AI Internships in 2026: The Application Calendar and the Honest Odds

Read the calendar before you read anything else

It is July 2026. Summer 2026 internships are already running, which means if you’re searching “ai internships” right now, you’re really shopping for summer 2027 — and that cycle opens in a matter of weeks. Amazon posts first, in July–August. Microsoft follows around mid-August. By the time Google STEP opens its famously short window in mid-October, the rolling-admission companies have already been taking applications for weeks — and rolling means earlier applicants face materially less competition.

That’s the whole reason this page exists. Every other result for this search is a job board counting open listings or a listicle with no dates. Neither tells you the one thing that actually decides outcomes: when to apply. Rolling admissions is the default in this market, and applying in the first two weeks of a window faces measurably less competition than applying in week eight. If you’re not checking careers pages by September, you’ve already missed half the FAANG openings.

So here’s the calendar, then the honest part — your real odds, what interns actually earn, and the concrete ladder to climb if you get rejected everywhere. Which, statistically, most applicants will be.

The summer 2027 application calendar

All dates as of July 2026 — windows move, so verify on the employer’s own careers page before you plan around one. Rolling means exactly that: apply within two weeks of opening or accept worse odds.

Two structural notes on the 2027 cycle worth knowing before you apply. First, this is reportedly the first intern cycle where AI/ML openings outpace traditional software-engineering intern spots. Second, Google and Meta now permit AI assistants during coding interviews — the bar didn’t drop, it moved: they’re now testing whether you can work with the tools, not whether you memorized a binary tree.

If your target is specifically research-track ML — NSF REU site selection, research internships like DeepMind’s Student Researcher program, or the quant-ML pipeline in depth — that’s its own game with its own skill bar, and we cover it separately in machine learning internships.

Still looking for summer 2026? Read this paragraph, then move on

Honestly: the corporate summer 2026 intern class was hired months ago. Your live options this late are rolling programs that never fully close — Ai2 takes research interns year-round, MLH Fellowship runs seasonal remote batches — plus research assistant work with a professor, which has no application season at all. Otherwise, the highest-value use of the rest of this summer is building the proof-of-work that gets you into the 2027 class, or doing paid AI work that fits around a student schedule — the options for that are in AI jobs for students. Don’t burn August refreshing listings for a season that’s over; burn it preparing for the one that’s opening.

The Anthropic/OpenAI internship that doesn’t exist

Here’s the misconception this page most wants to kill: students spend whole application seasons hunting “Anthropic internships” and “OpenAI internships” that do not exist in the form they imagine. The frontier labs mostly do not run traditional undergrad summer internships. What they run instead:

• Anthropic Claude Corps — a 12-month, paid, in-person fellowship at $85,000/year plus benefits and relocation. Open to anyone 18+ with US work authorization, under two years of full-time work experience, any educational background. Cohort 1 (100 fellows) closes July 17, 2026 for an October start; cohorts 2 and 3 follow in January and August 2027. Note what it is: a year-long commitment, not a summer you slot between semesters.
• Anthropic Fellows — an AI-safety research fellowship paying a $3,850/week stipend plus roughly $15k/month in compute. Four months, full-time. Built for early-career researchers, not enrolled students on break.
• OpenAI Residency — six months, full salary (around $18,300/month reported), and you cannot be enrolled in school to do it. The 2026 cohort’s applications are already closed as of this writing; watch for the next cycle.

The pattern: labs recruit people who can commit full-time — which usually means graduated, on leave, or between programs. If you’re an enrolled undergrad who wants a summer at a top AI org, your realistic targets are the big-tech AI/ML internships in the calendar above, plus research routes like Ai2 and DeepMind’s Student Researcher program. Chasing a lab internship that doesn’t exist costs you the weeks when the real windows are open.

Your actual odds (and what beats a 4.0)

Nobody ranking for this keyword will print these numbers, so we will. At the elite end, internship admissions are lottery-adjacent math: Jane Street takes roughly 250 interns from 50,000+ applications — under 1%. For calibration, Goldman Sachs ran 0.7% (360,000 applicants for 2,600 seats) and JPMorgan about 0.8%. The most selective tech and quant intern programs live in that same league. This is not a “study harder” problem. A rejection from that tier is the expected outcome for nearly everyone who applies, including people who deserved a seat.

Even being plainly qualified doesn’t convert the way students assume — estimates put qualified-candidate conversion at brand-name tech firms around 15–20%. The rational response isn’t applying to five dream companies. It’s applying early (see the calendar), applying across tiers — big tech and mid-size AI companies and research programs — and treating every application as a cheap ticket rather than a referendum on your worth.

As for what actually moves you through a screen: shipped work beats GPA. NVIDIA says this in public on its own recruiting pages — practical skills, projects, and research work are valued alongside grades, not beneath them. A real repository, a Kaggle placement, a merged pull request to a known ML project, a deployed thing with users: these survive a resume screen because they’re checkable. A 3.9 is not. This is the same proof-of-work principle that runs through everything on this site, and the step-by-step method for building it from zero is in AI jobs with no experience. If you have one summer and no internship, that method is your internship.

What AI internships pay

Pay in this market is bimodal — a broad big-tech band, and a quant outlier ceiling that skews every average you’ll see quoted.

Big tech (Google, Microsoft, Apple, NVIDIA, Amazon, Meta): AI/ML interns cluster at roughly $45–62/hour, which works out to about $8,000–11,000/month, typically with housing or relocation support on top. Treat these as estimate bands, not offers — most of these companies confirm “paid” officially without publishing rates, so the numbers come from aggregators and self-reports.

Quant firms (the ceiling): the top of this market is quant ML, where weekly intern pay reaches the equivalent of a ~$300k annualized salary plus housing and flights — real, documented figures, attached to the sub-1% acceptance rates above. The firm-by-firm breakdown of that pipeline lives in machine learning internships.

Research (NSF REU): about $6,300 for the summer plus housing and travel — a fraction of corporate pay, but you’re buying a faculty recommendation letter and a research line, which corporate internships don’t sell at any price.

Ranges compiled from platform listings, job postings, and worker reports · last verified July 2026.

One rights note while we’re on money: unpaid AI internships at for-profit companies are only legal in the US if they pass the Department of Labor’s seven-factor “primary beneficiary test” — genuine training, tied to your education, defined dates, no promise of a job. If you’re producing profit-generating work for free, the company likely owes you at least minimum wage. Know that before you accept “great exposure” as compensation.

Rejected everywhere? Here’s the actual ladder

Most applicants — including good ones — end the cycle with zero offers. The difference between students who break through the following year and students who don’t is what they do with the gap. In rough order of impact:

1. Research assistant with a professor. Email faculty whose papers you’ve actually read, with one specific sentence about their work. Often unpaid or work-study, but it’s the single most reliable route to a real recommendation letter and a research line on your resume — the two things a rejected application season can’t take from you.
2. NSF REU. The structured, paid version of the same thing: ~$6,300 plus housing for a summer of mentored research. Deadlines run January–March for the following summer, US citizens and permanent residents only. Apply to several sites.
3. MLH Fellowship or Ai2. Remote, mentored open-source work (MLH, beginner-friendly, seasonal batches) or a rolling research internship (Ai2, Seattle). Both read as internships on a resume because both are internships in everything but the recruiting-season branding.
4. Open source, Kaggle, and shipped projects. Proof-of-work you control completely — no acceptance rate, no application window. A placed Kaggle notebook or a merged PR to a known ML repository is exactly the checkable evidence the previous section said beats GPA.
5. Paid AI-training work. Platforms pay students right now to evaluate model outputs, write and rate prompts, and label data — real, hands-on model-evaluation experience that goes on a resume honestly as AI work, and it pays this month instead of next summer. Think of it as the paid internship alternative nobody markets: the vetted platforms are in AI training jobs and data annotation jobs, and the broader map of roles that don’t need an internship at all is in entry-level AI jobs.

A student who spends the year on rungs 1 or 4 plus rung 5 shows up to the next cycle with research or shipped work and paid AI experience. That profile beats “reapplied with a higher GPA” every time.

Red flags: the fake internship economy

The gap between demand (“AI internship” on a student resume) and supply (sub-1% acceptance) has spawned a scam layer. Three rules:

• Anyone charging you is a scam. No exceptions. No legitimate internship charges a registration fee, training fee, security deposit, or “materials” cost. Employers pay you. The moment money flows toward the company, close the tab.
• “AI internship certificate” mills are worthless. The pitch: complete a short online “internship,” receive a certificate, often for a small upfront fee — with no real work, no mentor, and no employer on the other end. Recruiters recognize these instantly, and not favorably. A certificate is not an internship; work is an internship.
• Impersonators spoof real companies. Scammers send fake Google and Meta offer letters over WhatsApp and personal email, then ask for bank details or an upfront fee. Real recruiting arrives through official company domains and applicant portals. A chat app asking for payment is a scam every single time.

Applying only through the official careers pages listed in the calendar above filters out essentially all of this.

Tools that get the interview

The calendar tells you when; volume and tailoring decide the rest, because rolling admissions rewards students who can send strong applications fast in the first two weeks of a window. Our current picks for that part — with the honest caveats and what each actually costs — live on one page: the tools we actually recommend.

FAQ

When do summer 2027 AI internships open? The cycle starts now: Amazon opens July–August 2026, Microsoft around August 15, quant firms and NVIDIA in August, Meta in early September, Apple September–November, and Google STEP in a short mid-October window. Most are rolling — applying in the first two weeks of a window meaningfully improves your odds. Dates as of July 2026; verify on each careers page.

Do AI internships pay? Legitimate ones, yes — and well. Big-tech AI/ML interns earn an estimated $45–62/hour (roughly $8–11k/month), quant firms pay $4,300–6,000/week at the extreme, and NSF REU research pays about $6,300 for the summer plus housing. Any “internship” that charges you a fee instead is a scam.

Do Anthropic and OpenAI offer internships? Not traditional undergrad summer internships. Anthropic runs Claude Corps (a 12-month, $85k fellowship) and the Anthropic Fellows research program; OpenAI runs a six-month Residency that requires you not to be enrolled in school. If you want a summer at a top AI company while enrolled, target the big-tech AI/ML internships and research programs instead.

Can freshmen get AI internships? Yes — two programs exist specifically for them: Google STEP and Microsoft Explore both target first- and second-year undergrads. Research routes (professor’s lab, REU sites) also take early students, and they care more about demonstrated interest and basic Python than credits completed. Everyone else’s postings mostly favor penultimate-year students.

What are my realistic chances? At the elite end, brutal: Jane Street accepts under 1% (~250 interns from 50,000+ applications), and top tech programs sit in the same league as Goldman’s 0.7%. Even qualified candidates convert at an estimated 15–20% at brand-name firms. Apply early, apply across tiers, and expect rejection as the default — it’s the market, not you.

What if I get rejected everywhere? Run the ladder: research assistant with a professor, NSF REU applications in January–March, MLH Fellowship or Ai2, open-source and Kaggle proof-of-work, and paid AI-training work that builds real resume experience now. The rejected-year plan is in this guide above, and the paid-now options are in AI training jobs.

Related guides

• Machine learning internships — the research-track and technical-ML cut: REU depth, the ML skill bar, and the quant pipeline.
• AI jobs with no experience — the proof-of-work method that beats GPA in internship screens, step by step.
• AI jobs for students — paid AI work that fits around classes while you wait for the next application window.