A hydrogen electrolyzer is a plant that splits water into hydrogen and oxygen using direct current: stacks of electrochemical cells, an alkaline liquid electrolyte or a proton exchange membrane between electrodes, driven by a rectifier at currents that push hydrogen off one electrode and oxygen off the other, with balance of plant around them feeding purified water, managing heat, separating the product gases from their carrier liquids, drying and delivering hydrogen at the pressure the customer needs. It is the machine that converts electricity into chemical energy, and its purchase decision is dominated by one ratio, kilowatt hours of electricity per kilogram or normal cubic meter of hydrogen, crossed with the price of that electricity and the hours the plant will run.
This page treats the hydrogen electrolyzer as a purchasing object: how the three architectures, alkaline, PEM, and solid oxide, trade efficiency, current density, pressure, and maturity; how specific energy and its efficiency map decompose into stack voltage losses and balance of plant loads; how water quality, turndown, and ramp behavior match renewable power; and how purity, drying, oxygen handling, and hydrogen's unforgiving safety physics complete the machine. It is not an air separation machine: the oxygen generator and nitrogen generator on related pages split air, while this one splits water.
The decision concentrates in five places: the hydrogen demand profile, flow, purity, pressure, and its hourly shape; the electricity supply, its price, and whether it swings with wind and sun; the architecture that matches that duty, alkaline for steady cheap power, PEM for dense dynamic duty, solid oxide where process heat pays; the water and its treatment; and the safety architecture hydrogen demands everywhere. Get those right and stack count, rectifier duty, and balance of plant follow.
Chapter 1 / 06
What a Hydrogen Electrolyzer Is and What Is Being Decided
The core physics is the electrolysis of water, two water molecules plus electrical energy becoming two hydrogen molecules and one oxygen. Thermodynamically the reaction demands a floor near 39 kilowatt hours per kilogram of hydrogen at room temperature, and real machines deliver 47 to 56 in the alkaline class, 50 to 58 in PEM, with balance of plant pushing complete plants toward the upper end of each band; the gap between floor and reality is voltage loss, electrode overpotential, membrane or diaphragm resistance, and gas bubble blanketing, and it is the engineering content of the machine. Every cell runs around 1.8 to 2.2 volts in service, cells stack in series electrically and in parallel hydraulically to make flow, and the stack is the replaceable heart inside a plant of pumps, exchangers, separators, dryers, and power electronics.
The plant around the stack is half the purchase and most of the reliability. Water arrives, is polished to the conductivity the architecture demands, deionized to below 1 microsiemens for PEM and cleaned for alkaline lye, and feeds the circulation loops that carry it past the electrodes and away with product gas. Heat from inefficiency leaves through cooling, a plate heat exchanger against a cooling tower or chilled loop sized on the losses. Gas leaves each electrode side carrying water vapor or lye mist, separates in gas liquid separators, and hydrogen continues to drying, typically dehumidification to a dew point the customer's process sets, measured by the dew point transmitter and moisture analyzer that close the purity loop. Oxygen vents or is captured, and both sides carry the pressure control that keeps the differential across the membrane or diaphragm near zero.
What is being decided, in purchasing terms, is five things. First the demand: kilograms per hour or normal cubic meters per hour, purity down to parts per billion for electronics grade or percent level for combustion, pressure delivered, and the load shape, steady or following a wind farm. Second the power: price per megawatt hour, its availability, and its dynamics, because the architecture's tolerance of swings decides how much of the cheap electricity the plant can actually absorb. Third the architecture itself: alkaline, PEM, or solid oxide, the chapter that follows. Fourth the water and site: quality, quantity, and the cooling that rejects the losses. Fifth the safety and integration: hydrogen detection, ventilation, hazardous areas, and the delivery interface, compression or storage, that follows the plant. The paragraphs that follow work these in order.
A scale tour grounds the machine. A laboratory runs a 1 kilowatt bench stack teaching the chemistry. A metals plant runs a 5 megawatt alkaline train feeding an annealing furnace at 300 normal cubic meters per hour. A bus depot runs containerized PEM plants at 1 to 5 megawatt filling 350 or 700 bar vehicle storage overnight. A refinery runs 20 megawatt and larger trains displacing reformed hydrogen with renewable power. The stacks differ by four orders of magnitude; the voltages per cell barely differ at all, and the purchase arithmetic, electricity price times specific energy, rules every scale.
Chapter 2 / 06
How Cells, Voltage Losses, and Balance of Plant Do the Work
The cell is where the money is spent. Two electrodes, an anode and a cathode, separated by a diaphragm in alkaline machines or a membrane in PEM, with an electrolyte conducting ions between them: hydroxyl ions in potassium hydroxide solution for alkaline, protons through the hydrated polymer membrane for PEM. Push direct current and the reactions run, water oxidized to oxygen at the anode, protons reduced to hydrogen at the cathode, and the cell voltage stacks its demanded thermodynamic potential with overpotential at each electrode, ohmic drops through electrolyte and membrane, and concentration losses as gas bubbles crowd the electrode surfaces. Vendors fight the losses with catalysis, thinner membranes, electrode geometry, and temperature, and the buyer sees the whole fight as one number, volts per cell at the operating current density.
Current density is the intensity dial. More amperes per square centimeter means more hydrogen per square meter of cell area, smaller stacks per megawatt, and higher cell voltage, worse efficiency per kilogram: the trade is capital against electricity, and each architecture sits at its own point. Alkaline machines run 0.2 to 0.6 amperes per square centimeter class with generous electrode area and cheap materials; PEM runs 1 to 3, compact and fast; solid oxide, operating on steam at 700 to 850 deg C, runs high current density with electricity partly replaced by heat. The specification should ask for the efficiency curve across the load range, not a single point, because the plant will live at many points.
Efficiency decomposes into stack and balance of plant. Stack efficiency is the cell voltage story: 1.48 volts corresponds to the higher heating value convention, and every hundredth of a volt per cell is money across a plant's life. Balance of plant is everything electric and thermal around it: rectifier losses, a few percent converting grid ac to stack dc; circulation pumps for lye and water; cooling pumps and fans; controls and instruments; dryers, where thermal swing or pressure swing adsorption spends its own energy; and compression where the delivery pressure demands it. A 52 kilowatt hour per kilogram stack inside a 56 kilowatt hour per kilogram plant tells the buyer exactly where the next improvement lives, and the specification should demand both numbers separately.
Water is the feedstock and its quality is a warranty condition. PEM wants deionized water below 1 microsiemens per centimeter, delivered by ion exchange or reverse osmosis polishing, with a conductivity meter guarding the inlet continuously; alkaline wants demineralized make up into its potassium hydroxide loop, and the lye itself is managed, filtered, and occasionally replaced as carbonate builds from carbon dioxide ingress. Consumption is stoichiometric plus losses, roughly 9 to 10 liters of purified water per kilogram of hydrogen before cooling tower make up, and the water treatment train, its regeneration waste, and its reliability belong inside the plant's specification rather than as an afterthought on the site utilities list.
Safety architecture is not an appendix because hydrogen is unforgiving: wide flammable range, low ignition energy, invisible flame, and buoyant leakage into roof spaces. The plant design ventilates enclosures, detects leakage with a gas detector network alarm before the lower explosive limit's action levels, grounds and bonds everything electrical, and manages the hydrogen oxygen interface in the separator logic with the differential pressure control that keeps the gases apart. Relief paths, a safety relief valve on every pressured volume and flame arrester where vent lines could carry a flame back, are designed in, and hazardous area classification follows the leak points. The safety case is a document the buyer should ask to see, and the vendor's answers about it separate the serious from the assembling.
The separator and loop logic deserves one paragraph of its own because it is where the two product gases live closest to each other. Each side of the stack discharges its gas carrying liquid, into a separator whose level transmitter controls the return pump or valve, and the two levels, plus the differential pressure transmitter comparing the gas headers, are the instrumentation that keeps the membrane or diaphragm unloaded: hydrogen at a few millibar above oxygen on PEM, the reverse on some alkaline logics, always small and always deliberate. The controls act on valve trim and a make up line, and their response time is part of the safety case, because a level or pressure upset that crosses the interface is the failure mode this whole architecture exists to prevent. Buyers should ask vendors to walk them through that logic loop on a drawing, and a vendor who cannot draw it from memory has not owned it.
Chapter 3 / 06
Alkaline, PEM, and Solid Oxide Designs and Duty Boundaries
The alkaline electrolyzer is the mature architecture, a century of industrial service, and it leads on cost per megawatt and durability. Nickel based electrodes in potassium hydroxide solution, separated by a composite diaphragm, assembled into large stacks that run near atmospheric pressure or pressurized to 30 bar class; lye circulates through both sides, gas separators disengage hydrogen and oxygen from the liquor, and the plant tolerates impurities and rough water better than its rivals. Its historical weakness is dynamics: the liquid electrolyte and gas blanket behavior made old designs slow to follow load, though modern controls have narrowed that gap to minutes class turndown. It wins on steady, cheap, large electricity.
The PEM electrolyzer trades precious metal catalysts and titanium components for intensity and agility. A proton conducting polymer membrane, iridium and platinum catalyzed electrodes, compact stacks running at 1 to 3 amperes per square centimeter, differential pressure operation that can deliver hydrogen at 30 bar and above directly from the cathode, and response to load changes in seconds, which makes PEM the architecture that follows solar and wind curves naturally. Its costs are the metals, the water discipline, and a shorter stack life record than alkaline's, improving yearly but still the line item to interrogate in any bid.
The solid oxide electrolyzer, soec, runs the same oxide ion chemistry as its fuel cell sibling in reverse, electrolyzing steam at 700 to 850 deg C with electricity plus process heat. Where that heat exists, industrial waste heat, nuclear, or concentrated solar, the electricity demand per kilogram drops meaningfully below the low temperature architectures, and the machine suits hydrogen or syngas duties bolted to a heat source. Its constraints are the ceramics' thermal cycle tolerance, materials life at temperature, and integration complexity, and it is the architecture of integrated energy projects rather than container installs.
Duty boundaries against the neighbors deserve plain language. This machine is not the oxygen generator or nitrogen generator of the compressed gas trade: those separate air by pressure swing adsorption or membranes and produce no hydrogen at all, and a site that wants hydrogen has no use for them. It is not a fuel cell, which runs the same electrochemistry in reverse to make electricity from hydrogen, though the families share components and vocabulary. It is not a steam methane reformer, which makes hydrogen from natural gas with carbon emissions the electrolyzer avoids at the cost of electricity. And delivered hydrogen, tube trailer or liquid, competes with it economically below some utilization threshold: the on site plant wins when demand and power price are steady enough to load its capital, and the delivered molecule wins below perhaps 30 to 40 percent utilization, a boundary each site must run its own numbers on.
Precious metal catalysts; strict water quality; stack cost
Solid oxide (soec)
Below 40 kWh/kg electric with heat
Highest efficiency where process heat exists
Thermal cycling limits; integration complexity
Chapter 4 / 06
Key Specification Parameters
The specification is built from the hydrogen demand outward. The process set states hydrogen flow in normal cubic meters per hour or kilograms per day, purity with its oxygen, water, and nitrogen allowances, delivery pressure, and the load profile with minimum turndown and ramp rate. The performance set states specific energy at rated and at part loads, guaranteed at stack and plant boundaries, plus stack life or degradation rate, the percent per thousand hours the voltage, and so the efficiency, will drift. The design set states architecture, stack count and module arrangement, rectifier and transformer arrangement, cooling duty and medium, water consumption and quality, and the drying technology. The safety set states hazardous area classification, detection and ventilation philosophy, and relief and flame arrest provisions. The commercial set states stack replacement pricing and the service model.
Parameter
Typical range
Why it matters
Specific energy, plant
47 to 58 kWh/kg by architecture
Times power price times hours, the lifecycle number
Turndown
20 to 100 percent class, pem deeper
How much cheap swing power the plant absorbs
Ramp rate
Seconds in pem, minutes in alkaline
Following wind and solar versus steady grid
Delivery pressure
Atmospheric to 30 bar plus
Compression stages and energy avoided or added
Hydrogen purity after drying
99.9 to 99.999 percent class
Set by the customer's process, fuel cell strictest
Water consumption
9 to 10 L/kg plus cooling make up
Feed train sizing and site water budget
Stack degradation
0.1 to 1 percent per 1000 h
Efficiency drift and replacement timing
Footprint
Containerized to building scale
Site integration and plot
Two lines in that table decide the project's economics. Specific energy, quoted at the plant boundary with the dryers and cooling honestly inside it, because power price times that number times operating hours is most of the cost of the hydrogen ever made here; a bid that quotes stack efficiency and hides the balance of plant loads is quoting the engine of a car and omitting the wheels. Stack degradation and replacement price, because electrolyzer economics run 10 and 20 years: a stack that drifts a percent per thousand hours and costs half the plant to replace turns a good bid into a poor one mid life, and the guarantee should state both the drift and the price.
Instrumentation closes the machine: conductivity meter chains on feed water, pressure transmitters on both gas sides with differential control, temperature transmitters on stack inlet and outlet, hydrogen purity analysis, oxygen in hydrogen and vice versa, with the dew point transmitter on dried product, gas detector coverage in every enclosed space, and a flow meter on hydrogen out and another on feed water in. The plc runs the load following, the differential pressure protection, and the safety matrix, and its logs are the plant's degradation record. Remote monitoring is standard in this market and the vendor's data habit matters: the plant that trends its own voltages per cell sees its stack's health story years early.
Chapter 5 / 06
Selection Decision Factors and RFQ Language
Selection starts with the load shape and the power price, together. A steady industrial demand at a flat contracted power price suits alkaline economics: lowest capital per megawatt, efficiency adequate, stack life long. A demand that follows renewable generation, or power bought hour by hour from a market that pays the plant to be flexible, suits PEM's seconds class response and deep turndown. A site with real process heat, steel, chemicals, nuclear, should price solid oxide before assuming the low temperature default. The utilization arithmetic, annual kilograms divided by rated capacity, decides whether an on site plant beats delivered hydrogen at all, and it should be the first calculation in the file, not the last.
Purity and pressure then trim the architecture. Fuel cell vehicle fuel demands the strictest water and sulfur specifications and favors PEM's inherently pure cathode product with polishing; industrial combustion and annealing accept simpler trains; electronics grade adds palladium or getter polishing beyond the electrolyzer itself. Delivery pressure interacts with storage: 30 bar direct from the stack feeds moderate storage cheaply, while 350 and 700 bar vehicle filling adds diaphragm compression stages whose energy and maintenance the project must own, and whose interstage cooling loads the same cooling tower as the stack.
Water and site close the physical selection. Ten liters per kilogram purified, plus cooling make up that can double the raw draw in warm climates, needs a water treatment train, ion exchange or reverse osmosis polishing, sized with its regeneration waste plan; the cooling system, cooling tower or dry cooler by climate, rejects the majority of the input power as heat and its approach temperature shows up in efficiency; and the plot wants ventilation, hazardous area discipline, and distance decisions the safety case makes explicit. None of this is exotic, and all of it is under-specified in more bids than not.
A complete RFQ states, on one process data page: hydrogen demand and its hourly profile with minimum and peak; purity specification with each impurity allowance; delivery pressure and storage arrangement; power supply, voltage, capacity, price, and its dynamics; architecture preference or openness; required specific energy at plant boundary, rated and part load; stack life and degradation guarantee with replacement pricing; water available and its analysis; cooling medium and conditions; indoor or outdoor, hazardous area requirements; oxygen venting or capture; and the applicable hydrogen safety practice and permitting regime. Vendors quoting against that page quote the same plant, and their energies, degradation rates, and stack replacement prices become comparable line items.
Integration scope belongs in the same decision boundary. Compression, storage, and dispensing or process delivery follow the electrolyzer and can dominate the plot; the power electronics interface, transformer, rectifier, and their harmonic footprint on the site grid, is part of the electrical package; and the control integration, the plant's plc following an external setpoint from the renewable asset or the furnace demand, is what makes the hydrogen actually arrive when wanted. Buying the stack without these interfaces defined is buying the engine of a delivery truck, and the project that splits the scope should split it deliberately, with the interfaces written.
The acceptance test deserves negotiation at purchase because specific energy is the product. Run at agreed load points across the range, with water quality, purity, and ambient conditions recorded, the test measures kilowatt hours per kilogram at the plant boundary against the guarantee curve, plus purity after drying, delivery pressure, and turndown demonstrated down to its guaranteed floor. Ambient matters and should be stated: cooling approach temperature moves the efficiency a measurable percent between winter and summer, and a guarantee quoted at 15 deg C cooling water in July is not the same machine as one quoted at 25. A well written test converts the vendor's efficiency claims from brochure to contract, and every serious bidder will accept it.
Chapter 6 / 06
Installation, Standards, Failure Modes, and Maintenance
Installation is power plant and gas plant work together: heavy rectifiers and transformers set and terminated, stacks and skids placed with their gas and water tie-ins, ventilation and detection commissioned, and the safety systems proven, leak testing with inert gas before hydrogen ever arrives, detector alarms walked, and relief paths verified. Commissioning purges, replaces air with nitrogen and nitrogen with hydrogen in ordered sequence, brings stacks up on water and lye at low current, and steps load while cell voltages, differential pressures, and purities are recorded. That first weeks' data is the baseline the degradation story is told against for the next decade.
The governing references are the hydrogen safety and installation practices, national fire and electrical codes applied to hydrogen service, pressure vessel and piping codes for the gas boundaries, and the grid codes the power electronics answers to; the US Department of Energy's hydrogen production and electrolysis documentation represents the technical basis and the technology assessments the market quotes, and vendors like Nel publish the applied plant practice for alkaline and PEM machines. Certification pressure is rising, and the buyer's job is to demand the safety case and the code mapping in the bid rather than after delivery.
Failure modes are chemical, electrical, and mechanical. Voltage drift rising across the stack is degradation, catalyst wear or membrane thinning in PEM, or lye carbonate and electrode changes in alkaline, and its rate against the guarantee is the warranty conversation, best held with the per cell trend data both parties kept. Differential pressure excursions threaten the membrane or diaphragm, and their controls, the pressure transmitters and the valve logic, are the protection. Water quality upsets poison PEM silently until the conductivity meter that should have alarmed is found bypassed. Dryer breakthrough shows as dew point drift and is answered by regeneration or desiccant, not by hope. Leakage, anywhere, is the emergency the gas detector network exists to catch early, and the ventilation and shutdown matrix is rehearsed, not just documented.
Symptom
Likely cause
First action
Stack voltage drifting up
Degradation per design or accelerated
Trend per cell; confirm water quality history; review load and trip history with vendor
Verify dp transmitters and valve response; check separator levels
Product dew point drifting
Dryer regeneration fault or desiccant exhaustion
Verify dryer cycle and heat; test check valves; replace desiccant
Oxygen in hydrogen rising
Membrane or diaphragm wear
Reduce load if beyond trip; schedule inspection; review warranty terms
Feed conductivity alarms
Polishing train exhaustion or bypass
Regenerate or replace resins; verify meter calibration; halt feed if confirmed
Routine maintenance is water, gas, and power discipline: polishing train regenerations on schedule, lye filtration and analysis on alkaline, dryer desiccant and regeneration system care, calibration of the conductivity, purity, and dew point instruments, thermographic review of power connections, and the detector network's bump tests. Stack replacement, planned at the degradation curve's end rather than forced by failure, is the defining outage, and its contract price was set at purchase because that was when the buyer had leverage. Plants that trend their cell voltages schedule it years out, which is the only comfortable way to buy a replacement heart.
The institutional item is power price review against operating plan, because this machine is first an electricity customer. Half yearly, the plant's actual specific energy and hours should meet the electricity contract that justified the project: markets move, renewable profiles shift as more solar joins the grid, and the plant that replans its dispatch twice a year captures the cheap hours the original model only hoped for. The electrolyzer is the most fungible load many sites will ever own, and treating its dispatch as an operating decision rather than a fixed habit is worth real money.
Training and spares close the commercial reality of these plants. Carry instrument spares for the critical loop, conductivity cells, pressure and differential transmitters, purity sensor cells, plus valve trim kits for the pressure control functions, dryer desiccant, and the polishing train's resins; on alkaline add lye filters and analysis support. Training should reach two audiences, the operators who will run load following and water discipline daily, and the maintenance crew who will own the instruments and the power connections, and the vendor's commissioning period is the cheapest classroom the project will ever book. The market is young enough that the service organization's depth, parts warehouses, response contracts, remote monitoring desks, differs more between vendors than the stacks do.
FAQ
Which architecture should I buy, alkaline, PEM, or solid oxide?
Match it to your power and heat: steady cheap power and large scale favor alkaline on capital and durability; dynamic renewable following and compact sites favor PEM's fast response and pressure capability; genuine process heat availability favors solid oxide's electric efficiency. The utilization arithmetic against delivered hydrogen comes first, and the architecture choice follows the power profile.
How much electricity does a kilogram of hydrogen take?
Complete plants deliver roughly 47 to 58 kilowatt hours per kilogram today, alkaline at the lower end, PEM toward the upper, against a thermodynamic floor near 39. The number that matters is at the plant boundary with dryers and cooling included, quoted across the load range, because that is the electricity bill the hydrogen carries.
Is an electrolyzer the same machine as an oxygen or nitrogen generator?
No, they only share the gas skid aesthetic. Oxygen and nitrogen generators separate air by adsorption or membranes; the electrolyzer splits water with direct current and makes hydrogen that air separation cannot. A site wanting hydrogen has no use for the air machines, and a site wanting only oxygen or nitrogen should not buy electrolysis for it.
How pure is electrolytic hydrogen, and what polishing does it need?
Alkaline product carries oxygen and water in the parts per thousand to percent class cleaned by deoxo and dryers; PEM cathode product is inherently purer and typically needs drying only. Fuel cell fuel, 99.97 percent class with strict water and sulfur limits, adds polishing stages; combustion and annealing duty accepts the simpler train. Specify the impurity allowances, not a bare percentage.
What water does the plant need?
PEM demands deionized feed below about 1 microsiemens per centimeter from ion exchange or reverse osmosis polishing; alkaline needs demineralized make up into its potassium hydroxide loop. Consumption runs 9 to 10 liters per kilogram of hydrogen before cooling make up, and the treatment train, its wastes, and its reliability are part of the plant, with a conductivity meter guarding the inlet as a warranty condition.
How do plants handle the oxygen byproduct?
Most vent it after demisting and pressure letdown; capture for wastewater, medical, or combustion duty is straightforward engineering, compression or psa, that some sites justify economically. The separator and differential pressure logic that keeps hydrogen out of the oxygen stream is safety critical in either case, and the vent line carries its flame arrester where codes require.
When does on site electrolysis beat delivered hydrogen?
When utilization and power price load the capital: steady demand above roughly 30 to 40 percent utilization at a competitive power price pays the plant out against tube trailers; intermittent or tiny demands pay the delivered molecule's premium forever. Run your own numbers on your power contract before believing either side's brochure.