The EB-2 National Interest Waiver (NIW) is the one employment-based US green card an AI or ML engineer can file entirely on their own: no job offer, no sponsoring employer, and no PERM labor certification. You petition on the basis that your own work is valuable enough to the country that the government should waive the usual "prove no American can do this job" step. For a researcher or engineer with published work, real deployed models, or a track record of shipping AI at scale, it is often the most realistic self-directed path. This post is informational, not legal, immigration, tax, or financial advice.
What the EB-2 NIW actually is
EB-2 is the second employment-based preference category, open to people with an advanced degree (a master's or higher, or a bachelor's plus five years of progressive experience) or "exceptional ability." Normally an EB-2 needs an employer to sponsor you and to run a labor-market test. The National Interest Waiver removes both of those requirements when you can show your work serves the US national interest. You file Form I-140 yourself and you are your own petitioner.
Since a 2016 administrative appeals decision called Matter of Dhanasar, USCIS judges every NIW against three prongs, all of which must be met:
- Prong 1: your proposed endeavor has both substantial merit and national importance. AI and ML work in science, technology, healthcare, or security tends to clear "merit" easily; the harder part is showing "national importance" reaches beyond a single employer or region.
- Prong 2: you are well positioned to advance that endeavor. This is where your degree, publications, citations, patents, and shipped work do the heavy lifting.
- Prong 3: on balance it benefits the US to waive the job-offer and labor-certification requirement in your case.
The full framework lives in the the official USCIS Policy Manual chapter on advanced degree and exceptional ability, and the category basics are on the official USCIS EB-2 page.
What evidence an AI or ML profile uses
USCIS wants documentary proof, not adjectives. An AI/ML petition is usually built from some mix of the following, and you rarely need all of it:
- Published papers at venues like NeurIPS, ICML, CVPR, or ACL, with the papers themselves attached.
- Citation record from Google Scholar or Semantic Scholar, framed against typical counts in your subfield rather than as a raw number.
- Deployed production ML at scale: models serving real users, measurable business or safety impact, and letters or internal documents that confirm your role.
- Patents or filed applications tied to your technical contributions.
- Open-source impact: a widely used library, meaningful commits to a major framework, download or star counts as context.
- Recommendation letters, ideally a blend of people who know your work directly and independent experts who do not. These carry a lot of weight when they explain why the work matters nationally, not just that it is good.
- Peer review, media coverage, invited talks, or grant funding as supporting signals.
The strongest petitions tie every exhibit back to a clear proposed endeavor ("advancing efficient large-model inference," say) so the reviewer sees a coherent story rather than a pile of accolades.
Premium processing and timing
You can pay for premium processing on the I-140, which commits USCIS to act on that form within a set number of business days (currently 45 for I-140 NIW cases, as of 2026). That gets you a faster decision on the petition itself, but it does not shorten the second, larger wait: the priority date.
Once your I-140 is approved you still need an immigrant visa number to be current before you can file for the green card. For people born in India or China, the EB-2 queue is heavily backlogged and can run many years, sometimes over a decade. Check the monthly movement on the official State Department Visa Bulletin before you assume a timeline. Premium processing speeds the paperwork, not the queue.
How to start this month
If the NIW looks like a fit, here is a concrete order of operations:
- Confirm EB-2 eligibility first. Verify you hold an advanced degree, or a bachelor's plus five years of progressive experience, and gather the diplomas and evaluations that prove it.
- Write your proposed endeavor in one paragraph. Name the specific AI/ML problem you will keep working on and why it matters to the US, not just to one company.
- Build the evidence index. List every paper, citation report, patent, deployment, and repo, and note which Dhanasar prong each one supports.
- Line up 4 to 6 recommenders early, mixing people who worked with you and independent experts, and draft detailed letters for them to edit.
- Check your priority date reality on the Visa Bulletin for your country of birth so you know whether approval means "soon" or "years."
- Decide on premium processing based on whether a fast I-140 decision actually helps you (for example, to switch or extend status).
Wondering whether the US is even the right destination given the queue? You can compare all 21 countries on after-tax pay, visas, and years-to-PR, and the AI Relocation Guide lays out the alternatives side by side.
The honest takeaway
The NIW is genuinely self-service and it does not need an employer, which makes it powerful for founders, researchers between jobs, and engineers who do not want to be tied to a sponsor. But the approval bar is real: a thin record with two citations and no clear national-importance story gets denied, and paying for premium processing will not fix a weak case. If you were born in India or China, treat the priority-date backlog as the main constraint, because a fast approval still lands you in a long line.
Two neighbors are worth comparing. The EB-1A green card for AI researchers has a higher evidentiary bar but usually a shorter or current queue, and the O-1 visa for AI researchers is a fast nonimmigrant option that buys time while an NIW or EB-1A is pending.
Rule of thumb: file the NIW when your record clearly shows national-level impact and you can wait out your country's queue; reach for EB-1A or O-1 if you need speed more than you need the lower bar.
