Here comes the AI capex shocker, Goldman Sachs says
By Gammatek ISPL, Industrial Safety & Compliance Analyst, Gammatek ISPL
Last updated: September 2026 | 11 min read
Author block: Gammatek ISPL advises manufacturing, industrial, and infrastructure clients on safety compliance systems at Gammatek ISPL. This analysis draws on Gammatek's work with industrial safety documentation, plus data from Goldman Sachs Research, ConstructConnect, the Associated Builders and Contractors, and the American Society of Safety Professionals (ASSP), current as of August 2026.
Why You Should Care
Goldman Sachs just told investors that AI capital spending is on track to blow past even the bank's own recent estimates — potentially exceeding $1 trillion globally in 2026, well above the roughly $800 billion consensus figure most analysts had penciled in. That's the headline number getting attention. What's getting far less coverage is what that money actually builds: physical data centers, at a pace and density the construction industry has never handled before, with a safety workforce that isn't growing anywhere near as fast. If you work in industrial construction, EHS, or safety compliance — this capex story is really a workforce and risk story, and it's happening right now.
The Numbers Behind the Headline
Goldman Sachs researchers now expect AI-related capital expenditure to climb from roughly 1.8% of U.S. GDP in 2026 to 2.5% in 2027 and 2.8% by 2028, a trajectory the bank compares to the peak investment periods of past general-purpose technology buildouts like electrification and the railroads. On the corporate side, the pattern is already visible: major hyperscalers raised their full-year capex guidance mid-2026 from a range of $180–190 billion up to $195–205 billion, with executives signaling a further "significant" jump in 2027. Even companies outside traditional tech are getting pulled in — one automaker committed roughly three times its historical capital spending level for 2026, largely tied to AI infrastructure and compute-heavy production goals.
Goldman's own TMT research team has been blunt about how long this lasts: their view is that AI supply and demand won't find balance until the first half of 2028, meaning capex, revenue growth, and supply chain constraints are set to keep climbing together for at least another two years — not a one-quarter spike.
The Part the Capex Story Skips: Who's Building All of This
Every dollar of AI capex eventually needs a physical building, and that build-out is happening at a pace the construction industry is visibly straining to match. U.S. data center construction starts hit a record $25.2 billion in a single month in January 2026. By March, monthly data center spending reached $9.5 billion, pushing year-to-date totals to $46.5 billion — more than six times the volume recorded at the same point in 2025. Full-year 2026 U.S. data center construction spend is now estimated at $85.3 billion, with hyperscalers committing over $400 billion in combined capital expenditure toward new facilities and expansions.
That's the money. Here's the strain it's putting on safety systems specifically: the Associated Builders and Contractors estimates the broader construction industry needs 350,000 net new workers in 2026 alone, rising to 450,000 in 2027 — while the National Center for Construction Education and Research projects 41% of the current workforce will retire by 2031. Experienced safety directors are being pulled toward the highest-paying, highest-profile data center jobs, leaving other sites — including standard industrial and manufacturing builds — thinner on safety leadership than they were two years ago. One industry account described a hyperscale data center site running 500 workers at peak with a single safety manager overseeing documentation for 700 subcontractors, manually collecting and forwarding paperwork for owner review.
A Simple Comparison: Then vs. Now
Typical Industrial Build (2023) | Hyperscale Data Center Build (2026) | |
Workforce scale | Dozens to low hundreds on-site | 500+ workers at peak |
Subcontractor count | Handful to a dozen | Up to 700 |
Safety documentation | Often manual, periodic review | Continuous, often manually collected and forwarded |
Compliance expectation | Regulatory minimum in most cases | Client-mandated proactive standard, driven by hyperscaler requirements |
Safety staffing ratio | Typically better-resourced per worker | Frequently one safety manager per 500+ workers |
The Associated Society of Safety Professionals' 2025 Construction Safety Challenges report found that 38% of firms still don't operate with a proactive safety and compliance approach — a gap that's becoming harder to sustain as hyperscaler clients increasingly set the operating standard other project types are now expected to match, even when it isn't formally required of them yet.
Why This Isn't Just a Data Center Story
This matters beyond the data center sector itself. The same limited pool of qualified safety directors, EHS managers, and compliance documentation specialists serves manufacturing, chemical, and pharma plants too. When the AI infrastructure boom pulls experienced safety leadership toward hyperscale projects — which pay more and carry higher visibility — general industrial and manufacturing operations are left competing for a shrinking pool of talent, often forced to rely more heavily on manual processes or under-resourced safety teams.
From Gammatek's own work advising plant operators, this is exactly the pattern we're seeing accelerate through 2026: clients who previously managed compliance documentation with modest in-house resources are now asking how to formalize and digitize those processes, precisely because they can no longer assume they'll be able to hire or retain a dedicated safety manager the way they could two or three years ago. The AI capex boom isn't just a Wall Street story — it's quietly reshaping who's available to keep other industrial sites safe and compliant.
What This Means If You Run an Industrial or Manufacturing Site
A few practical implications worth acting on now, not after a labor crunch or audit gap becomes a real incident:
Expect safety staffing costs to keep rising. As hyperscale projects out-compete for experienced safety directors, wages and retention costs for that talent will likely climb across all sectors, not just construction.
Manual compliance documentation is becoming a bigger liability, not a smaller one. The data center sector's shift toward continuous, demonstrable safety documentation — "safety performance needs to be demonstrated, not described," as one industry safety lead put it — reflects where client and regulatory expectations are heading generally.
Software-driven compliance tracking is shifting from "nice to have" to "necessary." With fewer dedicated safety personnel available per site, plants that rely on software to track training, incidents, and audit trails will be better positioned than those depending entirely on manual processes and a single safety manager's bandwidth.
This trend has multi-year legs. Goldman's own research suggests AI capex growth continues at least through 2027–2028, meaning the safety staffing squeeze this creates isn't a short-term blip — it's a multi-year planning consideration.
The Bottom Line
The headline number — AI capex potentially topping $1 trillion in 2026 — is the part getting attention. The more consequential number, for anyone running an industrial site, is the safety staffing gap that spending is creating underneath it. As experienced safety talent concentrates on the highest-paying data center projects, the rest of the industrial world is being asked to do more compliance work with less dedicated staff — which is exactly the environment where documentation gaps, missed audits, and preventable incidents start to creep in.
If your plant's compliance process still depends heavily on one person's manual tracking, this is a reasonable moment to ask whether that's still a safe bet for the next two to three years — not because of any single incident, but because of where the talent and attention in this industry are visibly heading.
[See how Gammatek's compliance platform helps plants maintain audit-ready documentation without depending on a single safety manager's bandwidth →https://www.gammateksolutions.com/post/the-rise-of-physical-ai-can-robots-save-us-manufacturing
Comments