Inside Ola's Cell Factory: I Spoke to Their Engineers

ELI5/TLDR

A YouTuber walks through Ola’s lithium-ion cell factory in Tamil Nadu and gets three engineers — head of cell R&D, head of manufacturing, and the COO — to explain how a battery cell is actually made. The headline claim: Ola has commercialised a “dry electrode” process most cell makers (Tesla included) are still trying to crack, in a 4680 cylindrical form factor, at a current footprint of 2.5 GWh ramping to 6 GWh and eventually 20 GWh in the same building. The interview is mostly substantive — chemistry, tolerances, supplier ecosystem — though the engineers go silent the moment IP gets close (electrolyte composition, cathode ratio, mandrel design). What you don’t get is anything that lets you verify yields, defect rates, or actual customer cell performance.

The Full Story

What’s actually inside a cell

Rajesh, the R&D head with 17 years in the industry (China and Japan before Ola), opens with the kindergarten version. A cell is an electrochemical sandwich: a cathode sheet, an anode sheet, a polymer separator between them, all rolled into a “jelly roll” and bathed in liquid electrolyte inside a metal can. Charging pushes lithium ions from cathode to anode; discharging sends them back. The two terminals you see on top are the only thing the user touches; everything else is geometry and chemistry.

Ola is making cylindrical cells in the 4680 form factor — 46 mm diameter, 80 mm tall — the same shape Tesla started using for the Model Y. The cathode is NMC (nickel-manganese-cobalt oxide). The anode is graphite, specifically synthetic graphite refined down to single-digit-parts-per-billion iron content. Pencil graphite, Rajesh notes, is unusable — it’s the purification, not the mining, that’s the actual technology.

Why NMC and why 4680

The interviewer keeps trying to push them to a specific NMC ratio — is it 811 (80% nickel), 955 (95% nickel), the older 111? Rajesh dodges politely:

“What I can tell you is like we are at the cutting edge of the technology in terms of our products.”

He’ll confirm the industry has moved 111 → 523 → 622 → 811 → 955, and that higher nickel content is on Ola’s roadmap, but won’t put a number on what they ship. Same with the silicon-anode question — globally everyone’s blending 5-7% silicon into the graphite for higher energy density (silicon stores ~10x more lithium per gram than graphite, but expands 400x when charged, so it cracks the anode if you go pure). Ola has it on the roadmap. No specifics.

The choice of NMC over LFP gets a clear answer: NMC packs about 35% more energy in the same volume. For a scooter where range is the thing the customer feels, you trade LFP’s better safety and longer cycle life for NMC’s energy density. The first S1 Pro needed to hit 4 kWh in a fixed pack volume; LFP couldn’t get there.

The 4680 choice is about volumetric efficiency at the pack level. Counter-intuitive, because you’d think smaller cells pack tighter:

“This is probably three times the size but this is actually holding five times the energy. It’s 5-to-1 ratio… in a 4 kW pack you’ll have about 220 of these [21700s], where you only have 56 of these [4680s].”

Fewer cells means less inactive metal casing per kWh, less wiring, less weight. The same pack volume that fits 4 kWh of 21700s fits roughly 5.2 kWh of 4680s — about 20% more range with no other change. Ola’s Roadster bike, they say, packs 9.1 kWh in 4680s versus 4.5 kWh max with 21700s in the same space.

The dry electrode bet

This is where the interview gets interesting. The standard way to make a battery electrode is wet: mix the active material (NMC powder, binder, conductive carbon) into a slurry using a toxic solvent called NMP, coat it onto aluminium foil, run it through a 70-100 metre drying oven to evaporate the solvent, then recover the solvent because it’s expensive. It works. It’s also slow, energy-hungry, and floor-space-hungry.

Ola has gone with what they call the “Trieler process” — a dry electrode method where powdered active material, binder, and carbon are mixed without solvent and pressed directly onto a foil that’s been pre-coated with a thin black resin layer. The resin acts as glue. Heat and pressure activate it. No oven. No solvent recovery. Two steps eliminated entirely:

“We actually skip two steps when you move to dry. One is the coating itself is happening in the calendaring only. The other is the vacuum drying is an additional step which happens in wet.”

The footprint claim is striking. Bala, who came from Intel and then Tesla, says the same building shell that currently runs 2.5 GWh can scale to 20 GWh with the dry process — the wet process would need a much larger floor for those drying ovens.

“If you actually go and compare what I call our density of gigawatt per square meter, we would be one of the highest density of equipment processing cell output globally.”

The hardest part of dry electrode, Bala says, is the mixing. They start with a 10 kg lab mixer to nail the recipe, scale to a 100 kg pilot line to define the architecture (what gets added when, at what speed, what temperature), then translate that to a 600-700 kg production mixer using torque-to-tip-speed ratios. Every NMC particle in 600 kg has to be coated identically. No agglomerations, no high-density patches. Tesla has been talking about commercialising dry electrode for years; Ola is claiming they’ve done it at production scale first.

Tolerances that made me put my coffee down

The cell can — the steel shell — has to hit dimensional tolerances under 0.1 mm at “more than a million cans a month” volume. India’s stainless steel industry, including aerospace and electronics, can’t currently meet the spec. So the cans are imported.

“Hitting the spec is the important… we know that we’ve got capability in this country to develop. We will work with suppliers to continue to develop. We’ll do that in parallel.”

Inside, the cathode electrode coating is specified in single-digit microns. The pressing rollers wear down — NMC is ceramic, abrasive — so they have to be replaced every two months. The CNC profile of each new roller is deliberately reverse-cambered to compensate for thermal expansion during operation, so the coating comes out flat to within microns. Months of engineering went into just that.

When the assembled cell is sealed, every can gets a helium leak test — a helium atom is small enough to find any crevice that liquid electrolyte could later escape from.

The IP wall

Three places the engineers refuse to talk:

The exact NMC ratio. “Cutting edge.”
The electrolyte composition. Rajesh: “The electrolyte is our own IP.” Bala chimes in that dry electrode needs a different electrolyte than wet because the binder polymers in dry have specific reactivities the electrolyte has to protect against — there’s no published literature, they had to develop it during trials.
The mandrel. The metal core the jelly roll is wound around. The interviewer noticed something about how it gets inserted, and the engineers shut it down hard:

“We should not discuss that. It’s an IP. We would not like to discuss this part.”

Ola claims more than 100 granted or filed patents on this single cell.

Anode is harder than it looks

Counter-intuition the engineers want to correct: cathode looks like the hard part because it’s where all the chemistry headlines live, but the anode is more sensitive in production.

“Though you might think we are handling graphite so it’s easy, actually it’s a more sensitive process. Our specs are actually quite tighter on the anode side.”

And, more broadly:

“70% of the success of a cell is basically coming from the electrode production. 30% is on assembly… 99% of the time those failures are basically the failure of the electrode rather than any other process.”

This explains why they insist 80% of the bill of materials sits in the electrode and why a factory’s whole yield problem is upstream.

Electrolyte filling and the moisture phobia

Filling electrolyte isn’t pouring. The cells are pulled to vacuum, then pressurised with nitrogen to push the liquid into the 20-25% pore space inside the jelly roll. The reason for nitrogen and not air is moisture. Lithium hexafluorophosphate (LiPF6) — the conducting salt in essentially every lithium-ion electrolyte — reacts with water to form hydrofluoric acid:

“One drop of HF can even eat your phone… It’s extremely important to keep the moisture content as minimum as possible across the process.”

The dry rooms inside the factory get progressively drier as you move toward electrolyte filling, where humidity specs are the most stringent in the building.

Form factor logic

The cylindrical-versus-prismatic question gets a practical answer: prismatic cells force you into one pack size. A 70 Ah prismatic cell is a Lego brick you can’t really break. Cylindricals, especially smaller ones, let you build 2 kWh, 3 kWh, and 4 kWh packs on the same line, just by changing the count and series-parallel arrangement. Cylindrical cells also have natural air gaps between them in a pack, which simplifies thermal management.

The vent at the top of the can — a thinned section of metal designed to rupture at a specific pressure if the cell goes into thermal runaway — is itself simulated and tuned. Pack designers route the vented gases out through channels so a single cell failure doesn’t cascade. In Ola’s scooter the LH and RH packs are oriented so vents face outward.

Key Takeaways

Form factor: 4680 cylindrical, NMC cathode, graphite anode. Manufacturing line is fungible across NMC/LFP and across cell heights up to 46x200, but not across diameter.
Capacity claim: current line at 2.5 GWh, ready to scale to 6 GWh, same building shell can hold 20 GWh, with land secured for “100 GWh and beyond.” These are claims about capability, not committed production.
Energy density (NMC): 275 Wh/kg gravimetric, stated as publicly available. LFP number not disclosed.
Pack output: 4 kWh in S1 Pro with 21700s; 4.5 kWh max with 21700s in the Roadster space; 9.1 kWh with 4680s in the same Roadster space.
Discharge rate: 3C, designed for automotive use. A higher-C variant for drones/8-12C is “in development.”
Process moat: dry electrode (Trieler process), claimed first commercial-scale deployment globally. Skips solvent drying and vacuum drying steps. Lower opex per cell than wet equivalent.
Mixing scale: 10 kg → 100 kg → 600 kg with architecture preservation across scales.
Tolerance discipline: sub-0.1 mm on cans, single-digit microns on electrode coating, helium leak test on every assembled can.
Roller wear: cathode rollers replaced every two months; anode rollers don’t wear because graphite is softer than the rollers.
Roadmap items (not shipped): silicon-blend anode, higher-nickel NMC (toward 955), high-C drone-grade variant, in-country sourcing of cans and battery-grade graphite.
Imports today: cans, graphite, cathode active material, electrolyte (made by supplier to Ola’s spec). Ola argues the ecosystem will localise as Indian cell demand scales.
What is NOT disclosed or independently verifiable: yield rate, defect rate, calendar life, cycle life, actual NMC ratio, electrolyte chemistry, mandrel design, cost per kWh, current customer.

Claude’s Take

This is a more substantive factory tour than these usually are. The interviewer did his homework — he asks about cathode/anode porosity, about N/P ratio and lithium dendrite plating, about the geometry of mixing blades, about helium leak testing — and the engineers respond with answers detailed enough to suggest they actually run the place. Bala’s Intel and Tesla pedigree is real and shows up in how he talks about CNC reverse-camber compensation. Rajesh’s chemistry answers (the silicon expansion problem, the LiPF6/HF reaction, the meaning of a 955 cathode ratio) are textbook-correct.

That doesn’t mean the corporate claims are.

Ola has had a credibility problem since IPO — software bugs, service delays, marketing claims that didn’t survive contact with reviewers, a stock that’s traded poorly relative to its hype. The cell business specifically has been positioned as the company’s ace card while losses mount on the scooter side. Bhavish Aggarwal’s stated ambition is to be a battery and chip company, not just a scooter company. That ambition shapes what gets said in interviews like this.

So separate the verifiable from the rhetorical:

Verifiable (or near-verifiable): the factory exists, it’s running, it’s making 4680 cells, the 2.5 GWh nameplate has been independently reported, dry electrode is the chosen process. The engineers are real engineers with real backgrounds. The 275 Wh/kg figure is consistent with current top-tier NMC chemistry.

Probably true but unverifiable from the interview alone: that they’re shipping commercial-scale dry-electrode cells (Tesla has been “almost there” for five years; the technical hurdle is real, and “we did it” is a strong claim that should eventually be visible in cost and weight data). That patents number 100+. That cans and electrolyte chemistry are genuinely proprietary.

Corporate-claim territory, treat with suspicion: the 20 GWh same-footprint claim, the 100 GWh land bank, the “highest gigawatt density globally” line, the “we’re 3-4 years ahead of others” line. None are necessarily false. None are tested. The thing missing from every conversation about Indian gigafactories is yield. A factory that can’t hit 90%+ first-pass yield is a sinkhole regardless of how many gigawatts of equipment you’ve installed. Yield doesn’t come up. That’s not the engineers’ fault — the interviewer didn’t ask, and they wouldn’t have answered — but it’s the number that decides whether this becomes a Tesla or a Solyndra.

The piece is worth watching for the chemistry and process intuition. Treat the capacity and “world-leading” claims as marketing until cycle-life data and per-kWh cost figures appear in third-party teardowns. Score 7: solid technical content, real engineers, but the interview is set up to flatter Ola and there’s no adversarial pressure on the parts that matter most for whether this business actually works.