What the May 2026 frontier-hiring data actually says.

If you ask the AI engineering job market what it's doing right now, you'll get four different answers depending on which compensation tracker you read. We tracked 5,800 senior-IC offers across 52 frontier labs and AI-native companies from January to April this year, and three of those answers turn out to be reasonable. The fourth, which is the headline number on the largest public comp board, is wrong by a margin big enough to make a real difference to anyone planning a move.

This is the May 2026 issue of the OpenTalent blog's frontier-hiring read. We do this every quarter. The methodology is at the bottom of the data page, and the underlying offers are network-verified — meaning at least one party to the offer (usually the candidate, often both sides) shared it with us privately, under contract. We do not pull from public scrapers and we do not use levels.fyi as a source.

If you'd rather skim, the four numbers are:

• +26% YoYfor inference engineers — the fastest-rising senior IC role in our dataset.
• +34% YoY for India-based senior IC offers across all tracks.
• 38% of senior offers now include a signing bonus above $200K. A year ago it was 21%.
• Median time from scope to written offer fell to 11 days. Two years ago that was 27.

And the number we think most other trackers got wrong: applied AI senior IC comp at frontier-adjacent product companiesis not flat. It's up, but quietly, in a way that the public boards aren't catching. We'll show you why.

// 01The inference number

Of the seven senior-IC engineering specializations we track inside ML engineering, only one cracked a 25% year-on-year compensation lift. The inference engineers — the people working on paged attention, speculative decoding, batching strategies, KV-cache reuse, and the rest of the modern serving stack — saw their senior IC median total comp climb from roughly $690K to $870K across our dataset.

Why this happened isn't mysterious. Frontier labs have moved past the phase where the binding constraint was “can we train this.” They're now in the phase where the binding constraint is “can we serve this at margin.” That changes who gets paid. Last year the equity refresh tiers at the top three frontier labs we track favored training-platform engineers; this year, in three out of three cases we've seen, the inference team is calibrated one tier higher.

The practical read for engineers: if you're currently on a training-platform team and you've spent the last year touching anything inference-adjacent (serving stacks, model graph optimization, decoding work), it's worth having a conversation with your manager about whether you're calibrated correctly. The data says you probably aren't, and it says your next move outside is going to clear the gap whether you want it to or not.

What this doesn't mean

This isn't a “pivot into inference” call. Specializations don't generally reward people who jumped in for the comp number. What it is, though, is permission to recognize that if you're already there, the market knows it now, and you should price accordingly.

// 02The India number

The single largest year-on-year compensation lift in our entire dataset wasn't a role. It was a geography. India-based senior IC offers climbed 34% YoY across all engineering tracks. The median India-based senior IC total comp in our dataset, USD-equivalent, is now $310K— up from about $231K a year ago.

Three things drove this, and they're all worth understanding because the trend is structural, not cyclical.

The first is Sarvam and Krutrim moving onto the comp leaderboard for India-based senior offers. Both labs were already paying competitively at the staff and principal levels a year ago; this year the senior IC bands moved up too, and Sarvam in particular widened its equity component meaningfully. The labs that anchor a geography's high-water mark set the ceiling for everyone else, and that's what we're seeing.

The second is frontier-lab remote roles becoming common for India-based candidates. A year ago we'd see maybe two senior IC US-frontier-lab roles per quarter where India-based candidates were genuinely competitive; this year it's running at roughly fifteen per quarter, and the comp band on those roles is the same as their US-based equivalents minus a real but shrinking geo adjustment (~30% rather than the historical ~55%).

The third, and most subtle, is the rise of dual-time-zone hiring. Several frontier labs we track have built out enough internal infrastructure (engineering management depth, eval review cycles, on-call rotation) to support engineers in IST as first-class team members rather than as remote add-ons. That changes the comp calculus — and once one lab makes that move, the others follow within two quarters, because the talent pool you're suddenly competing for got 5× larger.

// 03The signing-bonus number

If you negotiated a frontier-lab senior IC offer in 2024 and your offer came without a signing bonus, you weren't unusual — roughly four-fifths of senior offers didn't include one. This year the pattern is different. 38% of senior IC offers we tracked in Q1 included a signing bonus above $200K. Last year that number was 21%.

The instinctive read is that this is a competition-driven escalation, like the FAANG signing-bonus war of the late 2010s, and that the bonus number will normalize when the cycle cools. We don't think that's right. The structural reason this is happening is more interesting.

What we're seeing in the actual offer letters — and we read more of these than is good for anyone — is that the signing bonus is increasingly being used as a notice-period buyout instrument, not as a competitive top-line number. The candidate's existing offer comes with a long notice period (45–90 days), and the new lab uses the signing bonus to cover the gap. The bonus is materially the same size as the candidate's salary during their notice; the lab is essentially buying them out.

Why does this matter? Because it means the signing bonus money is not “extra.” It's compensation for the lab's faster start date. And because of that, the bonus number is highly negotiable in both directions: candidates with short notice periods leave a lot on the table by not asking for a signing bonus at all; candidates with long notice periods can often get more if they explicitly frame the ask around their existing notice obligations.

The practical advice we give network members is in the negotiation insights section of the comp page. We won't repeat it here. But the short version is: in 2026, signing bonus is not free money. It's compensation for a specific thing. Treat it that way.

// 04The shrinking-loop number

Median time from scope to written offer in our dataset, for senior IC frontier-lab roles, has fallen from 27 days in 2024 to 11 days in May 2026. That is, to a first approximation, the same loop running twice as fast.

What changed isn't the screening rigour — which has if anything intensified — but the latency between the stages. Two structural shifts.

First, frontier labs have realized that their old “take-home → review week → panel → debrief week → offer” cadence was costing them candidates to faster-moving labs. Several of the top labs in our dataset are now batching their reviews tighter (24–48 hour review SLAs instead of 5–7 days), and that one change compresses the loop by close to a week and a half.

Second, OpenTalent's network candidates are typically pre-vetted enough that the screening stages can be partially parallelized — which is something the labs only do when they're confident in the upstream filter. The candidates who join through us start their loops roughly five days aheadof candidates from the same lab's general pipeline, in our dataset.

You should care about this even if you're not in our network, because it changes how you should think about cycle planning. The window between “I'd like to move” and “I have a written offer” has compressed sharply. That tightens the case for running multiple loops in parallel — the old “let me focus on one” calculus made sense when each loop was six weeks; it doesn't survive when each loop is two.

// 05The number the public boards got wrong

The biggest comp tracker in our space is reporting that applied AI senior IC compensation at frontier-adjacent product companies was flat year-on-year. We have 1,860 applied AI senior IC offersin our dataset across that exact category, and they very clearly weren't flat.

Median total comp for senior IC applied AI roles moved from about $498K to $560Kin our window — a real +12% lift. Within applied AI, the sub-specialization that moved the most was agent product engineering (+24% YoY, the fastest-rising applied AI track).

Why does the public board show flat? Two reasons. First, public boards are dominated by self-reported new-grad and junior IC data, and at those bands applied AI is genuinely flatter — the labs were already paying competitively for entry-level applied AI work last year. Second, the senior IC end of applied AI is over-represented by frontier-adjacent product companies (Stripe AI, Notion AI, Glean, Sierra, Harvey, Cursor) that don't post comp publicly and don't show up on public scrapers. Their offers don't appear in the dataset that everyone is reading.

The practical read: if you're a senior applied AI engineer using public comp boards to anchor your negotiation, you're probably anchoring below market. By a meaningful margin.

The composition pattern

One more thing on the senior applied AI band, because we get this question often. The composition of these offers differs from frontier-lab senior IC offers in a structurally important way: the equity component is smaller (typically 28–38% of total vs. 45–55% at frontier labs), and the base is larger (35–45% of total vs. 25–35%). The reason is that frontier-adjacent product companies are often later-stage or post-IPO, which makes their equity more liquid but less explosive, and the labor market has priced that.

If you're moving from a frontier lab into a frontier-adjacent product company at a comparable headline total comp, you'll usually be moving from a higher equity / lower base mix to a lower equity / higher base mix. Your year-one cash will go up; your 4-year expected value depends on what you think the equity does. Worth being clear-eyed about which version of “$560K total comp” you're actually negotiating.

// 06What we'd plan around, if we were planning

We're not a career-coaching outfit, so we won't tell anyone what to do. But here's the planning lens we'd use if we were a senior IC engineer reading this in May 2026 and thinking about the next 12 months.

If you're inference- or serving-adjacent — even marginally — make sure your last 12 months of work surfaces in your resume and Cohire profile in the inference column. The market premium is real and the panels know it. Don't bury this in a “ML infrastructure, generally” line.

If you're India-based and senior — the band you should be negotiating in moved, fast, and most engineers haven't updated their internal anchor. Refresh it. The geography no longer pays a 50% discount to US frontier-lab equivalents; in 2026, the realistic spread is closer to 30% and shrinking.

If you've been holding off on a move because you didn't want to deal with the multi-month interview marathon — that calculus changed. Loops are 11 days now. Two-month exclusive focus on one lab is no longer the dominant strategy; running three loops in parallel is.

If you're senior applied AI — don't anchor on public boards. The senior IC band at frontier-adjacent product companies moved up about 12% in the past year and your future employer reads the same data we do. Use the network comp data in negotiation if you have access to it; if you don't, at least don't undersell yourself based on a number scraped from new-grad postings.

The next issue of this blog publishes in three weeks, covering the August 2026 dataset. If you'd like it to land in your inbox, subscribe at the bottom of the blog page. We send one email a week, on Wednesdays, and there is no other email from us.