3

Supply, Demand, and Sector-Specific Considerations

Estimating demand for doctorates in computing indirectly, this chapter develops an overall estimate of past and current demand for computing PhDs and how that demand may evolve in the future. The challenge of assessing “demand” for skills and occupations is not simply a problem of insufficient data but also an issue of conceptual limitations. Demand is not a directly observable quantity—it reflects employer intentions, preferences, and expectations, all of which are shaped by evolving business needs, market conditions, and institutional contexts. Consequently, demand cannot be precisely measured, only inferred indirectly through proxies like job postings or employment levels, which offer at best an incomplete and delayed signal of the underlying economic dynamics.

INDUSTRY DEMAND FOR COMPUTING DOCTORATES

For the purposes of this report, “industry” is defined broadly to encompass all sectors of commercial activity, including technology, bioscience, aerospace, manufacturing, health science, finance, and others. The committee uses the term “industry” primarily to refer to the for-profit technology sector, which includes commercial activities related to software and services, semiconductors, and semiconductor equipment. This broad definition includes major computing technology firms, like Microsoft, Alphabet (Google), Amazon, Apple, and Meta (Facebook); large firms with significant computing efforts, like Boeing, Dell, Intel, IBM, HP, Cisco; and all manner of smaller computing oriented firms, including start-ups. This broad definition includes government contractors that provide

various computing services to the government, including defense contractors; as such, government contractors fall into the industry category of demand for computing PhDs.

Industry’s demand for computing doctorates is primarily driven by a desire to advance innovation and translation in areas where economic value can be derived, resulting in a demand for (1) particular computing skills associated with specific computing research topics—for example, artificial intelligence (AI) and security, numerical computing, computing systems, program verification, etc., and (2) the skill of conducting computing research itself—for example, to develop new computing technologies or to gain business-relevant insights from applied or basic science. Historically, the computing industry has sought computing PhDs with expertise in areas like VLSI chip design, software engineering, database systems, and graph algorithms. Today, expertise in areas like AI and security are in the greatest demand. For instance, computing industry job advertisements mentioning computing PhD-level experience and experience in AI have grown substantially over the past 5 years in particular. In February 2025, Amazon reported hiring most of the experts specializing in automated reasoning over the last decade, including 97 interns with doctorates in 2024 and hundreds in total (Lin 2025).

Indirect Estimates of Industry Demand

Estimating the demand for computing PhDs from industry is complicated by two main factors. First, the cyclical nature of industry makes it difficult to forecast how demand for computing PhDs may change in the medium and long term, and industry demand for particular subfields tends to change faster than academia can adjust its supply, due to the 5–8-year training time for new PhDs. However, industry experts agree that computing’s pervasive integration into the modern economy may now imply a more sustained and likely growing demand in the future, particularly in the areas of security and AI. This expectation of sustained growth in demand is supported by the Bureau of Labor Statistics (BLS) decadal projections, which forecast that demand for computing researchers across degree levels is projected to rise 25.6 percent by 2033 compared to a 4 percent increase for all occupations (BLS 2024). Such an increase corresponds to an estimated need for 8,300 additional computing researchers, up from 36,500 total in 2022 to 44,800 in 2032. However, based on results of the SED 2023, approximately 10,000 computing doctorates entered industry employment, suggesting that the BLS projections underrepresent demand in this area.

Second, actual demand growth may be much higher if historical trends over the past 20 years continue. For instance, computing experts interviewed by the committee indicated that in the early 2000s, Google began hiring computing PhDs in large numbers and today has about 10,000 at this level across all divisions. This demand corresponds to an average of 500 new hires per year for two decades, or about 25 percent

of the total supply in 2022 by a single company. Amazon began a similar effort in the mid-2010s and now has about 6,000 employees at the PhD level across all divisions, which corresponds to an annual average demand of about 30 percent of the total supply in 2022 by a single company. Notably, these levels of hiring encompass PhDs of all kinds, including computing, and, for large or multinational firms, the counts can be expected to include international PhDs.¹ Hence, while the actual fractions of each year’s supply of computing PhDs hired by these two firms is lower than these estimates, and there are other firms also hiring, they indicate clearly the large and growing scale of industry demand for computing PhDs.

Work Visas for International Computing PhDs

With more than 60 percent of recent computing PhDs being international students on a temporary visa, industry’s demand to hire such skilled workers on a more permanent basis is likely to place an additional burden on both employers and employees to secure work visas, and to navigate the complexity of U.S. immigration policy. For example, quotas for H-1B visas make industry employment prospects less certain for international students in computing because they must enter the H-1B lottery on an equal basis with the far more numerous visa holders without a PhD, in order to extend the 3-year work allowance provided by their student visa. In recent years, a substantial increase in H-1B visa applications has been observed (USCIS 2025) as more companies are sponsoring high-demand workers for permanent residency (green card). For fiscal year (FY) 2025, there were approximately 52,700 unique employers submitting registrations. While this is not a direct measure of PhD demand, this increase indicates that industry demand is not being filled by domestic students broadly. Large companies or those with diversified locations may be able to navigate this uncertainty more effectively than smaller or more localized companies. The complexity and uncertainty of U.S. immigration policy likely reduce the likelihood that international students completing their doctorate in the United States find long-term employment there.

Demand for Specific Subareas of Computing

Over the past 5 years, industry demand has dramatically increased for computing PhDs with expertise in AI and computer security, in particular, and large technology

___________________

¹ In a survey of science, technology, engineering, and mathematics (STEM) PhDs (2010–2016), among PhDs who work at Google, 78 percent are in computer science, followed by 6 percent in electrical and computer engineering, 4.5 percent in physics, 3.2 percent in neuroscience, and the rest scattered across fields. Among PhDs who work at Amazon, 55 percent are in computer science, 10 percent in electrical and computer engineering, 15 percent in physics, and 5 percent in mechanical engineering. Among PhDs at Microsoft, 58 percent are in computer science, 21 percent are in electrical and computer engineering, 6 percent are in physics, and 6 percent are in neuroscience.

companies report both substantial shortfalls in the supply of qualified researchers in these areas and substantial competition for these individuals from highly funded growth-stage start-up companies. This undersupply relative to demand has resulted in higher industry salaries in these specific areas, compared to salaries in other areas of computing. Academia’s ability to meet this demand would require increasing the number of computing PhD trainees within these areas of high industry demand, and historically, AI (29 percent), databases (9 percent), and security (4 percent) make up about 42 percent of all new computing PhDs employed in industry (Zweben and Bizot 2024, Table D4). However, an increase focused on a specific area of industry demand would take approximately 7 years to materialize due to the time required to complete a PhD.

Industry Demand for Computing Faculty

Industry’s growing demand for computing PhDs also extends to computing faculty at universities, with some tenured and tenure-track faculty taking part-time and sometimes even full-time temporary positions in industry, and a smaller portion leaving academia entirely. Historically, temporary positions tended to be at small technology start-ups, often university spin-off companies intended to commercialize new technologies developed by computing faculty, and these encompassed about 1 percent of computing faculty per year.

Today, temporary positions in industry are substantially more common, and most are located at medium or large technology firms. These positions often focus on applying faculty’s computing expertise to applied settings or conducting computing research on more applied topics with a commercial, but sometimes also a basic science, focus. Sometimes faculty continue to conduct and publish their research while in these positions, and sometimes they continue to advise PhD students at their home institution, but not always.

These faculty engagements with industry are also unevenly distributed across universities. Many, perhaps even most, academic departments have only a handful of such arrangements, while others, particularly those at universities located close to major technology hubs, report having as many as 25 percent of their faculty working temporarily in industry.

Such temporary positions come with both benefits and costs to academic departments and to overall computing PhD production. When these arrangements work well, they produce a kind of “hybridization” of research between industry and academia, with fluidity of roles between the two sectors. Due to the scale of resources that can be needed to advance the state of the art in some technologies, and a general perception that industry is at the forefront of innovation in some areas, engagement with industry enables faculty—and in some instances their graduate students—to conduct cutting-edge

research not possible within academic departments. In addition, the freedom to publish and additional financial compensation can make industry engagement attractive to some faculty.

At the same time, these arrangements may come at a cost to university departments. The faculty undertaking these positions are often among the most productive and prominent faculty, which can create significant pressure on a university’s ability to produce cutting-edge computing research and to maintain its current level of production of new PhDs in computing.

For computing PhD production, these arrangements present a potential seed corn risk, as tenured and tenure-track faculty are responsible for training new computing PhDs. Faculty spending part of their time in industry often have reduced time for academic duties like recruiting, training, and collaborating with computing PhD students. As such, having more faculty deeply engaged with industry may reduce a department’s overall capacity for PhD production. At the same time, the computing faculty the committee interviewed generally viewed these arrangements as net beneficial, as they can provide their faculty with funding, access to unique resources and collaborators, and opportunities for computing research not available in academic settings, and they seek other ways to mitigate the costs related to increased teaching and committee service loads on other faculty.

Changing Work of Computing PhDs in Industry

Historically, like most science, technology, engineering, and mathematics (STEM) PhDs working in industry, computing PhDs were hired into positions focused on applied science questions whose results could accelerate technological advancement for companies—for example, via development work that would lead to patentable inventions (Agarwal and Ohyama 2013). Most computing researchers working in industry would not share their research and development (R&D) results with the larger academic computing community, as firms viewed the findings as business secrets whose primary purpose was to provide a competitive advantage in the marketplace. Even so, there were notable and very influential exceptions to this model of industry computing researchers, such as Microsoft Research, Bell Laboratories, and IBM Research, which produced a wide range of foundational advances in computing, including multiple Nobel Prizes.

Over the past 20 years, however, the tenor of work by computing PhDs in industry has changed significantly, and it is now increasingly common for industry researchers to actively engage with academic research by publishing their findings in academic computing conferences and journals. Many firms point to Bell Laboratories and Microsoft Research as successful models for R&D in fast-moving areas of computing, although this

shift has business-relevant benefits as well. Some experts interpret this opening-up of computing research in industry as yet another indication of industry’s growing demand for computing PhDs. Allowing or encouraging computing researchers in industry to publish their findings in academic conferences and journals and collaborate with academic faculty represents a significant shift in norms relative to the more closed stance historically.

This open engagement with academic research reflects certain aspects of industry’s growing demand for computing PhDs and the unique contributions they make to firms. Beyond contributions to accelerate technological advancement alone, greater engagement by industry researchers with academic publishing can (1) provide reputational benefits that enhance a firm’s ability to attract top-quality researchers (Cockburn and Henderson 1998); (2) help recruit top-quality doctoral students to industry to perform high-impact applied research (Sauermann and Roach 2013; Stern 2004); (3) serve as a recruiting tool for prospective employees, in which current industry researchers network with doctoral students at the conferences; and (4) further competitive interests, by making it more difficult for rival firms to patent inventions based on the published work by a firm’s researchers.

All of these patterns appear to apply to the current dynamic in industry’s high-demand areas like AI (Ahmed et al. 2023), in which many major technology companies have a substantial presence at the top academic conferences each year. The degree to which these patterns appear in other, less competitive areas of computing research for example, human–computer interaction, high-performance computing, and computational biology—is not as clear.

ACADEMIC DEMAND FOR COMPUTING DOCTORATES

For the purposes of this report, “academia” is defined broadly to encompass all forms of faculty employment in higher education, including tenured and tenured-track faculty, teaching faculty, and research faculty, as well as staff researcher positions. This broad definition includes both research-focused institutions, including public and private PhD-granting institutions, and teaching-focused institutions, including primarily undergraduate institutions, liberal arts colleges, regional universities, community colleges, and master’s degree–granting institutions. These institutions are highly variable in their particular missions, sizes, curricula, and the populations they serve, and hence also in their demand for computing PhDs. As such, there are different dimensions of demands and needs for each type of faculty at each type of institution—research versus teaching positions at doctoral versus non-doctoral-granting institutions. The Computing Research

Association (CRA) Taulbee Survey recognizes tenure-track research faculty, non-tenure-track research faculty, teaching faculty requiring a PhD, instructors, and postdocs. The committee focused on both dimensions of academic demand—at research-focused doctoral-granting institutions and teaching-focused, non-doctoral-granting institutions. Notably, although some demand can be derived from information collected on the CRA Taulbee Survey of doctoral-granting institutions, very little data are available to address the demand at non-doctoral-granting institutions, including several minority-serving institutions and large, public, non-doctoral-granting institutions that serve the majority of undergraduates in computing majors (NASEM 2018a).

Academia’s demand for computing doctorates is primarily driven by a demand for individuals who can fulfill needs for teaching in degree-granting programs and/or conducting original research—specifically, (1) faculty who can teach college-level courses on computing, (2) faculty who can teach graduate-level courses on computing to master’s-level students, (3) tenured and tenure-track faculty who can teach graduate-level courses on computing to doctoral-level students and who can supervise the training of computing PhDs, (4) faculty who can conduct original research on advanced computing topics, and, to a lesser extent, and (5) non-tenure track researchers with titles like “research professor” or “staff scientist” who can focus on conducting and applying research approaches.

In contrast to the significant growth of newly declared undergraduate majors in computing and demand for computing courses as service courses across academic departments, the number of tenure-track faculty in computing grew by only 17 percent. To compensate, departments have increased the share of non-tenure-track teaching faculty and instructors, rising from 18 percent of all faculty in 2012 to 25 percent in 2022 (Taulbee, excluding postdocs), in order to meet undergraduate teaching demands. Moreover, in 2022, 31.7 percent of open faculty positions were for teaching faculty or instructors, who typically teach more classes and of larger size than tenure-track faculty. A recent report suggests this divergence has produced significant pressure on the tenured and tenure-track faculty in computing’s ability to train PhD students, due to the rapid growth in need for undergraduate teaching, which has only been partially mitigated by modest growth in teaching faculty positions (CRA 2017).

At most institutions, regular tenured and tenure-track faculty fulfill the institution’s need for all teaching activities at all degree levels and all research activities. In other cases, institutions may hire faculty who focus exclusively on teaching, or exclusively on research, and these faculty are typically not eligible for tenure. In 2022 approximately 65.7 percent of computer science department personnel were tenured or tenure-track faculty, while 23.8 percent were teaching faculty or instructors, 4.1 percent were non-tenure-track researchers, and 6.4 percent were postdoctoral fellows (Zweben and Bizot

2024). Doctorates in computing are nearly always required for tenured and tenure-track faculty at R1 and R2 institutions and at selective institutions with other classifications, such as liberal arts colleges, and are often required or at least preferred for other faculty-level or research staff positions.

Academic demand for computing PhDs also extends beyond departments that grant degrees specifically in computing, and computing PhDs are sometimes hired as faculty or non-faculty researchers by departments in other STEM fields, as well as in medicine (Hannak et al. 2023). Computing PhDs are sometimes also hired by interdisciplinary academic programs like the digital humanities, media and arts, business and information systems, and computational social science fields. Academia is a large and diverse source of demand for computing PhDs, and its demand is expected to continue to grow steadily over the next decade.

Indirect Estimates of Academic Demand

Reflecting academia’s lesser demand for computing PhDs compared to industry, in 2023, 13.8 percent of recent computing PhDs with definite employment plans after graduation took positions as tenure-track faculty while 7.8 percent took positions as non-tenure-track research or teaching faculty (NCSES 2024).

Over the past decade, however, a smaller share of all computing doctorates have entered tenure and non-tenure-track academic positions after graduation, with the overall number of computing doctorates remaining in academia increasing 64 percent from 2015–2022 and rising from 213 in 2015 to 348 in 2020. Thus, despite concerns that fewer computing doctorates are remaining in academia, the fact is that their numbers are growing. Moreover, computer science departments responding to the 2022 Taulbee Survey reported hiring 397 tenure-track faculty to fill 457 openings, or a hiring success rate of 86.9 percent. In addition, members of the committee reported that at their institutions, the quality of faculty hires has increased in recent years, assuaging concerns that faculty positions are being filled with lower-quality applicants. Together these findings suggest that faculty hiring remains strong and much of the demand for tenure-track faculty is satisfied, although hiring shortfalls may be more problematic outside of the top departments.

From 2011 to 2022, the number of tenured and tenure-track faculty at doctoral computing departments increased 17 percent, rising from 4,416 faculty to 5,152, or approximately 74 new faculty each year, net of attrition (Wapman et al. 2022). Atop this growth, the average number of new computing PhDs produced per tenured or tenure-track computing faculty grew from about 0.41 per year to 0.46 per year, or about a 12 percent rise in per-faculty production. However, from 2023 to 2025, doctoral departments of computing project that they want to increase the number of tenure-track

faculty by 11.4 percent and teaching faculty by 19.9 percent (Zweben and Bizot 2024). This estimated increase represents the hiring of approximately 500 new tenure-track faculty each year, which is consistent with the steady observed increase over the past 10 years. That is, academic computing departments project a continued increase in demand to hire top-quality computing PhDs as faculty in the near term.

In addition to hiring temporary visa holders who completed their PhDs in the United States, approximately 15 percent of tenured and tenure-track faculty at doctoral computing departments received their PhDs outside the United States (Wapman et al. 2022). This level is comparable to both engineering and other STEM fields. Immigration policy does not pose as complex a barrier as in industry for meeting demand for computing PhDs by hiring either U.S.-trained temporary visa holders or internationally trained doctorates in computing, although it may constrain those faculty’s subsequent interactions with other sectors—for example, industry and government.

Effect of Undergraduate and Master’s Teaching on Academic Demand

Nearly all computing faculty, regardless of institution type, engage in teaching within degree-granting programs. As described previously, the production of undergraduate degrees in computing has grown dramatically over the past decade. A separate but substantial increase has occurred in the number of non-computing degree students who enroll in computing courses, in order to “upskill” themselves for post-graduation employment. These students are not reflected in the undergraduate computing degree counts but nevertheless increase the demand on computing faculty for teaching. And, in parallel, the production of master’s degrees in computing has also grown substantially over the same period, particularly for international students seeking employment in the computing industry after graduation.

Computing departments at doctoral universities have responded to these multiple dramatic increases in the demand for teaching through a mixture of various measures, including increasing class sizes (particularly among introductory computing courses), restricting enrollment, increasing teaching loads, and hiring (non-tenure-track) teaching faculty (CRA 2017; NASEM 2018a). These actions likely mitigated some of the potentially large negative impacts on computing PhD production, as increased teaching time tends to reduce time available for research and doctoral training, but clear data are lacking for estimating how many more computing PhDs would have been produced if the growth in teaching demand had been more modest or better distributed.

Teaching faculty represent a growing source of demand for computing PhDs. Whether a PhD is required for this type of position varies by institution, and experts indicate that industry experience can be considered sufficient. In 2022, the CRA data

indicate that 3.1 percent of recent doctorates were employed as teaching faculty (Zweben and Bizot 2024). The focus of these faculty on teaching can effectively shield tenured and tenure-track faculty’s efforts in computing PhD production and research, by fulfilling some of the demand on departments for teaching.

From 2006 to 2021, the average number of teaching faculty in a doctoral computing department increased from 3.6 to about 8.6, or an increase by 240 percent (Zweben and Bizot 2024). Over the same period, the average number of tenured or tenure-track faculty in these same departments also increased, but more modestly, from 22.4 to about 33.6, or an increase of about 50 percent. These increases imply that in the average doctoral computing department, teaching faculty now make up about 24 percent of all computing faculty. Most of these increases occurred during the 2012–2021 period, concurrent with the dramatic increase in undergraduate majors (Figure 2-2), which more than doubled. Hence, since 2012, departments have seen student-to-faculty ratios increase by about 200 percent on average, because the growth in student populations has not been paralleled by growth in the number of faculty and instructors hired by doctoral institutions.

Experts in higher education indicate that the effects of this dramatic increase in student demand over a relatively short period of time are complicated, and academic departments are not able to “hire their way out” of the problem. Experts also indicate that this demand is unevenly distributed across subfields in computing, with particularly large demand by students in areas of interest to industry, such as AI (broadly defined) and security. The concentration of demand on particular subareas of the curriculum can induce even larger pressures on departments and the smaller subset of faculty with that expertise.

Training the Next Generation of Computing Professionals

Tenured and tenure-track university faculty in computing play several critical roles in the production of computing PhDs—producing new knowledge and innovation in computing, training new computing PhDs through research, and introducing new courses that often lead to new degree programs. This, in turn, provides essential training, leading to a competitive workforce.

Receiving a doctorate in computing requires that a student conduct and publish original research at the forefront of the computing field, under the supervision of a tenured or tenure-track faculty with a formal appointment in a computing department. This doctoral training process is a time-intensive and multi-year process, taking typically 5–8 years to complete. Computing PhDs who take tenure-track positions in academia then typically take another 5–8 years to begin producing a new generation of computing

PhDs, due to the time required to build and fund a productive research group and for the initial members to complete their training.

Without sufficient university faculty to train new computing PhDs, PhD production will likely decline, in addition to likely reducing the amount and range of innovation in basic and applied computing research. Similarly, if faculty capacity for training new computing PhDs is eroded—for example, because of increasing demand for faculty to spend time on non-research activities like teaching or service, or because of a significant number of faculty spending increased time conducting research in industry—PhD production may decline. Hence, shortfalls relative to demand in academia can exacerbate shortfalls in industry and government and thus require special attention.

Competition Between Academic and Industry Demand

Academic demand among doctoral institutions for computing PhDs tends to be broadly distributed across computing subfields, spanning all forms of computing scholarship and the computing curriculum, including theoretical computer science, programming languages, systems, cybersecurity, human–computer interaction, databases, software engineering, machine learning, and more. At the same time, academic demand does vary across subfields and tends to track industry interests to some degree (Zweben and Bizot 2024). For instance, today, academic demand coincides with industry demand for expertise in AI, security, and data science. In contrast, industry demand tends to be more focused on particular areas of economic innovation and competition, which can change over time more rapidly than academic demand.

A common concern voiced by computing academics is that industry’s larger demand, combined with access to industrial-scale data and compute resources, opportunities to publish, and higher pay, is making it increasingly difficult to meet academia’s demand for computing PhDs, particularly with respect to tenure-track faculty who can train new computing PhDs. At the same time, as shown above, the number of doctorates entering tenure-track faculty positions has increased over the past decade and appears to be tracking with overall demand for tenure-track faculty positions at doctoral institutions.

One indicator of where supply is meeting demand is the fraction of faculty openings at doctoral institutions that are filled. Among doctoral institutions, approximately 85 percent of open tenure-track faculty positions in computing were filled in 2021 and 2022 (Zweben and Bizot 2024), including 89 percent of advertised searches in AI/machine learning, 93 percent in security, and 88 percent in systems and networking in 2021 (Wills 2022). However, expert opinion is mixed on these interpretations, with some indicating that faculty positions in hot areas like AI going unfilled while others at top departments indicating that they have a strong pool of highly qualified applicants to fill

faculty positions. Although there are no large-sample data on the quality of faculty who are hired or the number of offers made to hire a single faculty member, members of the committee indicated that in recent years the quality of tenure-track faculty hired at top research-focused universities has increased, suggesting that concerns of doctorates being lured away from academia into industry may be felt more strongly at teaching-focused or less prestigious research-focused institutions. Notably, no such data exists for faculty hiring at teaching-focused institutions, such as minority-serving institutions and large, public non-doctoral universities, that train the majority of undergraduates in computing.

Computing academics indicated to the committee that the tenure-track faculty hiring success rate is often higher for higher-ranked doctoral institutions, while much lower both for teaching faculty positions at all institutions and for all faculty positions at lower-tier universities. If a newly graduated doctorate in computing would prefer a research position in industry over a faculty position at a lower-ranked institution, then competition from industry demand may be of greater concern for meeting demand at lower-ranked doctoral institutions with very high research activity and at other research institutions, all of which provide fewer non-pecuniary benefits for computing PhDs and often also lower salaries relative to high-ranked R1 institutions. As such, the salary differential between academia and industry may be particularly problematic for meeting the demand for computing PhDs at these institutions. For instance, one large system of non-doctoral universities reports a success rate of only 41 percent in its tenure-track faculty hiring in computing from 2021 to 2023, and even highly ranked doctoral institutions have reported needing to make half a dozen offers to secure one tenure-track faculty hire. Reports suggest that industry’s growing demand for computing PhDs may be distorting the academic market in ways that are not uniform, having impact in subfields of particularly high industry demand and for institutions close to technology hubs, and on institutions outside of the high-prestige R1 institutions (Ahmed et al. 2023; Nix et al. 2024).

Another measure is faculty turnover and, in particular, computing PhDs leaving academia for industry. Data indicate that there has been increasing turnover among computing faculty at doctoral institutions in the past 10 years. From 2011 to 2020, 1,461 tenured or tenure-track computing faculty at doctoral institutions left academia (Wapman et al. 2022), inclusive of faculty retirements, while 2,197 were hired, implying a net increase of 736 faculty. Hence, to increase the size of the computing professoriate at doctoral institutions by 1 faculty required 2.98 new hires on average. However, this rate of faculty attrition at doctoral institutions is not significantly different from academia as a whole, or from STEM fields as a group (Spoon et al. 2023), suggesting that industry demand between 2011 and 2020 did not draw new or established faculty away from

academia at unusually large rates. (Even so, high turnover rates among tenured and tenure-track faculty remain concerning for a field in which current computing PhD production appears inadequate to meet current, let alone future, demand.)

NATIONAL SECURITY AND OTHER GOVERNMENT DEMAND FOR COMPUTING DOCTORATES

Historically, demand for computing PhDs in government has primarily come from agencies, centers, and laboratories focused on national security and from national laboratories. For this report, “government” demand encompasses that of the Department of Energy’s national laboratories (both the science and weapons laboratories), University Affiliated Research Centers, federally funded research and development centers, agencies of the Intelligence Community that conduct computer science research (e.g., the National Security Agency, the Central Intelligence Agency, and the National Geospatial-Intelligence Agency), and laboratories directly under the Department of Defense (e.g., the Army Research Laboratory, the Naval Research Laboratory, and others). However, demand for computing PhDs also extends to other parts of the U.S. government, including the National Institute of Standards and Technology, the U.S. Census Bureau, NASA, and other agencies. Activities by private firms that contract with government institutions, such as defense contractors, are considered part of industry demand in this report.

Demand for computing PhDs in government shares many characteristics with demand for academic research faculty positions. However, there are two aspects that make this demand different: (1) government positions in general tend to have lower salaries than industry, and lower even than many academic institutions, and (2) national security and national laboratory positions in particular have unique difficulties due to the sensitive nature of the research. Given the nature of the work, even small shortfalls in supply relative to demand can have large societal consequences.

Among computing PhDs with definite employment plans after graduation, 2.8 percent had plans for employment in government positions, including positions at national laboratories, or nonprofit organizations, such as research institutes. Government experts indicate that demand for computing PhDs in government is not being currently met.

Salary Constraints

Both government and academic institutions have a restricted ability to compensate computing PhDs competitively with industry. For all positions administered or funded by the

U.S. government, compensation is determined by policy or law and rarely allows much flexibility. In fact, government institutions seeking computing PhDs can be at a disadvantage even to academic institutions, as some academic institutions can increase compensation through endowments or allow joint employment with industry. In general, federal law prohibits government institutions from similar initiatives.

Security Clearances

Research positions within the national security community can additionally require a security clearance. As a minimum, U.S. citizenship is required for a security clearance, and a researcher must be currently eligible to hold a security clearance. Eligibility becomes stricter as the level of clearance increases. For example, a Top Secret clearance having access to Sensitive Compartmented Information requires a background and credit check, access to tax records, and may require a polygraph or drug test. Not every U.S. citizen can successfully qualify for the required security clearance. The requirement of a security clearance for a national security research position thus restricts the pool of potential hires to domestic PhD recipients.

Given that a growing majority of new computing PhDs are international students on temporary visas (Figure 2-3), the potential supply of computing PhDs for meeting government demand in general, and demand within national security in particular, is modest, being restricted to domestic computing PhDs who can meet the additional clearance requirements.

Demand for Specific Subareas of Computing

Experts indicate that government and national security demand also differs from industry demand in the particular subareas of greatest interest, including supercomputing research and quantum computing research. This difference may mean that government demand is not necessarily supplanted by the larger industry demand, although data does not exist that can directly estimate the degree to which this is true.

RECOMMENDATIONS

The committee conducted an in-depth analysis of hiring trends across computing disciplines, informed by institutional data and expert perspectives. It found that while the supply of doctoral recipients in computing has grown in recent years, significant challenges remain in meeting hiring demands across various institutional contexts.

Specifically, the committee concluded that (1) in high-demand areas such as AI and cybersecurity, nearly all institutions face difficulty in recruiting qualified doctoral recipients; (2) teaching-intensive universities encounter persistent challenges in filling both tenure- and non-tenure-track computing faculty positions; and (3) in other areas of computing, the growth in the number of doctorates has generally kept pace with demand at leading research-intensive universities and in industrial R&D. These conclusions inform the following recommendations aimed at addressing these workforce pressures.

Ensuring a Continued Robust Supply of Doctorates in Computing

This picture points to the need for targeted interventions in cases where there is unmet demand, along with actions to ensure continued robust supply of doctorates in computing.

Recommendation 1: Federal agencies supporting research and education, universities that perform research and education, and the larger computing industry should at a minimum sustain and, if possible, expand doctoral fellowships and assistantships in computing independently and in partnership. At the same time, they should increase the fraction of fellowships and assistantships awarded to domestic doctoral students studying computing.

This recommendation aims to achieve two coupled goals:

1. Ensure a sufficient supply of talent to sustain the United States’ world-class research and teaching workforce and ensure a sufficient supply of talent in areas that are important for U.S. economic competitiveness and national security.
2. Reverse the decline in the fraction of domestic students pursuing a doctoral degree in computing, ensuring supply that is more robust to changes in global labor markets and helping meet demand for national security roles that require U.S. citizenship.²

There ought to be no problem finding additional highly qualified international and domestic candidates to meet these goals, given the enormous growth in the number of U.S. undergraduates earning computing degrees and the very low acceptance rates in the most selective doctoral programs. Additional steps will be needed to attract more domestic students to graduate study including enhanced opportunities for undergraduate

___________________

² Another way to grow the number of PhDs with U.S. citizenship to fill national security roles would be to expedite paths toward citizenship. Changes to national immigration policy are, however, out of scope for this report.

students to become aware of, be exposed to, and experience computing research, as well as paying attention to the opportunity costs of attending graduate school. It may also require changes in recruiting and admission, and additional attention to doctoral student retention.

Better Measurement of Supply and Demand

More detailed information would help the National Science Foundation (NSF) and others supporting doctoral education in computing to detect specific supply shortfalls and tailor appropriate interventions to mitigate them.

Recommendation 2: The National Science Foundation should, in collaboration with other federal statistical agencies and the Computing Research Association, collect additional data to better measure supply and demand for doctoral degrees in computing and provide the information needed to tailor future interventions.

Recommendation 2-1: The National Science Foundation should collaborate with other federal statistical agencies and the Computing Research Association to identify and collect additional data on the supply of doctoral degrees in computing.

Data are not currently being collected in aggregate to reflect whether there is a retention problem within computing doctoral programs, either compared to other fields or to past retention trends, and the committee heard no briefings to this effect. Additionally, more information is needed to assess the in-flow and out-flow of doctoral applicants and recipients, especially for domestic students. Data to collect would include the following: (1) number of applicants (preferably distinct—which would require a trusted third party to cross-link by name or unique identifier) to doctoral programs in computing each year, (2) number of students admitted to doctoral programs in computing each year, (3) number of years to complete doctoral studies, and (4) completion and attrition rates for doctoral programs. The CRA Taulbee Survey (2023) has recently added questions that address the first two data points suggested, but longitudinal data are needed to assess the trends in applications and admissions and allow for appropriately tailored interventions in the future.

Recommendation 2-2: The National Science Foundation should collaborate with other federal statistical agencies and the Computing Research Association to identify and collect additional data that would allow them to better measure and project academic hiring demand for doctoral degrees in computing at non-doctoral-granting institutions.

Although some data are available on the success of faculty searches and hiring at doctoral-granting institutions, these alone do not show a complete picture of the challenges in meeting demand for faculty, especially those teaching undergraduate courses. Briefers to the committee suggested that it has not been possible to hire the faculty required to meet the enormous demand for teaching computing at various institution types. Data to collect would include faculty hiring data from an expansion of the current CRA Taulbee Survey to non-doctoral-granting institutions; NSF could encourage or require institutions that receive research funding from NSF to participate.

Recommendation 2-3: The National Science Foundation Survey of Earned Doctorates should add a set of questions to track the supply of and demand for PhDs in computing and related fields, disaggregated by subfield.

Differentiating doctoral recipients by research area, interest, capability, and citizenship will help ensure demand is met while avoiding oversupply. Doctoral students are not necessarily interchangeable between subdisciplines or topic areas, students interested in research careers may not be seeking faculty positions at teaching-focused institutions, and not all students may be interested or eligible to pursue national security careers.