Jobs for Physics PhDs: 10 Hours a Week Now Pays a Postdoc Salary

The NIH first-year postdoctoral stipend for FY2026 is $63,480.

micro1's live physics listings pay $80 to $200 an hour, and the schedule on them reads 5 to 10 plus hours a week, fully remote.

Ten hours a week at the middle of that band is $62,400 a year.

So a physics PhD can now match a full-time postdoc salary in roughly a day a week, and the work is closer to their thesis than most postdocs are. That is a genuinely new fact about jobs for physics PhDs, it is about eighteen months old, and no careers office has caught up with it.

Here is why it happened, and it is not that AI companies are generous. Every other employer that hires physics PhDs buys the transferable part: the coding, the statistics, the comfort with hard problems. The AI labs are the first buyers in the economy who want the untransferable part. They are posting for the Nishimori line, for first-order Palatini gravity, for Yb-171 sigma and pi transitions. The specialities you were told to stop mentioning at interviews are now line items on a job requisition.

This guide sorts jobs for physics PhDs by that one question, with real pay on each tier, the fourteen live physics listings and what each one wants, the arithmetic for graduate students, and the immigration rule that roughly half of you have to clear before you accept a dollar.

The number nobody in your department will say out loud

AIP's Statistical Research Center follows physics PhDs after graduation. For the combined 2015 and 2016 classes, the winter after the degree:

• 47% were in a postdoc
• 40% were in a potentially permanent position
• 7% were in another temporary position
• 6% were unemployed and seeking work

Now the part that matters. Of those in potentially permanent positions, only about 16% were in an academic setting. Roughly three-quarters were in the private sector.

Multiply it through. About 6% of new physics PhDs hold a potentially permanent academic position one year out.

The trend is not a blip either. The postdoc share fell from 67% for the class of 2004 to 45% for the class of 2016, and the offset went to permanent private sector jobs, not to faculty lines.

The same survey gives the money, and the spread is the whole story:

• Private sector, potentially permanent: median $105,000
• Government lab postdoc: median $68,500
• University postdoc: median $50,000
• Academic, potentially permanent: median $58,000

Read those four numbers again. A potentially permanent academic job paid a median of $58,000, which is less than a government lab postdoc. The permanent academic position, the thing the entire pipeline is optimised for and that 6% of you will hold, pays below the temporary job at a national lab.

None of this is an argument against academia. It is an argument against letting a 6% outcome be your only plan while you are paid $29,364 to wait for it.

The axis: does the buyer pay for your subfield, or ask you to drop it?

Every careers guide written for physics PhDs says the same thing. Your skills transfer. Emphasise the Python, the statistics, the problem solving, the project management.

That advice is exactly why physics PhDs take a pay cut.

It tells you to lead with the part of you that is common and hide the part that is scarce. There are perhaps a few hundred people alive who can referee a claim about noncontractible loop defects in a replicated random-bond Ising model. There are hundreds of thousands who can write Python, and most of them have been doing it full time while you were writing a thesis.

So sort the market by whose problem you solve with which half of yourself.

Tier one pays for the subfield. The thing on your thesis cover is the product. Narrow is an advantage.

Tier two pays for the toolkit and discards the subfield. Highest ceiling on the page. You also restart as a junior in somebody else's field.

Tier three pays for the credential. The letters matter, the speciality does not.

Most physics PhDs go straight to tier two because it is the only tier their department talks about. Look at all three before you decide.

Tier one: work that pays for your speciality

AI model evaluation. Published bands of $80 to $200 an hour, by subfield, remote, contract, 5 to 10 hours a week. Full listing breakdown below. This is the only entry on this page where being a gravitation theorist pays more than being a generalist.

National labs. The Los Alamos postdoc scale starts at $97,000. AIP puts the median government lab postdoc at $68,500 against $50,000 at a university. Same title, same person, roughly 1.5 times the pay, and the research is closer to physics than most industrial R&D. There is a catch and it is a clock, covered below.

Quantum hardware, photonics and semiconductors. The industrial employers who genuinely need the physics rather than the numeracy. An AMO PhD who has actually stabilised a laser, a condensed matter experimentalist who has run a dilution fridge, a materials theorist who has done real DFT. These roles are scarce, they are specific, and they pay for the specificity.

Medical physics, with a gate you have to see first. Every list recommends this and almost none of them mention that a physics PhD does not qualify you. Since 2012 the ABR Part I exam has required CAMPEP-accredited education and since 2014 Part II has required a CAMPEP-accredited residency. So a straight physics PhD needs a CAMPEP certificate programme, then a residency, and residencies are the bottleneck: as CAMPEP graduate output rose from 210 to 332 between 2009 and 2015, residency slots rose only from 60 to 134, and just 36% of CAMPEP graduates secured a residency in 2015, with a median placement rate across programmes of 22%. This is a real destination and a multi-year, gated one. Treat it as a degree, not as a pivot.

Tier two: work that pays for the toolkit

Quantitative finance. The highest ceiling available to you. Reported first-year total compensation for quant researchers at top funds and prop shops runs roughly $250,000 to $450,000, with the broader market nearer $150,000 to $250,000 and a levels.fyi median around $200,000 for the title. PhD entrants typically start a level above a masters hire.

Data science and machine learning. levels.fyi puts the overall US data scientist median near $180,000, with hedge fund and top tech bands far above that.

Software engineering. Deep, liquid, and indifferent to your thesis.

Be clear-eyed about the trade. In tier two your subfield is worth zero on day one. Nobody at the fund cares about AdS/CFT. What they are buying is that you can hold a hard problem in your head for six months without losing the thread, and they will pay a lot for it. But you are now a beginner in someone else's discipline, competing with people who have five extra years in it, and the physics only comes back as a personality trait.

That is a fine trade to make. It is a bad trade to make by default, which is what happens when tier one never gets mentioned.

Tier three: work that pays for the credential

Teaching and tutoring, science writing and editing, patent work as a technical specialist or agent, textbook and problem authoring, expert network calls, science policy and programme management.

Real work, real ceilings in a few cases, and the credential is doing the lifting. Useful as a bridge, especially while a tier one or tier two search runs. Expert networks in particular pay well per hour and cannot be planned around, because call volume is somebody else's pipeline.

The fourteen listings, and what they actually ask for

These are live listings from Skillora's AI training jobs feed, which aggregates Mercor, micro1, Alignerr, Outlier, DataAnnotation and Terac. Rates are as each marketplace publishes them.

The micro1 physics family, which is the largest single block of physics work on the feed:

• Physics Expert PhD (Statistical Physics / Quantum Information / Condensed Matter), $100 to $200 an hour. Wants replicated random-bond Ising and Ashkin-Teller models, the toric-code threshold, the Nishimori line, Kramers-Wannier duality, quenched disorder averaging, square-lattice self-duality, domain-wall free energy, 4-state Potts numerics, noncontractible loop defects.
• Physics Expert PhD (AMO / Optical Properties of Materials), $100 to $200. Ab initio quantum chemistry, configuration interaction, polarizability calculations, Yb-171 and Yb-174 isotopes, sigma and pi transition identification, atomic spectroscopy.
• Physics Expert (PhD / Postdoc), $100 to $200. General theory and computation: LaTeX, SymPy, Python, Jupyter, across high energy, mathematical physics, biophysics and statistical physics.
• Physics Expert (Condensed Matter / Quantum Information), $100 to $170. Exact diagonalisation, QuSpin, symmetry block-diagonalisation, quantum many-body scars, Rydberg blockade.
• Physics Expert (Condensed Matter / Magnetic Materials), $80 to $160. Magnetic structure factor calculations, propagation-vector analysis, magnetic space groups, MAGNDATA, BNS notation, neutron scattering and diffraction.
• Physics Expert (AMO / Quantum Information), $80 to $160. Cascaded optical parametric amplifiers, SU(1,1) interferometry, two-mode squeezing, Bogoliubov transformations, covariance-matrix methods, homodyne detection.
• Physics Expert (Gravitation / Cosmology / Astrophysics), $80 to $160. First-order Palatini gravity, tetrad and spin connection formalisms, dynamical Chern-Simons gravity, Pontryagin density, FRW cosmology, Friedmann ODEs with torsion, Planck-unit normalisation.
• Physics Expert (Biophysics / Statistical Physics), $80 to $160. Two-state Markov processes, gamma-distributed waiting times, the Euler-Lotka equation, renewal theory, first-passage-time analysis, bacterial population dynamics.
• Physics Expert (High Energy / Mathematical Physics), $80 to $160. AdS/CFT, holography, on-shell gravitational actions, Weyl anomalies, the Fefferman-Graham expansion and Weyl-Fefferman-Graham gauge.
• Physics Expert (Professor / Principal Investigator), $80 to $160. Research leadership and expert judgment, explicitly priced for people who supervise rather than execute.
• Physics Expert (Postdoc / Junior professor), $80 to $150. Scientific review and error detection, derivations, dimensional analysis, computational validation.
• PhD Physicist, $70 to $90. Broader: theory, experiment, data analysis, content evaluation, teaching.
• Physical Sciences Research Assistant, $30 to $50. Rubric design, reference curation, protocol fidelity. The entry rung.

And the same work at the other marketplaces:

• Physicist, DataAnnotation: $75 to $125
• Physicist, PhD, Alignerr: $90 to $120
• Physics Expert, Alignerr: $20 to $120, a wide band that should tell you the screen decides the rate
• Physicist Talent Network, Mercor: $60 to $80
• Applied Physics Benchmark Specialist, Mercor: $61 to $77
• Physics PhD Coding Experts, Mercor: $70
• Applied Physics Specialist, Alignerr: $40 to $80
• Physics Masters, Alignerr: $35 to $55

Browse the current set at science, which is where physics sits on the board. The counts move daily.

What the listings tell you that no careers page will

Read past the rates and four things fall out.

The subfield is the product. Nowhere else will you see a requisition asking for the Nishimori line by name. Every other employer asks for "strong analytical background." These labs are building benchmarks that frontier models cannot yet pass, which means they need problems only a specialist can set and only a specialist can grade. Your narrowness, the thing that made the academic job market brutal, is the reason the rate is what it is.

You are graded, audited, or you adjudicate. micro1's physics projects assign experts as Solver, Auditor or Adjudicator depending on subfield strength. That is a research group's structure, not a crowdwork queue, and it is why the professor and principal investigator listing exists as its own line at $80 to $160.

No AI experience is required, and they say so. The listings state it plainly: domain knowledge is what matters. There is no machine learning prerequisite to clear, which is the opposite of the tier two path.

Referrals pay real money. micro1 shows a refer-and-earn amount on each posting, and the figure varies by listing: $1,000 on the gravitation and cosmology role, $2,000 on the statistical physics and quantum information role. If your old group has six people with the same speciality, that is worth knowing before you apply alone.

One honest caution. The bands are wide, and the bottom of a band is a real outcome. Alignerr's general physics listing runs from $20 to $120 for the same title. What decides where you land is the screen, and the screen is a subfield interview.

Side hustles for PhD students: three permissions, in this order

If you are still in the programme, the arithmetic is even starker. The NIH FY2026 predoctoral stipend is $29,364 a year, about $2,447 a month. At $120 an hour, the midpoint of the main physics band, that is under five hours a week.

Which is exactly why you have to get the permissions right rather than signing up tonight.

Permission one: immigration status, and this is the one that ends careers. AIP found that 47% of physics PhDs were conferred on non-US citizens, so this is about half of you. On an F-1 visa, off-campus employment requires authorisation. On-campus work is capped at 20 hours a week while school is in session. Off-campus work needs CPT, authorised by your designated school official on your I-20 and tied to your curriculum, or post-completion OPT with an approved I-765 and an EAD in hand. And the consequence is not a warning: if a student engages in unauthorised employment, the DSO must terminate the SEVIS record.

Contract work for an AI marketplace, paid on a 1099, performed from your apartment in the US, is off-campus employment. A listing that says "open to candidates worldwide" is describing the company's hiring reach, not your immigration status. Ask your international student office before you accept, not after the first invoice. If you are a US citizen or permanent resident, you clear this gate automatically and can skip to permission two.

Permission two: your appointment. Research and teaching assistantships almost always cap outside work and require the graduate school's or your advisor's sign-off. This is a conversation, not a form, and it goes better when you have the numbers: five hours a week, evenings, fixed rate, no travel.

Permission three: your intellectual property agreement. Anything you produce that touches your dissertation area raises a question your funding agreement has already answered. The safe rule, and it is the same rule this series gives lawyers and accountants: write from your knowledge, never from your group's unpublished work. No data, no figures, no results that have not been published, nothing from a collaboration you are bound by.

Same three checks apply to a postdoc, with the added item of your institution's outside professional activity policy, which typically allows something like one day a week with disclosure.

The clock: five years

Every profession in this series has a clock, and most people only find theirs after it has run out.

For physics PhDs it is the postdoc eligibility window. Los Alamos requires the postdoc application within five years of completing all requirements for the PhD, with exceptions for family-related career breaks. Other labs and fellowships set their own versions.

So the national lab route, which is the best paid tier one research job on this page, closes on a schedule. The academic queue does not close, which is precisely why it absorbs people. You can keep applying for faculty jobs forever. You cannot keep applying for a lab postdoc forever.

Compare that with the rest of the series and the pattern is clear. Nurses have a multistate licence compact. Accountants have practice privilege. Physicians have re-entry standards that bite after an average of 2.9 years away. Lawyers have nothing that expires. Mechanical engineers have portable paperwork and non-portable licences. Physics PhDs have no licence at all and one hard deadline nobody tells them about.

Two consequences. If a national lab is on your list, apply early and do not spend year four deciding. And if you want the optionality, the tier three and AI evaluation work has no clock whatsoever, which is why it is the sensible thing to run underneath everything else.

Your first month

1. Write down the 6% number and your own plan against it. If academia is the goal, fine. The question is what else runs in parallel, not whether to give up.
2. Check your immigration status before anything else. If you are on F-1, talk to your international student office. Nothing on this page is worth a terminated SEVIS record.
3. Read your outside activity and IP policies. Institution policy and funding agreement. Ten minutes, and it settles what you can accept.
4. Apply to two or three platforms, not one. Each is a queue owned by somebody else's client contracts, and a quiet queue is not a verdict on you.
5. Lead with the subfield, not the toolkit. Your application should name the speciality, the methods you have actually run, and the papers. This is the one market where "I work on Rydberg blockade" beats "proficient in Python."
6. Prepare for the screen as a physics conversation. State the assumption, name the governing result, do the derivation, say what would change your answer. That is what an auditor does, and auditing is what they are hiring for.
7. If your old group has people in your speciality, mention the referral. $1,000 to $2,000 per listing is not a rounding error on a postdoc salary.
8. Price everything per hour, against $30.52. That is the FY2026 first-year postdoc stipend over 2,080 hours, and it is the real benchmark. Not zero.

Where the physics listings are

Skillora's AI training jobs board collects live listings from Mercor, micro1, Alignerr, Outlier, DataAnnotation and Terac into one feed, refreshed daily and filterable by pay, skill and experience level. The science listings carry the physics, materials and chemistry work, and the data and machine learning listings carry the adjacent evaluation roles that a computational physicist clears easily. Every listing links back to the marketplace that posted it, you apply and get paid by them directly, and browsing is free. Where a marketplace runs a referral programme Skillora may earn a fee if you sign up through our link. It costs you nothing and has no effect on how listings are ranked.

Two things before you apply. These marketplaces screen with a recorded or live AI interview rather than a resume panel, which is the specific reason a long postdoc chain, a non-US institution or a two-year gap does not cost you here the way it does in a faculty search. The micro1 AI interview guide and the Mercor AI interview guide cover that format, and the screening guide compares how many attempts each platform gives you before you are locked out, which decides the order you should apply in. The best AI training jobs roundup ranks the platforms on pay, entry bar and the catch on each, and the micro1 review covers the one running most of the physics work.

Other credentialled fields run the same sort against a different gate. Freelance mechanical engineer sorts by whose product the deliverable is for. Side hustles for accountants sorts by whether you are selling your signature, your seat or your knowledge. Side hustles for lawyers sorts by whether the work creates a client. Jobs for former teachers prices every exit per day actually worked.

The short version

About 6% of new physics PhDs are in a potentially permanent academic position a year out, and that position pays a median of $58,000, less than a national lab postdoc.

The standard advice is to lead with your transferable skills. That advice moves you into the one market where your training is worth nothing and your competition has a five-year head start.

There is now exactly one buyer that prices the untransferable part, by name, at $80 to $200 an hour, for five to ten hours a week, remote, with no AI background required. It will not last forever, because benchmark work has a shelf life and the models keep improving. It is unusually good while it lasts.

Ten hours a week matches a postdoc. Five hours matches a graduate stipend. Check your visa status, read your IP agreement, then go and get paid for the thing you spent six years learning.