PSAvsVSA vs Membrane Oxygen Generation:Which Technology Fits Your Application?

On-site oxygen generation is now a genuine cost decision, not a theoretical one. US bulk liquid oxygen averaged USD 258 per metric ton in March 2026 and USD 263 per metric ton in June 2026, with Canadian bulk LOX at USD 230 and USD 233 per metric ton over the same months, according to IMARC's bulk liquid oxygen price data for the first half of 2026. Against a delivered price that moves with the market, the case for generating on site gets stronger every quarter.

The catch is that no single technology wins. PSA, VSA and membrane systems all sit on the same purity-versus-flow plane, and each one trades purity against flow to hold its position on it, which is why the three technologies compete at all rather than one simply replacing the others.

This PSA vs VSA vs membrane oxygen generation comparison is research-only. It draws on published benchmarks, OEM datasheets and standards bodies, and claims no first-hand test data. The sections that follow evaluate all three on purity, flow and scale, energy and total cost of ownership, footprint and maintenance, and feed-air and certification readiness.

Summary comparison: PSA vs VSA vs membrane oxygen generation at a glance

How to read the comparison:

The performance ranges shown above should be treated as indicative engineering ranges rather than guaranteed values. Actual oxygen purity, recovery, specific energy consumption and operating cost vary with feed-air conditions, oxygen flow, delivery pressure, ambient temperature, adsorption media or membrane characteristics, operating cycle and system configuration.

Where a value is derived from a manufacturer's published specification, it should not be interpreted as an independently verified Sollant test result. Where we provide Sollant-specific data, the applicable model, operating conditions and measurement method should be stated.

PSA delivers 90 to 95 vol.% oxygen in 18 sizes from 1.5 to 197.4 Nm³/h (published PSA oxygen specifications and cycle data, retrieved 2026). VPSA reaches 700 to 10,000 Nm³/h single-train, equal to 25 to 345 tpd, at up to 94 vol.% (large-scale VPSA plant capacities, assets dated 2024-2025), and 80 to 400 Nm³/h at 93% in smaller frames (VPSA units in the 80 to 400 Nm³/h band, retrieved 2026). VPSA runs at 0.28 to 0.38 kWh/Nm³ against PSA's 0.45 to 0.60 (benchmark energy use per cubic metre of oxygen, site updated August 2026). Membrane output is an enrichment band, not a setpoint, because O₂/N₂ selectivity in practical polymers is modest (measured O2/N2 selectivity in hollow-fiber membranes, 2020-08).

Decision rule: choose PSA below roughly 100 Nm³/h, VPSA above it, and membrane only when an enrichment band is enough.

Read every figure as a range. Feed-air quality, altitude, duty cycle and cycle design all move these numbers, so supplier headline figures are not comparable unless the test conditions are named.

What actually separates these three technologies?

All three take the same feed air and separate it differently, and that difference sets every downstream trade-off. Oxygen PSA passes air through a zeolite-type molecular sieve that preferentially adsorbs nitrogen under pressure, leaving oxygen in the product stream; reducing pressure regenerates the bed (Atlas Copco, updated 2026-07-07). A pressure swing adsorption oxygen generator therefore produces oxygen by holding nitrogen back, not by pulling oxygen forward. The common confusion is carbon molecular sieve, which separates by kinetic diffusion, oxygen entering the pores faster than nitrogen. That is the classic nitrogen-PSA mechanism, a different air-separation mode, so any "CMS bed makes oxygen" phrasing is wrong.

VSA and VPSA run the same adsorption chemistry but regenerate on the vacuum side, which changes the energy and scale profile. Membranes work on an entirely different principle: solution-diffusion through hollow fibers, where oxygen is more permeable than nitrogen, so the permeate is oxygen-enriched air, not purified oxygen.

That single fact explains why the three technologies compete at all. In any PSA vs VSA vs membrane oxygen generation comparison, they occupy different boxes of the same purity-versus-flow plane, and each one trades purity against flow. For PSA, cycle timing sets the purity ceiling: longer adsorption phases yield higher flow but lower purity, while shorter cycles prioritize purity over volume, and higher purity also needs more compressed air and a larger footprint (Atlas Copco, updated 2026-07-07).

Before quoting anything, a distributor needs five inputs from the customer: required purity band, required flow in Nm³/h and SCFM, working pressure, altitude, and duty cycle. Those five numbers decide which technology is even in the running.

Pro Tip: A purity figure without its sampling point and test conditions is a marketing number, not a specification. Apply that test to every datasheet you receive, including the ones you forward.

Purity: what can each technology actually deliver?

PSA and VPSA both reach the 90 to 95 vol.% band, membrane does not, and nothing short of a downstream purifier breaks the 95% ceiling.

Important note on oxygen purity data

Oxygen purity is not a fixed property of a PSA, VSA or membrane system. Actual purity depends on feed-air conditions, flow rate, operating pressure, cycle settings, adsorbent or membrane characteristics and system configuration.

The ranges in the chart are therefore intended for technology-level comparison, not as a guaranteed output specification for every system.

PSA purity is selectable rather than fixed. One published PSA specification offers grades at 90, 92, 93, 94 and 95%, and flow falls as the grade rises: 197.4 Nm³/h at 90% down to 179.4 Nm³/h at 95% (PSA purity grades and the flow they cost, 2025). VPSA runs to a similar ceiling, up to 94 vol.% on the reference data GASGEN publishes (updated August 2026).

The ceiling is a molecular fact, and it belongs to the zeolite mechanism. Under pressure the sieve's zeolite preferentially adsorbs nitrogen, and oxygen is left in the product gas (the molecular-size basis of sieve separation, updated 7 July 2026). Argon is close enough to oxygen in size and adsorption behaviour that the sieve cannot separate the two economically.

Why 95% is a wall, not a setting: pushing past 95% means adding a downstream purifier, not turning the unit up.

Membrane oxygen separation purity behaves differently. Single-stage O2/N2 selectivity is modest, so the product is an enrichment band rather than a setpoint you dial in, and feed pressure, temperature and stage count move the result more than any control setting does. That is not a defect where the application wants enrichment; it is a disqualification where the application wants a grade.

The application sets the useful filter. Wastewater aeration and aquaculture run at 90 to 93%, glass oxy-fuel and steel EAF enrichment at 90 to 95%, and ozone generation at 90 to 95% (minimum oxygen purity by application, updated August 2026).

Medical oxygen settles the membrane question outright. WHO states that PSA medical oxygen is 93% plus or minus 3%, while oxygen produced by air liquefaction for medical use must be not less than 99.5% (WHO's medical oxygen purity requirement, 2020, standards page still current). For medical oxygen applications, users should always confirm compliance with local regulations and applicable standards before selecting equipment. Oxygen generator specifications alone do not determine regulatory approval.A membrane system's enrichment band sits far below 93% and is therefore disqualified from medical use, and a PSA unit cannot simply be pushed to 99.5% without a downstream purifier.

The cost side of purity is mechanical, not commercial. Longer adsorption phases yield higher flow at lower purity; higher purity needs more compressed air and a larger footprint to process it.

Flow and scale: where does each technology stop being economic?

The crossover sits at roughly 100 Nm³/h. Above it, a vacuum swing adsorption oxygen system saves 30 to 50% of the energy a PSA unit would burn for the same output; below it, PSA's mechanical simplicity wins outright (GASGEN oxygen generator energy benchmark, updated August 2026).

The published envelopes map that split cleanly. PSA generators run from 1.5 to 197.4 Nm³/h (OXYBERG), while VPSA units in the 80 to 400 Nm³/h band cover 47 to 235 scfm, or 2.5 to 12.6 tons per day (Atlas Copco OGV+). Large-scale VPSA plant capacities reach 700 to 10,000 Nm³/h in a single train, equal to 25 to 345 tons per day of 100% O₂ (Linde Engineering).

Why does PSA stop scaling smoothly? Each tower completes a cycle every 60 to 120 seconds, and a downstream buffer tank converts that pulsed bed output into continuous flow (OXYBERG). Turndown is therefore bounded by cycle timing and buffer volume, not by demand alone.

That envelope explains the demand map behind large VPSA plants: the industry segments they serve include steel and non-ferrous metals, glass, pulp and paper, mining, combustion and chemicals, lithium battery cathode roasting, wastewater treatment and cement oxyfuel (Linde Engineering).

Two honest caveats. Below the crossover, a small VPSA carries infrastructure the application does not need. At the very top of the range, cryogenic separation remains the incumbent.

Energy and total cost of ownership: which is cheaper to run?

VPSA is the lowest-energy option at scale and PSA the lowest at small scale, so on-site oxygen generation cost depends on where you sit on the flow axis.

Those are benchmark energy use per cubic metre of oxygen from GASGEN's technology comparison, updated August 2026. Purity carries its own cost: 90% to 93% adds roughly 5-8% energy, and 93% to 95% another 10-15%, so the energy penalty of each purity step compounds.

Sieve life is the second line item. Nitroquanta's CMS replacement guidance (2026-05-28) gives 5 to 10 years under ideal conditions, 3 to 4 years with contaminated air or operation at upper temperature limits, and 12 years or more from a single charge with excellent pretreatment. Hydrocarbon and oil aerosol pore blockage, moisture, mechanical attrition and dusting, and thermal cycling drive the difference, so realistic sieve replacement interval is a pretreatment question, not a vendor question.

Against delivered gas, IMARC's bulk liquid oxygen price for the first half of 2026 ran USD 258 per metric ton in March and USD 263/MT in June in the USA, and USD 230/MT and USD 233/MT in Canada. What this research window could not re-verify on a live page is the demand and break-even figures for on-site generation versus delivered LOX, so no payback period is quoted here.

Measurement rule: log actual kWh over one week and divide by total Nm³ produced, compared at your purity level. That number, not a datasheet, is what procurement should sign.

Footprint, maintenance and feed-air readiness: what does each demand on site?

The technology that wins on footprint and maintenance is the one whose feed-air pretreatment is correctly specified, because inadequate pretreatment shortens sieve life and voids performance claims across all three. The feed-air chain each technology requires is the real constraint: PSA needs clean, dry compressed air from a compressor, dryer and filtration train, with coalescing and oil-vapour filters depending on compressor type, commonly specified to an ISO 8573-1 class (Atlas Copco, no page date). VPSA units eliminate the separate dryer because an integral drying layer removes moisture and CO2 ahead of the adsorption layer, so low-pressure blower systems (300 to 1000 mBar, or 4 to 15 psi) carry a lighter balance of plant.

Maintenance profiles differ by mechanism. A pressure swing adsorption oxygen generator cycles its beds roughly every 60 to 120 seconds, so output is pulsed and needs a buffer tank, and cycling-driven sieve attrition is a genuine recurring cost. Membrane modules have no cycling bed to attrit, while VPSA adds blower and vacuum-side service load instead.

Certificate, not brochure: ask for the test report naming the exact model, not a generic datasheet.

For a distributor qualifying a supplier, demand CE marking with a PED 2014/68/EU declaration, ISO 9001 as the general quality certificate, ISO 13485 where the unit is sold for medical use, and third-party performance test reports naming the exact model for purity, flow and pressure from an accredited or notified body. Add ISO 8573-1 compressed-air test evidence for the feed-air side, with the WHO and UNDP medical oxygen plant specifications as reference documents. One option in the PSA-favoured zone below the ~100 Nm³/h crossover is Sollant Treatment's SLTZO-1MZ to SLTZO-30MZ range, covering 1 to 30 Nm³/h at 93% ± 3%, which supports internal air distribution, CMS selection per working conditions and a spring- or cylinder-compression device that compensates as the sieve settles.

Who should choose which technology?

If you need 90–93% oxygen at moderate flow, choose PSA or a small VPSA unit. If you need 90–95% at high continuous flow, choose VPSA. If you need medical oxygen at 93% ± 3%, choose PSA. If you want enrichment rather than a purity setpoint at large flow, choose membrane.

Wastewater or aquaculture operators, 90–93% at moderate flow: PSA is usually the fit, and small VPSA becomes competitive as flow rises toward the crossover.

Glass, steel or cement plants, 90–95% at high continuous flow: VPSA wins on the flow threshold where VPSA overtakes PSA, roughly 100 Nm³/h, above which it saves 30–50% of energy.

Medical oxygen buyers, 93% ± 3% under ISO 7396: PSA meets WHO's medical oxygen purity requirement; liquefaction-grade supply is held to 99.5%.

Enrichment at large flow: Membrane suits applications that need an enrichment band rather than a purity setpoint, because measured O2/N2 selectivity in hollow-fiber membranes sets an output band, not a fixed point. Membrane oxygen separation purity is therefore not a medical-grade oxygen source.

Below PSA's small end or above VPSA's single-train envelope, the answer is delivered liquid oxygen or cryogenic on-site generation.

Sollant develops and manufactures compressed-air treatment equipment and gas generators, including oxygen and nitrogen generation systems. If you are sizing a system for a specific customer, request a sizing review and we will work through the flow, purity and duty-cycle numbers with you.

Conclusion

The PSA vs VSA vs membrane oxygen generation decision comes down to four variables, not a single best technology: required purity band, scale, duty cycle and site altitude. Membrane systems are the simplest to install and the least demanding to maintain, but they cap out at lower purity and struggle when high flow is also required. PSA covers the widest practical range for small-to-medium demand, while VPSA becomes the more economical choice once continuous demand climbs past roughly 100 Nm³/h. That threshold is where the pressure swing adsorption oxygen generator gives way to vacuum swing adsorption on running cost, and it is the single number most worth checking before you quote.

For a distributor, the next step is not a technology pitch but a data collection exercise. Gather five inputs from the customer: peak and average flow, required purity, daily duty cycle, site altitude and available feed-air quality. With those in hand, request a sizing review against the actual operating conditions rather than a catalogue headline figure. Ask the supplier for the sieve specification, the stated specific energy consumption at your customer's duty point, and the certification documents that apply to the intended end use. A supplier who can answer all three in writing is a supplier you can put in front of a hospital or a food processor with confidence.

For a fuller side-by-side of the three technologies across purity, scale and cost, see our oxygen generation technology comparison.