Goldman Sachs’ latest AI Adoption Tracker puts a sharp number on an unusual labor split in the U.S. economy. AI is now estimated to be eliminating about 11,000 jobs a month on a net basis, while data center construction has been moving the other way. Since 2022, that segment has added 212,000 jobs and is still creating roughly 9,000 positions each month.
The jobs in short supply are not the ones many people first think of. Goldman’s description points to electrical contractors, heating, ventilation and air conditioning technicians, commissioning staff, utility workers and commercial construction labor rather than machine learning engineers. For people looking at a career tied to AI infrastructure, the hiring demand is centered far more on physical build-out and facility uptime than on software.
Goldman’s numbers show office job losses and infrastructure hiring moving in opposite directions
In the August edition of the AI Adoption Tracker, Goldman Sachs economists Sarah Dong and Joseph Briggs said AI is net eliminating around 11,000 U.S. jobs a month. The previous version of the estimate was 16,000. Over the past three years, AI-linked layoffs have cumulatively reached 136,000, according to the report cited in the article.
In the same report, data center construction is shown adding 212,000 jobs since 2022, with around 9,000 more still being created every month. Goldman characterizes those openings as jobs in power work, HVAC, utilities and commercial construction. The takeaway in the article is simple: money is flowing out of some office-based functions and into job sites, and the skills required are very different.
A separate Goldman Sachs report from July adds a broader capital spending backdrop. It estimated global AI infrastructure investment at about $7.6 trillion between 2026 and 2031. In that assessment, the bottleneck has shifted away from chips and toward power, equipment and labor. In some core markets, the queue to connect data centers to the grid is estimated at eight to 12 years, which is longer than the replacement cycle for GPUs.
Two job tracks come first: construction and operations
The article argues that anyone trying to enter AI data center work needs to separate the market into two categories before thinking about salary or training.
Construction-phase roles are tied to building the site itself. That includes electricians, HVAC technicians and commissioning engineers. Commissioning in this context means testing and validating that power and cooling systems can hold under full-load and fault conditions before handover. The work is project-based by nature. When one build ends, workers often move to the next site.
Operations-phase roles begin after the facility goes live. They are tied to keeping the site running around the clock, usually in shift schedules. Jobs in this group include data center technicians, critical facilities engineers, network operations center analysts and security personnel. The work location is more stable, but the rhythm is often 24/7 operations.
The article says the pay scales, entry barriers and job duration are very different across the two tracks, and that difference matters more than many new entrants realize.
Six roles and how the pay ladder looks
1. Data Center Technician
This is presented as the most common entry role on the operations side. The U.S. average annual pay is about $67,900, with the 25th percentile at $55,500 and the 75th percentile at $83,500. For workers with less than one year of experience, the entry hourly rate is about $24.8.
The job usually covers racking servers, running cabling, swapping drives and performing hardware diagnostics.
2. Senior Data Center Technician
This is also an operations role, with pay around $75,000 to $110,000. Responsibilities expand into shift leadership and change management.
3. Critical Facilities Engineer
Critical facilities engineers sit on the operations side as well, with base pay around $117,000 to $147,000. They are responsible for systems that cannot fail without consequences, including uninterruptible power systems, generators and chilled water equipment.
The article says many workers in this role hold associate or bachelor’s degrees in mechanical or electrical fields, though some come up through apprenticeships.
4. Commissioning Engineer
This is a construction-phase role. Average base pay is about $112,000, entry level is near $95,000, and senior roles on hyperscale projects can exceed $215,000. In markets with heavy project concentration, the article says the salary ramp on this path can outpace operations jobs.
5. Electricians and HVAC Technicians
These are core construction roles. Using ZipRecruiter data from June 2026, the article puts average apprentice pay at about $25.98 an hour. Verified electrician apprentices are paid in tiers, with hourly wages ranging from about $26.7 to $44.5. In major markets, union electricians are listed with base wages of roughly $60 an hour.
6. Operations Manager
This operations-side role is listed at about $117,000 to $198,000 a year. For senior engineers and director-level staff, stock compensation can push total pay above $200,000.
Three entry routes: apprenticeships, certifications and lateral transfers
Route A: Apprenticeships for people starting from zero
The first route is paid apprenticeship. The point is to learn on the job without paying for certifications upfront.
The article lists several examples. Amazon offers a 12-month paid technical apprenticeship. Microsoft typically works with local community colleges to run its Datacenter Academy when it builds a new facility. Google’s STAR Program is described as training people from underserved communities for construction and technical trades.
Applicants are also urged to look at local registered apprenticeships, especially in electrical work and HVAC. Citing U.S. Department of Labor data, the article says electricians who complete registered apprenticeships earn 49% more over their careers than peers who do not.
Another point is wage progression. Electrician apprentice pay is tiered by verified experience, and the gap between the highest and lowest tier is close to 2x. The article also notes that an AI-integrated technician track launched in 2025 and continued expanding in 2026.
Route B: Certifications for faster switching
Certifications are presented as a direct way to improve screening odds and pay. Certified data center technicians are listed with a median salary of $82,000, versus $68,000 for those without formal certification, a difference of $14,000. The gap is described as widening at more senior levels.
At large operators such as Equinix, Digital Realty and QTS, applicants with certifications are said to be three times more likely to pass initial screening, with job-search timelines shortened to four to 12 weeks.
The article separates certification choices by career direction:
- DCCA from Schneider Electric for entry-level facilities knowledge focused on power and cooling.
- CompTIA Server+ for entry-level IT, centered on server hardware and systems.
- CDCP, or Certified Data Centre Professional, developed by EPI as a foundational credential. The article notes that Chinese-language classes are available in Taiwan.
- BICSI DCDC, or Data Centre Design Consultant, for the design track. It is said to be relatively uncommon in Taiwan. Total costs for coursework, standards materials and the exam are listed at about $2,500 to $3,000, with a discount for BICSI members paying a $199 annual fee.
- Uptime Institute ATD, or Accredited Tier Designer, for people aiming at design and consulting rather than entry-level operations.
The article repeatedly warns that this is where people often lose time: facilities and IT are not interchangeable lanes. Facilities work covers power, cooling and electromechanical systems. IT covers servers, networks and systems. Picking the wrong certification can cost a year.
Route C: Move over from adjacent skills
The fastest route, according to the article, is often a lateral move from a neighboring trade or technical background. It specifically mentions electricians, air conditioning technicians, fire protection or security system technicians, military communications personnel and workers from IT operations.
About half of data center technicians, along with a meaningful share of facilities engineers, are said to come from vocational schools, military service or community college certificate programs rather than computer science degrees.
At the end of 2025, the U.S. had 439,000 unfilled data center-related jobs, the article says. The deepest shortages were in MEP, or mechanical, electrical and plumbing, commissioning, and power and cooling infrastructure. Another estimate cited in the piece puts permanent jobs at as many as 650,000 in 2026, with 340,000 of them left unfilled because of skill gaps.
Those shortages map to established industrial and infrastructure skills, not to entirely new categories of work.
Taiwan: real openings, smaller scale
The Taiwan side looks different mainly because the domestic build-out is much smaller than in the U.S., though telecom and hardware companies are still hiring around the same themes.
The article lists several projects now expanding. Chunghwa Telecom is converting IDC capacity across Taiwan into AIDC, and a subsidiary plans to invest more than NT$3 billion to build a fourth AIDC facility in Central Taiwan Science Park. Google is expanding its Changbin cloud region. Microsoft is expanding in Taoyuan. Foxconn and NVIDIA deployed a 27 MW GB300 data center in Taiwan in the first half of 2026.
Hiring channels in Taiwan
Chunghwa Telecom is described as one of the most structured entry points. Its recruitment is split between business and technical tracks, and the listed projects include roles in power and air conditioning operations management as well as telecommunications network planning, design and maintenance. Those align closely with the facilities side and the network side of a data center.
The technical track starting salary has been raised to NT$50,020, while the business track starts at NT$47,995. Training program compensation, including meal and transport allowances, is listed at about NT$55,520. In one recent recruitment round, the company sought 686 people in total, including 399 regular hires and 287 alternates.
Delta Electronics is recruiting more than 1,700 people in Taiwan in 2026 and nearly 8,000 globally, with a stated focus on AI power supply and thermal management. The article frames that as the electromechanical core of data center work.
Google Taiwan also has a dedicated data center hiring page for its Changhua County site, using the same job categories as in the U.S.
Where certifications are offered
For certifications, the article points to CDCP courses delivered in Taiwan by iSpan. The class runs for two days and is available in Taipei, Taoyuan, Taichung and Kaohsiung. Registration is online. Participants receive an email within seven days before the course confirming whether the class will open, and payment is made only after the class is confirmed.
Pricing varies by session. The article says a discounted price for January to June sessions had been NT$56,700 and included one CDCP exam attempt, though applicants are told to confirm the current price directly with customer service.
The scale difference still matters
An estimate commissioned by Taiwan’s Administration for Digital Industries and carried out by the Institute for Information Industry puts annual official demand for AI professionals in Taiwan at about 3,750 people in 2026. That is a nationwide annual figure, not something comparable to the U.S. adding around 9,000 construction jobs a month in data centers.
Still, electricity use by data centers in Taiwan, while currently under 1% of total power consumption, is projected to rise from 60 MW in 2025 to 450 MW in 2029. The article gives that growth an annual average rate of 66% and says hiring should follow that curve, even if the starting base is small.
The biggest risk is not replacement by AI but the expiration date on construction work
The article makes this its main warning. Goldman Sachs estimates that peak build-out activity could create about 4.7 million temporary construction jobs, while permanent operations jobs after facilities open would total only about 697,000. That is close to a 7-to-1 ratio.
In practical terms, the roles that currently look most attractive on pay and entry difficulty, especially construction-side jobs such as electricians, steel work and commissioning, are project jobs. Once a site is finished, workers either move to the next project or try to transition into operations roles at the facility they helped build.
Goldman also cautioned that if construction is excluded, the employment picture in industries where AI is actually being embedded, including marketing, graphic design, customer service, document processing and software, looks weaker than the aggregate numbers suggest. Data center construction is helping support headline employment now, but that support has a time limit.
Additional resources and the article’s FAQ points
The source article also lists follow-up resources, including Goldman Sachs Research pages on data center capacity and AI demand, the monthly AI Adoption Tracker, Uptime Institute education pages, iSpan’s CDCP course information in Taiwan, Google’s data center recruiting pages and Chinese-language guidance on BICSI DCDC.
Its FAQ section adds several practical points. A related degree is not required to enter the field; about half of data center technicians come from vocational schools, military service or community college certificate programs. People with no experience are steered toward apprenticeships or entry technician jobs first, then certifications later. Those with backgrounds in electrical work, HVAC or IT operations can use a targeted certification to improve initial screening odds. Facilities and IT remain separate tracks, with the former centered on power, cooling and generators, and the latter on servers, networks and systems. Taiwan does have openings, but on a much smaller scale than the U.S. And in the near term, the immediate risk to these roles is less about AI automation than about project completion reducing construction demand.
The article also notes that Goldman’s reported AI adoption rate of 21.5% applies specifically to companies with 20 to 49 employees, not to all companies. The latest figure for all firms is 20.6%, while the rate for firms with more than 250 employees is 36.1%.