | Operating |
| Northvolt Ett | Northvolt (Sweden) | Battery | 16 GWh target | Bankrupt 2025 |
| Sichuan | Tongwei (China) | Solar cells | 150 GW/yr | Operating |
| Lake Township | First Solar (USA) | Solar | ~6 GW/yr (US total) | Operating |
| Colossus 2 | xAI (Memphis, USA) | AI compute | 2 GW · 555k GPUs | Ramping 2026 |
| Stargate Abilene | OpenAI + Oracle (USA) | AI compute | ~1.2 GW | 2026 |
| Prometheus | Meta (Ohio, USA) | AI compute | 1 GW · 500k GPUs | 2026 |
| Hyperion | Meta (Louisiana, USA) | AI compute | 5 GW (planned) | ~2030 |
Scale is the strategy. A gigafactory's whole point is that cost falls as volume rises. Making cells, panels or compute a hundred times bigger than the old normal is what dragged battery prices down ~90% in a decade and put solar below every fossil fuel — the single biggest reason the energy transition is now cheaper than the alternative.
Power is the new bottleneck. Whether the output is stored energy, generated electricity or trained intelligence, every gigafactory is really a bet on gigawatts of power. That is why battery, solar, nuclear and AI have collided into one story — and why the map of where these plants get built is increasingly the map of where the grid can feed them.
Ramping is brutal. Announcing a gigafactory is easy; making one actually produce at nameplate is one of the hardest feats in manufacturing, as Northvolt's collapse showed. The winners — CATL, BYD, Tesla — are the ones who solved the boring problem of yield at scale, not just the headline of breaking ground.
> Sources: company disclosures and public reporting on CATL, BYD, Tesla, Northvolt, Tongwei and First Solar; hyperscaler AI data-center announcements (xAI, OpenAI/Oracle, Meta); IDTechEx / Benchmark battery-capacity data. Figures are latest stated or reported values and shift with each build.