Data Science Is Growing 34.6%. The Entry Door Is Still Getting Narrower.
The latest BLS projections point to strong demand through 2035. New workforce evidence shows why that does not translate into an easier first job.
The newest US employment projections contain a number that looks like uncomplicated good news for anyone trying to enter data science: employment for data scientists is projected to grow 34.6% from 2025 to 2035.
That puts data scientists among the ten fastest-growing detailed occupations in the economy. The Bureau of Labor Statistics also expects computer and mathematical occupations as a group to grow 7.3%, more than twice the 3.5% rate projected across all occupations.
So why does the first data job still feel unusually hard to get?
Because occupational growth and entry-level access are different measurements. One describes how many jobs may exist. The other describes who can pass through the hiring funnel, what employers expect on day one, and whether organizations still provide a place to learn the work.
That distinction matters more as AI absorbs the routine tasks that used to train junior analysts and data scientists.
What the 34.6% actually says
The August 2026 BLS release projects US data-scientist employment to grow from 2025 through 2035. BLS attributes demand across computer and mathematical work largely to the spread of digital tools and AI systems. Organizations still need people who can develop, improve and apply those systems to business decisions.
That is a demand signal for the occupation. It is not evidence that:
- 34.6% more entry-level openings will appear;
- every role carrying the title “data scientist” will grow equally;
- a degree or bootcamp certificate will be sufficient;
- hiring will be smooth in every year, industry or geography; or
- the tasks inside the occupation will stay the same.
The projection describes a stock of employment across experience levels. A hiring funnel describes a flow of candidates into specific jobs. You can have a fast-growing occupation and a difficult entry point at the same time.
The missing rung in the ladder
Entry-level work has traditionally served two purposes. It produces useful output, and it gives a novice repeated exposure to the small decisions that eventually become professional judgment.
In data work, those early tasks might include:
- cleaning a messy extract;
- checking row counts after a join;
- writing a first-pass query;
- documenting a metric;
- reproducing an existing dashboard; or
- preparing a baseline analysis for review.
AI can accelerate many of these tasks. That is useful. But if an employer removes the task without redesigning the learning around it, the organization also removes a training loop.
A June 2026 World Economic Forum and PwC study helps quantify that pressure. It drew on global data, senior-leader interviews and responses from more than 9,000 entry-level workers across 48 countries. Its findings include:
- 37% of young workers are in occupations with medium-to-high exposure to AI-driven task change;
- entry-level occupations in the highest AI-exposure quartile show about 2.2 times more net skill change than those in the lowest-exposure quartile;
- 28% of entry-level workers believe half or fewer of their current skills will remain relevant in three years; and
- 45% report spending more time working because of AI, a useful counterweight to the assumption that automation immediately reduces workload.
These results do not prove that AI caused fewer junior data-science openings. The research mixes occupations and countries, and the worker responses capture perceptions as well as observed labor-market data. What it does show is that the entry layer is being redesigned while people are trying to enter it.
A growing market can still raise its entry bar
Imagine a data team that once hired two junior analysts and one senior analyst.
The juniors prepared datasets, ran standard queries and produced initial charts. The senior analyst reviewed assumptions, resolved ambiguity and presented recommendations.
Now the team adopts capable AI tools. One junior analyst can produce more first-pass work, while the senior analyst can review a larger volume. The company may still grow the team because demand for analysis is rising. But the next junior hire is expected to do more than produce syntax. They must also:
- recognize when the generated query uses the wrong grain;
- test whether a metric definition matches the business question;
- challenge a plausible but unsupported explanation;
- protect sensitive data; and
- communicate uncertainty to a decision-maker.
The total number of data jobs can rise even as the minimum acceptable level of judgment rises. That is not a contradiction. It is a composition change.
What this changes for an early-career portfolio
A portfolio built only to prove that you can generate code is losing signal. Employers can already generate plausible SQL, Python and charts cheaply. The stronger portfolio proves that you can make a result trustworthy.
For one project, show the full chain:
1. Define the decision
State who needs the analysis and what decision it should inform. “Predict churn” is a modeling task. “Identify which retention action should be tested next” is a decision problem.
2. Document the grain
Explain what one row represents, what uniquely identifies it and how joins change that grain. Include at least one check that would catch duplication or leakage.
3. Build a baseline before complexity
Compare an advanced model with a simple baseline. If the more complex approach wins, show where and by how much. If it does not, say so.
4. Keep an evidence trail
Record data sources, assumptions, exclusions, validation rules and changes. If you used AI, identify where it helped and how you verified the output.
5. Write the decision memo
End with a short recommendation, the uncertainty around it and the next test. A notebook is evidence of activity. A decision memo is evidence of professional judgment.
The employer side is not optional
Candidates cannot solve this transition alone. If companies demand senior judgment from people who have never been allowed to build it, they create a circular requirement: experience is necessary to get the role, but the role is necessary to get experience.
The WEF/PwC framework places responsibility across four areas: job access, job design, talent pipelines and education alignment. For a data team, that can translate into concrete practices:
- preserve low-risk work where junior employees can make and correct decisions;
- pair AI-enabled production with structured review;
- make evaluation criteria explicit;
- reward documentation and error detection, not only speed;
- use apprenticeships, internships and scoped rotations as real production pathways; and
- measure whether automation is improving learning or merely concentrating work among senior staff.
AI can shorten the time required to produce an answer. It cannot eliminate the need to learn why an answer deserves trust.
What would make me wrong
This argument would weaken if future hiring data showed that the increase in data-science employment is concentrated at the entry level, with stable qualification requirements and broad access across industries and regions.
It would also weaken if longitudinal evidence showed that AI tools reliably help novices acquire judgment faster without reducing mentoring, feedback or early-career openings.
We do not have that evidence yet. The BLS release is a national ten-year projection, while the WEF/PwC study covers a broad international set of entry-level occupations. Neither directly measures the US entry-level data-science funnel. The conclusion here is therefore an inference from two compatible signals, not a causal estimate.
The practical conclusion
Data science is not disappearing. The latest official projection points in the opposite direction.
But “the occupation is growing” is not the same as “the path into it is easy.” The opportunity is moving toward people who can combine technical fluency with problem framing, verification, domain context and clear communication.
For early-career practitioners, the goal is not to compete with AI at producing the first answer. It is to demonstrate that you can determine whether the answer should be trusted and what should happen next.
For employers, the challenge is more structural: do not automate away the apprenticeship and then complain that experienced talent is scarce.
Sources
Found this useful? Passing it on to someone who builds is the best way to help the publication grow.