The Credential Boom: Certificates, Bootcamps, and the Reskilling Economy

A boom and a leak, in the same industry

The reskilling economy has two honest stories running at once, and most coverage of it tells only one. The first is a genuine boom in scale and revenue. Coursera closed 2025 with 197 million registered learners and $757.5 million in full-year revenue, up 9 percent on the year before, a learner base that has roughly doubled in about four years. Google's Career Certificate program passed 1 million graduates worldwide, more than 350,000 of them in the United States. Demand for generative-AI specific courses is the sharpest edge of the boom: Coursera reported more than 8 million GenAI-course enrollments in 2025, up 195 percent year over year, across roughly 700 courses, with 10 million cumulative enrollments in the category. By any measure of volume and dollars, the market for short, non-degree credentials is one of the fastest-growing corners of education.

The second story is a leak running underneath the first, and it explains why the industry's own numbers can be simultaneously true and misleading. Registration is not completion. Completion is not competence. Competence is not employment. Each of those is a different, measurable thing, and the gap between the first and the last is where the reskilling economy's credibility is actually decided. A course that 8 million people enroll in and that 400,000 finish is a real phenomenon at both ends, but it is two different phenomena, and the marketing language for the industry almost always quotes the first number.

This study holds those two stories side by side rather than picking a winner. It follows the money and the enrollment first, then follows the leak through completion rates, then asks the sharpest question the evidence can actually answer: does any of it raise a worker's pay. The strongest single piece of evidence in the whole space, a December 2025 Brookings analysis of 37.7 million resumes, answers that question with more precision than almost anything else published on reskilling to date, and the answer is neither the industry's slogan nor its critics' dismissal. It depends, specifically and measurably, on what the credential is.

Registration is not completion. Completion is not competence. Competence is not employment. The reskilling economy's credibility is decided in the gaps between those four words.

The scale is not in dispute

Start with what is well established. Coursera's 2025 results show a platform still growing but decelerating: 197 million registered learners against roughly 100 million a few years earlier, and $757.5 million in revenue growing at 9 percent, a single-digit pace after several years of faster expansion. That deceleration matters as context for everything that follows: even the category leader's revenue growth has slowed even as enrollment volume keeps climbing, a sign that the market is maturing from land-grab to something closer to a normal subscription business.

Google's Career Certificate program offers the cleanest single-employer credential story. It passed 1 million graduates globally by 2025, with more than 350,000 of those in the United States, and Google reports that more than 70 percent of US certificate holders self-report a positive career outcome, such as a new job, a promotion, or a raise, within six months. That outcome figure needs a caveat stated plainly: it is graduate self-report, collected by the company that sells the certificate, with no control group and no independent verification. It belongs in the same category as a satisfaction survey, useful as a signal of sentiment, not as proof of causal wage effect. The 1-million-graduate milestone itself, by contrast, is a hard count and stands as established.

Beyond the two clearest cases, the picture thins. Udemy reports roughly 85 million learners, but that figure is secondary, reported through company communications rather than an audited filing comparable to Coursera's. And a genuine data gap sits at the center of the credential economy: there is no reliable, independently verified public count of how many people hold an active AWS or Microsoft certification, despite those two vendors running some of the most widely referenced technical credentials in hiring. Vendors publish exam volumes and marketing figures inconsistently, and no third party audits them the way Coursera's revenue is audited for a public filing. That gap should temper any claim about the total size of the credentialed technical workforce; for the largest single credential category in tech hiring, no one outside the vendors actually knows the number.

The completion leak

The most durable finding in this entire literature, corroborated across platforms and years rather than resting on one study, is that completion rates for free, self-paced online courses run in the single digits to low teens, commonly cited in a 3 to 15 percent range. That is not a temporary embarrassment the industry has been closing; it has held roughly steady since the first MOOC-completion studies over a decade ago, which is why it counts as established rather than emerging.

What moves the number is stakes, and the pattern is close to linear. Paid, structured courses, where a learner has spent money and faces a visible sunk cost, complete at roughly 60 percent. Cohort-based programs, where a learner is moving through material alongside peers on a shared schedule, complete near 72 percent. Employer-mandated corporate training, where completion is a condition of the job rather than a personal choice, completes at a similar 72 percent. The gap between the top and bottom of that range, roughly a five-fold difference between a free self-paced course and a mandated corporate one, is the clearest evidence available that "millions of learners" and "millions of people who finished and can use the skill" are not the same claim, even when both numbers describe the same platform.

This has a direct bearing on how to read every enrollment headline in this study. Coursera's 8 million GenAI enrollments in 2025 is a real and fast-growing number, but it is an enrollment number, and the completion-rate evidence says a large majority of a free or lightly-priced cohort of that size will not finish. None of the sources in this study's spine publish a GenAI-specific completion rate, which is itself worth naming: the fastest-growing course category in the reskilling economy is reported to the public almost entirely through its top-of-funnel number.

Micro-credentials: recognized on paper, uneven in practice

Employers say they are on board. In one vendor-run survey, about 68 percent of employers report they recognize micro-credentials as a legitimate signal in hiring, a figure the industry cites often as proof the market has arrived. But a second data point sits right next to it and complicates the story: 58 percent of hiring managers say they still weight the reputation of the issuing institution heavily when they evaluate any credential, meaning the badge itself is doing less independent work than the survey headline suggests. A micro-credential from a recognized university or a name-brand technical program appears to travel very differently from an identical-sounding credential issued by an unknown provider, even when the stated skill content is the same.

The OECD's 2023 review of micro-credentials for lifelong learning and employability reaches a similar, more cautious verdict from the policy side: their labour-market value is real but uneven, and the field lacks standardized definitions, quality assurance, or a common framework for what a given badge actually certifies. That is a structural problem, not a marketing problem. Two employers can look at the same certificate and read entirely different things into it, because no shared standard tells either of them what skill level it represents or how it was assessed.

Hold both readings together rather than resolving them. It is true that employer receptivity to micro-credentials has risen and that a majority now say they recognize them. It is also true that recognition without standardization means a micro-credential's value is still substantially borrowed from the reputation of whoever issued it, which is close to the opposite of what a standardized skills signal is supposed to provide. The debate about whether micro-credentials are a genuine parallel currency to a degree, or a weaker signal wearing a degree-like name, remains open on the evidence in hand.

The bootcamp shakeout

Coding bootcamps are the part of the reskilling economy that has already been tested against a real market cycle, and the test did not go the way the sector's early marketing promised. The commonly quoted figures, roughly 90 percent placement and a $69,000 median starting salary, are marketing claims rather than audited outcomes, and it matters that almost no one still publishes audited numbers to check them against. The Council on Integrity in Results Reporting, the industry's own audit standard, now has only three schools still publishing CIRR-audited outcomes; those audited figures put placement closer to 71 percent and median first salary in a range of roughly $66,000 to $70,700, both meaningfully below the marketing claims, though still a real and positive outcome for the graduates the audited programs track.

The shakeout was not gradual. 2U, the parent of the edX bootcamp business, pivoted away from the bootcamp model entirely in December 2024, and at least a dozen US bootcamps closed between 2023 and 2024, including Codeup, Kenzie Academy, Rithm School, Code Fellows, and Women Who Code among others. That is not a story of the whole model failing; the sector graduated roughly 69,000 US bootcamp students in 2024 alone, and the number of bootcamp programs worldwide has grown from fewer than 100 in 2015 to more than 600 today. It is a story of a specific version of the model, the version built on unaudited placement claims and low barriers to entry, running into a labor market that stopped absorbing new graduates at the pace bootcamps needed, and breaking first.

The bootcamp shakeout was not the whole model failing. It was the version built on unaudited placement claims running into a labor market that stopped absorbing graduates at the pace it needed.

The reskilling imperative, and a myth inside it

The pressure behind the boom is not manufactured. The World Economic Forum's Future of Jobs 2025 survey, which polled more than 1,000 employers across 55 economies, finds that 39 percent of workers' core skills are expected to change by 2030, down somewhat from 44 percent in the 2023 edition of the same survey, a small softening but still a large fraction of the global skill base in flux over five years. The same report projects that 59 of every 100 workers worldwide will need retraining by 2030, and that roughly 11 of those 59 are unlikely to receive any, implying more than 120 million workers globally sit at medium-term skills risk on the survey's own numbers. These are stated by the WEF as a projection from an employer survey, not a measured outcome, and this study treats them accordingly, as a contested but well-sourced scenario rather than a fact already in the record.

One popular justification for the urgency deserves direct scrutiny: the claim that skills now have a five-year half-life, so roughly half of what a worker knows becomes obsolete on a fixed five-year clock. That figure is folklore rather than a measured constant. It traces back to a single 2011 book's framing rather than to a replicated empirical study, and it does not hold up against a simple test: Dice's 2025 ranking of the fastest-growing-salary technical skills includes COBOL, a language from 1959, and Ruby, from 1993, both still commanding rising pay 35 to 66 years after their creation. A skill can decay fast in one domain and barely decay at all in another; there is no single half-life, and the reskilling industry's frequent use of the five-year figure as a universal law is not supported by the evidence in this study's spine.

Employers say the need is real even as they have pulled back on funding it. US corporate learning spend fell to $98 billion in 2024, down 3.7 percent, with per-learner spending falling from $954 to $774, before rebounding to $102.8 billion in 2025. That is a genuine tension worth naming rather than smoothing over: the same employers telling surveys that reskilling is essential cut the budget for it in the very year the pressure was rising, and only partially restored it the year after.

What actually pays: the Brookings anchor

The single strongest piece of evidence in this study, and the cleanest anchor for the whole debate, is a Brookings analysis published in December 2025 that used machine-learning methods to read 37.7 million US worker resumes rather than relying on self-reported outcomes or a small survey sample. Its central finding cuts directly against the industry's undifferentiated marketing: not all credentials pay, and the ones that do pay in a specific, measurable pattern.

A first non-degree credential raises wages by about 3.8 percent when it is relevant to the job the worker actually holds, versus roughly 1.8 percent when it is not, more than double the return for relevance alone. Occupation-specific certifications, concentrated in IT, healthcare, and the skilled trades, pay the most of any credential type measured: for non-college workers, a first relevant certification is worth about 7.1 percent, with each additional relevant certification adding roughly another 3.5 percent. Generic badges and stacked certificates, the low-tier, unrelated credentials that make up a large share of what platforms sell in bulk, pay far less across the board, and the study finds that continuing to accumulate certificates that are nominally relevant but low-tier and unfocused can carry an actual wage penalty near 3.0 percent, evidence that stacking credentials can actively work against a worker, not merely deliver a weaker version of the same strategy.

One more pattern in the Brookings data deserves its own emphasis: the wage premiums it measures are 1.5 to 2 times larger for non-college workers than for college graduates earning the same credential. That is the opposite of what a purely credentialist reading of the reskilling boom would predict. A relevant certification does more for a worker's pay precisely when that worker does not already hold a four-year degree, which suggests occupation-specific credentials function less as a signal layered on top of existing education and more as a genuine substitute pathway into gated, better-paid work.

Not all credentials pay. A job-relevant certification is worth roughly double an unrelated one, and stacking low-tier badges can carry an actual wage penalty rather than a smaller reward.

India: the skilling-employability gap, at national scale

India offers the clearest and largest version of the same gap between training volume and employment outcome. The National Skill Development Corporation, through the flagship Pradhan Mantri Kaushal Vikas Yojana program, trained 3.3 crore, roughly 33 million people, over ten years, backed by a fourth phase carrying an outlay of Rs 12,000 crore. Measured purely as an enrollment and training-delivery number, that is one of the largest workforce-skilling efforts ever run by any government.

Placement is where the picture splits, and the spread is the actual story rather than any single number in it. Placement rates by phase of the program have themselves varied widely, from about 18.4 percent to 23.4 percent to 10.1 percent across different rounds, and the two headline aggregate figures now cited for the program diverge sharply: official government reporting cites a placement rate near 54 percent, while independent policy analysis of the same broad cohort puts the figure closer to 22 percent. Neither number is fabricated; they appear to be counting different things, official figures likely including a broader definition of positive outcome such as self-employment or further training, independent estimates applying a stricter test of formal wage employment. But a spread of that size, from 54 percent down to 22 percent for describing the same national effort, is itself evidence that outcome measurement in mass skilling programs remains contested rather than settled, in India and, on the evidence assembled in this study, everywhere else the reskilling economy operates too.

The limits of this reading

Several limits bound this study. The Brookings resume analysis, the strongest single piece of evidence here, is itself an observational study of correlations in resume and wage data; it controls carefully for relevance and credential type, but it cannot fully rule out that workers who choose relevant, occupation-specific certifications differ from other workers in ways the data does not capture, such as prior motivation or career focus. The wage premiums it reports are the most rigorous available, not a randomized-trial-level proof of causation.

Several figures in this study are vendor-reported rather than independently audited, and are tiered down accordingly: Coursera's GenAI-enrollment growth of 195 percent comes from the company's own Global Skills Report rather than its audited financial results, Google's 70-percent-positive-outcome figure is graduate self-report with no control group, and the 68-percent employer-recognition figure for micro-credentials comes from a vendor survey rather than a neutral third party. Udemy's 85-million-learner figure is likewise secondary and unaudited. None of these figures is treated in this study as equivalent to the audited Coursera revenue release or the Brookings resume analysis, and each is labeled accordingly in the findings below.

A genuine data gap runs through the technical-certification market specifically: there is no reliable, independently verified public total for active AWS or Microsoft certification holders, despite those credentials sitting at the center of technical hiring. And India's placement-rate dispute, a spread from 22 to 54 percent for the same program, is presented here as an open measurement disagreement rather than resolved in either direction, because this study did not have access to the underlying survey methodology of either the official or the independent figure.

What survives all of that is a narrow, well-founded claim rather than a verdict on the whole reskilling industry. The boom in enrollment and revenue is real and independently corroborated. The completion leak beneath it is durable and well established. And the payoff, where it exists, is concentrated, specific, and conditional on relevance, not a general property of holding more credentials. A worker or an employer reading this evidence should expect a job-relevant, occupation-gated certification to be worth pursuing, and should treat a stack of generic, unrelated badges as, at best, a weak signal and, on the strongest evidence available, occasionally a negative one.