Your new equipment needs three-phase power and the building does not have it

This is one of the few electrical questions where two competent people give opposite answers and neither is wrong. The dealer who sold you the machine has watched hundreds of shops run that model behind a phase converter. The electrician looking at your panel has watched converters fail in shops that needed real three-phase power. They are describing different buildings.

What separates those buildings is not opinion. It is a short list of facts about the load: how many machines, how hard they work, how much imbalance the equipment tolerates, and whether your existing service can carry the converter at all. Work through those and the answer stops being a matter of who you trust.

Before anything else, confirm what the building already has, because a surprising number of older commercial buildings in Santa Clara County are already fed three-phase and nobody on site knows it. The field check (a 2-pole versus a 3-pole main, and what you read between hot conductors) and the voltage systems those readings point to are on our three-phase electrical service page. If you already have it, everything below is moot and you need a circuit, not a project.

Assuming you do not, there are four real paths, and one thing that gets sold as a fifth:

A variable frequency drive. Single-phase in, three-phase out, for one motor. The cheapest answer when it applies, and the one most often missed. A rotary phase converter. A spinning idler motor manufactures a third leg, and a whole subpanel of machines can run behind it. A solid-state converter. Same job, electronics instead of a rotor, much tighter output, several times the price. Utility three-phase. A real service, with a real application, a real allowance calculation, and a calendar measured in months.

The fifth thing is the static converter, and it deserves its own warning before you see the price.

Find your situation in the left column. Where two rows describe you, the lower one wins, because these are thresholds and not preferences.

This is the answer that gets skipped, usually because the conversation started at the phrase "phase converter" and never left it. If the new machine is one motor, and you do not need other three-phase machines running at the same time, a variable frequency drive takes your single-phase supply and produces three-phase output for that motor. It is normally the cheapest path, and unlike a converter it gives you something you did not have before: soft start instead of a full inrush hit, and speed control on a machine that had one speed.

The commodity ceiling for a drive with a native single-phase input is 3 HP at 230V. That covers the AutomationDirect GS10, GS20, and GS30 lines, the Fuji FRENIC-Mini C2, the Lenze SMV, and the Hitachi WJ200 single-phase models. Under $300 buys a 3 HP unit.

Above 3 HP you will read everywhere that no such thing exists. That is folklore. Invertek publishes single-phase-input Optidrive E3 models up to 7.5 kW, which is 10 HP, in specific configurations, and Yaskawa publishes single-phase input tables for the HV600 reaching 40 HP at 240V and 75 HP at 480V. The Yaskawa caveat is real and worth stating: the HV600 is an HVAC variable-torque drive, not a machine-tool drive, so those tables are not a green light for a lathe. But "nobody makes one above 5 HP" is simply not true, and it costs shops money.

The path most often used from 5 to 10 HP is different: derating a three-phase-input drive onto single-phase supply. The physics is that a six-pulse rectifier fed from one phase sees 120 Hz ripple on the DC bus instead of 360 Hz, which stresses the rectifier and the bus capacitors, so the drive is rated down. Several manufacturers permit it in writing. Invertek allows it at 50 percent of rated output current with the supply on L1 and L2. Fuji publishes UL-approved single-phase rating tables per drive family, with the supply on L1 and L3, the input phase-loss detection turned off, and a DC reactor strongly recommended on some families. AutomationDirect simply prints the derate in the product title, so its 230V GS20 models are sold as, for example, 10 hp on three-phase input and 5 hp on single-phase input.

Three things to carry out of that. The supply terminals differ between manufacturers, so the only correct instruction is to follow the manual for the drive you bought. At least one manufacturer manual contains both a warning against single-phase operation and a conditional allowance for larger models, which is a good reason to get the answer from the distributor in writing rather than from a forum. And the folk rule of "just double the horsepower" is explicitly rejected by at least one manufacturer: the correct method is to read the published single-phase current table and match it to your motor full-load amps.

Then the limitation that ends this path, and it is the one that decides most shops. A drive is a one-motor device. Drive manufacturers are consistent and explicit that a drive output must never feed non-motor loads, power factor capacitors, or a building distribution network, and that contactors, relays, and disconnects do not belong between the drive and the motor in normal operation. Running several motors from one drive is possible but constrained: each motor needs its own thermal overload protection, the drive runs in volts-per-hertz mode rather than any vector mode, and at least one major manufacturer recommends against line starting and stopping motors while the drive is running. That last constraint describes the ordinary multi-machine shop exactly. So a drive cannot power a shop bus. That is converter or utility territory.

One code point that quietly saves money: a drive lives under Article 430 with the rest of motor circuits, while a phase converter lives under Article 455 and its own sizing regime. For the same motor, the drive path lands on materially smaller conductors and a smaller breaker.

A rotary phase converter is an induction machine, called an idler, running unloaded. The spinning rotor induces the third leg, and your two utility legs plus that manufactured leg make three-phase power for a panel full of machines. It is not a motor-generator set, and it does not change voltage: 230 volts single-phase in gives 230 volts three-phase out. If your machine wants 460 volts, you need a step-up transformer or a unit built for 460.

The behavior that causes the most wasted service calls: on a 230V rotary the manufactured leg reads 10 to 15 percent high at no load by design, so 260 to 290 volts on an idling converter is correct, not a fault. As load comes on it settles into 2 to 5 percent balance. Line-to-ground readings tell you nothing useful about a converter at all. Measure phase to phase, under the load you actually run.

Sizing is where the marketing and the engineering diverge. There is no single multiplier. One manufacturer publishes a five-tier scheme running from easy loads (one size up) to very hard loads (2.5 times), with CNC at twice the load amps. Another bakes roughly 1.5 to 1.67 times into its own published "largest motor HP" ratings, so its numbers are not comparable to a competitor's at the same nameplate. Rigid tapping and instant reversing carry a 2 times minimum. Compressors and submersible pumps run 2.5 to 3 times. As a working floor: 1.5 times for general shop work, 2 times for CNC, 2.5 times for compressors.

The vendors are candid about this once you read past the product page. American Rotary states that a typical 10 hp phase converter can power loads up to approximately 5 hp. Phoenix publishes a per-model load-class table where its 10 HP unit is a 5 HP unit for CNC spindle and other hard loads. Which is the single most useful buying instruction in this guide: compare converters in usable horsepower for your load class, never nameplate to nameplate.

Idle draw is real operating money, and the published figures conflict by a factor of two to four, so treat any single number with suspicion. Phase-A-Matic publishes 10 amps of idle current for its R-20, which is roughly 2.3 kVA. Phase Technologies claims 1,350 watts for a 20 HP rotary. Phoenix describes it as 2 to 5 percent of rated horsepower. A solid-state unit sits at 80 watts on standby. On a load that runs around the clock, that standby gap is the entire economic argument for the expensive unit. On an intermittent shop load, you switch the rotary off at night and the argument mostly evaporates.

On service life, vendors claim 20 to 30 years and more, and the wear item is the capacitors, where the published intervals disagree sharply: one manufacturer says replace every one to three years, another says ten to fifteen, and a third warrants capacitors for one year, which tells you something about where the honest number sits. Two installation details from the manuals are worth knowing before somebody bolts your converter to a slab: mount the unit on rubber isolators and do not rigidly fasten the idler, or you will kill bearings, and if the converter does not come up to speed within about two seconds, shut it off rather than letting it labor.

A solid-state converter does the same job with no rotating parts. The published specifications are a different class of machine: 2 percent or better output voltage imbalance guaranteed, 95 to 98.7 percent efficiency, sinusoidal output to IEEE 519, near unity power factor, 80 watts on standby, regeneration on overhauling loads such as elevators, and voltage-doubling models that take 230V single phase directly to 460V three phase, which removes a step-up transformer and its losses from the project. The price premium at the same nameplate runs two to four times, partly offset because you can often buy a smaller one: the manufacturer's own scope comparison put a 10 HP solid-state unit against a 20 HP rotary marketed for the same blower.

Every shop that has run a converter for a while has the same story: the old lathe and the drill press have been fine for years, then the new machine went in and it throws faults nobody can explain. That is not bad luck and it is not a defective converter. It is one number, and it is small enough that people dismiss it.

Voltage unbalance is defined as 100 times the largest deviation of any phase from the average of the three, divided by that average. So a set of readings that looks almost identical produces a low single-digit percentage. The reason that matters out of proportion to its size is the multiplier: the US Department of Energy, citing NEMA MG-1, puts current unbalance at 6 to 10 times voltage unbalance at full load, with a worked example where 2.5 percent voltage unbalance produced 27.7 percent current unbalance. It is worse at light load, where measurements have shown 3.13 percent voltage unbalance producing 37 to 40 percent current unbalance.

NEMA recommends holding 1 percent or less at the motor terminals. Above that, the motor has to be derated (roughly 0.88 at about 3 percent, and 0.75 to 0.76 at 5 percent), the Department of Energy notes that operating there will void most manufacturers' warranties, and operation above 5 percent is not recommended at all. One practical consequence for anyone setting protection: size the overload to the highest measured running current, not the average and not the nameplate, because the average is exactly what hides the phase doing the damage.

Now put the equipment tiers against those numbers. A general-purpose rotary publishes 5 to 8 percent imbalance. A digitally controlled rotary publishes 1 to 5 percent. A solid-state unit guarantees 2 percent or better. Meanwhile a CNC machine is not one motor: it contains its own drives and servo amplifiers, and some of those internal drives are designed to trip above roughly 3 percent imbalance to protect themselves. A converter that sits comfortably inside a plain induction motor's tolerance sits outside the CNC's. That is the whole mechanism.

One manufacturer publishes the resulting rule as a cross-sell, which is as close to a neutral recommendation as this market produces: equipment that needs 2 to 5 percent gets a rotary, equipment that needs 1 to 2 percent gets the solid-state unit.

What Haas actually says, since every converter vendor has a Haas landing page implying the opposite. Its published input requirements are grounded delta or wye power (the machines will not function properly on ungrounded power), voltage fluctuation within plus or minus 5 percent, 47 to 66 Hz, and harmonic distortion of 10 percent THD or less, with a service acceptance window of 195 to 260 volts. It publishes no numeric voltage-imbalance limit at all, so every percentage you see attributed to Haas is somebody's inference. And the phase converter section of its current pre-install guides opens with this:

"Do not use a phase converter unless it is necessary." (Haas Automation, machine pre-installation guides)

It goes on to note that a converter can leave the machine operating at less than full power, and its older service documentation treats a converter as a last resort when no other method is available. Converters are allowed. They are not endorsed. For scale while you are sizing: a 20 HP VF-series spindle draws 40 full-load amps in that 195 to 260 volt window, and a 30 HP inline spindle draws 70.

Compressors are the other sensitive load, and the mechanism is documented rather than folkloric. A major compressor manufacturer's application bulletin gives current unbalance at 4 to 10 times voltage unbalance, puts winding temperature rise at twice the square of the percent voltage unbalance, and names a single-phase load connected between two of the three phases as a common source of unbalanced voltage. A manufactured leg is precisely that. Attribute the thresholds correctly, because they come from different places: the 2 percent hard maximum is HVAC field convention, published in equipment installation manuals, while the 1 percent recommendation is the general motor guidance from NEMA by way of the Department of Energy. And if somebody proposes a drive on a hermetic compressor, it has to be a compressor-rated drive, not a shop drive: the compressor manufacturers publish their own drive families for it. A static converter is not a candidate for continuous compressor duty at all.

Phase converters have their own article in the electrical code, which surprises almost every owner who has priced one. It is Article 455, and it is carried into the 2025 California Electrical Code (Title 24 Part 3, based on the 2023 NEC) that took effect January 1, 2026. We found no California amendment to it, though local amendments are always possible and the authority having jurisdiction over your address is the one that decides.

This matters commercially, not just technically. The article is what turns a $999 converter into a project, and it is also the reason an owner-installed converter so often fails inspection. Paraphrasing the requirements that drive the work, with the section numbers so you can check any line:

Conductor sizing (455.6(A)(1)). The single-phase supply conductors are sized at not less than 125 percent of the converter nameplate single-phase input full-load amps. The converter nameplate, not the motor. A separate path at 250 percent of the sum of three-phase full-load currents exists for certain fixed loads, and only where that ampacity comes out below the 125 percent figure (455.6(A)(2)). The two are alternatives, not additive.

Overcurrent protection (455.7). Capped at 125 percent of nameplate input full-load amps on variable loads. That is a ceiling. Vendor advice that points you at a much larger breaker is pointing at a device the code does not permit here.

Disconnecting means (455.8). Opens all ungrounded single-phase supply conductors simultaneously, readily accessible, and in sight from the phase converter, meaning visible and within 50 feet, rated at not less than 115 percent of input full-load amps. One converter manual describes this disconnect as recommended. The code makes it mandatory.

The manufactured phase (455.9). No single-phase load connects to it.

Rotary sequencing (455.21 and 455.22). No power reaches the equipment until the converter has started, and equipment must drop out when power is interrupted. On a static installation, the input disconnect may serve the converter plus a single load in sight (455.20). Nameplate marking carries seven required items (455.4), the last of which is the rotary unit's three-phase full-load amps.

Fire pumps (695.3(I)). A phase converter is not permitted to supply power to a fire pump. If your building has one, that feed is out of scope for any converter, whatever its size.

There is a point where the converter conversation should stop, and it is not a matter of taste. It comes from a chain of three facts that compound:

One. Single-phase input current is physically larger than the three-phase load current it feeds, running roughly 1.73 to 2 times before any margin at all. Two. The vendors then size the converter well above the load: a genuinely hard 10 HP load wants a converter in the 20 to 25 HP range by their own published tables. Three. Section 455.6(A)(1) then sizes your feeder at 125 percent of that converter's nameplate input full-load amps.

Multiply those together against a 200 amp single-phase service that already runs lighting, HVAC, and a shop full of existing equipment, and the panel fills up fast. This is where owners discover that the converter they were quoted cannot actually be connected without a service upgrade, which was the thing the converter was supposed to avoid.

Two things decide whether the utility can serve you: whether three-phase primary runs on your street, and whether there is spare transformer capacity on it. Under PG&E's Rule 2, three-phase is supplied on request at 3 horsepower and above where existing transformer capacity is available. Where it is not, the job routes into either a line extension or special facilities, and the difference between those two words is worth tens of thousands of dollars.

The clause almost nobody writing about this knows: PG&E's Rule 15 names conversion of existing single-phase lines to three-phase lines, where required, inside the refundable amount of a distribution line extension. So a load-driven phase conversion is normally an extension with an allowance and a ten-year refund attached, not automatically a non-refundable special facility. That is materially better news than the internet consensus on this topic.

For non-residential work the allowance is a formula, based on the net revenue your new load produces divided by a cost-of-service factor, which means it scales with the size and revenue of the load rather than sitting at a flat figure. You advance only the excess above the allowance, that advance is refundable over ten years, a 50 percent non-refundable option exists, and an income tax component of contribution rides on contributions. Two clauses to plan around: there is a monthly charge on unearned refund balances after twelve months on non-residential accounts, and there is a claw-back if the contracted load is not actually used within a year, which is a reason not to inflate the forecast to chase a bigger allowance.

The hinge is 100 kVA. Where the utility determines your demand exceeds what a single-phase 100 kVA transformer can serve, its own construction standard says three-phase service must be supplied, and the allowance machinery attaches. Below that line, a three-phase request in an area where three-phase is not readily available can be treated as customer preference and routed to special facilities, where you pay the entire cost plus an ongoing monthly cost-of-ownership charge (0.49 percent per month customer financed, 1.23 percent utility financed). On a $50,000 advance that second figure is roughly $245 every month for as long as the facilities exist. Which is why a documented load calculation is the cheapest thing an owner can buy on this kind of project: it is the document that decides which side of the hinge you land on.

On cost, separate gross construction cost from what you advance. PG&E publishes a unit cost guide (March 2025, written for interconnection scope and explicitly non-binding), and the figures are larger than most contractor pages admit: an overhead primary service including one span at $32,700, urban overhead conductor at $234 per foot, an underground primary service up to 200 feet at $76,748, new underground line at $854 per foot, transformer and secondary packages in the 75 to 300 kVA range between roughly $52,000 and $62,000, and secondary metering at $5,150. Those are construction costs, not your bill. The allowance and refund mechanism sits between them and your net advance, and the two numbers can be very far apart on the same job.

Which is also why you should be skeptical of any page quoting a tidy flat range for an overhead extension. The commonly repeated "$10,000 to $20,000" figure has no utility-published support behind it, and the closest published match we could find for it is phase converter vendor marketing, a party with an interest in making utility power look expensive. Underground extension costs genuinely do reach the $30,000 to $100,000 and beyond range that our service page quotes, and the published unit costs support that end of it.

One sizing trap worth knowing before you pick a service size. Three-phase self-contained metering tops out at 225 amps in the utility construction standard. There is no 320 amp step on the three-phase side the way there is on single phase, so a service sized at 250 to 300 amps three-phase is a CT-metered service, with the metering enclosure and switchgear that implies. Sizing to 225 or accepting CT metering is a decision, not an accident, and it is cheaper to make it at design than at the meter set. In the other direction, a single-phase service above 400 amps triggers a conversation with your assigned job owner about whether the service should be three-phase in the first place.

Two municipal notes for the cities that are not PG&E territory. In the City of Santa Clara, Silicon Valley Power serves the meter, and its structure differs from the investor-owned model: the customer provides required facilities and pays a load increase fee plus per-kVA commercial load development fees, with current amounts in the city fee schedule, so confirm figures directly with SVP engineering and service planning at 408-615-6610. In Palo Alto, City of Palo Alto Utilities has the applicant pay all costs except the material for the transformer, switches, and meter, with a non-refundable advance engineering fee, a 1,600 amp ceiling on 208Y/120 service, and upgrades at 400 amps and above going underground with a pad-mounted transformer and an easement. Reach CPAU at 650-566-4500. Neither municipal structure carries the allowance-and-refund mechanism PG&E's rules do, so do not assume a municipal utility is the cheaper or faster path.

Six things decide how smoothly this goes, and five of them are yours:

This is where a machine purchase turns into a production problem, and it is the part most worth reading before you commit to a delivery date.

California's energization law and the CPUC decision implementing it set targets that have been binding since September 2024: for line-extension-class work, an average of 182 days and a maximum of 306 to 357 days, with a decision on whether your application is even complete inside 10 or 45 days depending on the path. Where a project triggers upstream work the targets stretch a long way: a maximum of 684 days for a new circuit, 1,021 for a substation upgrade.

PG&E's own filed performance is longer than the target. Line-extension work has been averaging roughly 307 to 314 calendar days end to end, and jobs falling under both extension rules have averaged 389 to 427 days, which is about ten to fourteen months. Projects that trigger upstream capacity work add about 492 days on average, and the upstream categories missed their targets outright.

Now the number that should change how you run the project. The utility-controlled portion of that calendar is about 117 to 118 days, and the 182-day clock explicitly excludes customer time: contract payment, permits, and site readiness are yours. The distance between 118 days and ten months is largely not the utility waiting on itself. It is applications sitting incomplete, documents waiting for a signature, and advances waiting to be paid. Your own turnaround is the single biggest lever you hold on this schedule.

There is a hard edge on that. Documents and the engineering advance have a 35 calendar day return window, and missing it cancels the application and restarts the clock. Losing five weeks to a piece of mail that went to the wrong person is a genuinely common way for a project like this to slip a quarter. On the other side, the utility is committed under the same law to complete intake within 66 days.

On equipment lead time, be accurate rather than alarmed. Distribution pad-mount transformers were running about 30 weeks as of the second quarter of 2025 and improving, and switchgear around 44 weeks. The 128-week and multi-year transformer figures circulating widely describe transmission and generator step-up class units, not the 75 to 300 kVA pad-mount that serves a shop. The utility construction standard itself warns to plan for several months of transformer lead time, which is the right instruction.

For context on the queue, both sides are on the record and they do not agree. PG&E projected a backlog on the order of 19,000 projects by the end of 2026 in its own regulatory filing. It has separately stated (March 2026) that its average is 118 days and that the number of customers waiting has fallen roughly 97 percent since the second quarter of 2023. Both are published. Plan against the longer end and treat the shorter one as upside.

The practical conclusion is the same one that applies to every utility-dependent project: open the application at the front of the job, not after the drawings are finished, and treat any date that depends on energization as unconfirmed until you have it in writing.