The Toxic Blueprint: How North Carolina Learned—and Forgot—the Semiconductor Industry’s Burden
🌐 The AI Inflection Point (TAIIP) provides an outside/in view of how tech infrastructure, policy, and global power dynamics are shaping the AI transformation — and what it means for your work and life. Curated monthly insights across 4 forces reshaping AI: Global power struggles, innovation vs. protection, talent migration, and infrastructure.
When I arrived at Intel’s New Mexico fab startup in 1983, Joseph “Chip” Hughes was thousands of miles away in North Carolina, standing in courtrooms and community halls, tracing the impacts on the people living beside the semiconductor fabs.
We were part of the same story without knowing it. Inside and outside. Two vantage points on the same industry. We followed the same advocates, pored over the same reports, worried about the same chemicals, and asked strikingly similar questions.
We just never met.
Not until this August—forty-two years later—when we connected through a discussion space seeded by Michael Spencer, an AI-focused community that had grown into a wider network thinking together about how technology touches real lives. Within minutes, it was obvious: our paths had been running parallel the whole time.
When Chip sent me this essay, I recognized the landscape immediately—the regulatory fragmentation, the missing data, the evasions around long-term exposure, the relentless pressure to keep production moving. But I wasn’t in the rooms where those gaps became human consequences.
Chip was.
Part One: The North Carolina Question (1981)
In 1981, North Carolina stood at a threshold the state didn’t fully recognize it was crossing.1 The General Assembly had just appropriated $24.4 million to establish the Microelectronics Center of North Carolina.2 It was sold as the future—the “new industrial revolution,” in the breathless language of boosters.3 But a smaller group of people, working in legal aid offices and health centers across the state, were asking a different question: What are we actually inviting in?
I was among them. The semiconductor industry was already the fourth-largest industrial employer in North Carolina, moving fast toward the top tier.4 Yet almost nobody was asking what those fabrication plants would do to the workers inside them, or the groundwater beneath them.
The state had no approved sites for hazardous waste disposal. IBM’s Research Triangle Park facility had already begun leaking toxic waste into the water table—a fact that came out quietly, after the damage was already done.5
And nobody had a real answer to the obvious question: What happens when dozens of semiconductor plants start operating across the state, each generating chemical byproducts that current regulations barely contemplated?
My work with the pesticide project at Farmworker Legal Services gave me a foothold in occupational health. But the semiconductor industry wasn’t agriculture. It required a different kind of translation—turning chemical safety data into understandable risk, connecting worker testimony to environmental science, building a case that economic development and human health weren’t separate ledgers to balance. They were the same ledger.
That’s when California activists became indispensable.
Part Two: Learning From the West
The Silicon Valley Toxics Coalition had been formed just a year before, in 1982, when Ted Smith and others pulled together the scattered threads that groups like SCCOSH and PHASE had been working with for years.6
They had a clarity we needed: the high-tech hazard chain—the idea that the same system that endangered workers in the cleanroom endangered the community’s water supply, that occupational disease and environmental contamination weren’t separate problems but symptoms of a single industrial logic.
SVTC had already pushed Santa Clara County to pass one of the nation’s first Right-to-Know ordinances in 1983, forcing semiconductor firms to disclose the chemicals they used and stored.7 They had mapped contamination, documented worker illness, connected the dots between corporate secrecy and public risk. They were doing what environmental justice work should do: making the invisible visible, putting a name and address on abstract hazard.
What impressed me most wasn’t their victories alone—though forcing IBM and Fairchild Semiconductor to acknowledge contamination was no small thing—but their method.
They didn’t argue that the industry couldn’t come to California.8 That would have failed. Instead, they argued that if it came, it would come with conditions, with accountability, with the understanding that “innovation” that leaves behind poisoned aquifers and damaged workers isn’t actually progress.
They provided a framework we adapted here. By the mid-1980s, Silicon Valley had accumulated 29 federal Superfund sites—the densest concentration in the country.9 The epidemiology would later link glycol ether exposure to spontaneous abortion in semiconductor workers.10 When NCOSH and others held hearings on the microelectronics industry in North Carolina, we pointed to this evidence and said plainly: this is what’s coming if we don’t build precaution into the industry from day one.
Part Three: What We Actually Know (If We Look)
The documents told a story that industry spokespeople worked hard to obscure. Take George Herbert, chairman of the Microelectronics Center Board. In testimony to the Joint Appropriations Committee in May 1981, he deployed a clever rhetorical move: acknowledge that chemicals are used, but insist they pose “no unusual hazards to workers nor to the environment when compared to other major manufacturing sectors.”11
It was technically evasive rather than false. Semiconductors didn’t use more chemicals by volume than refineries or steel mills. But the nature of those chemicals and the intensity of exposure were different. Herbert could cite the semiconductor industry’s 1978 occupational injury rate—6.4 cases per 100 workers, lower than the all-industry average of 9.4. Good news, right?
Except that the devil lived in the categories. The California Department of Health Services, tracking physician-reported occupational illnesses that same year, found something Herbert’s comparison obscured: the semiconductor industry had a chemical burn rate of 34.1 percent of all reported illnesses, compared to 12.9 percent across all manufacturing.12 More than two and a half times higher.
The overall illness rate in California data was four times higher than all employment sectors, two and a half times higher than the manufacturing average.
Those weren’t “carefully extracted statistics,” as Herbert claimed about health advocates’ warnings. They were what mandatory state reporting systems documented when you actually paid attention.13
The problem was that federal OSHA data showed one picture, California’s more rigorous physician-reporting system showed another, and state industrial recruiters preferred the former.
This is where SVTC’s work became crucial—not just locally but epistemologically. They insisted on triangulation: worker testimony, environmental testing, health department data, internal corporate documents, independent epidemiology.
They wouldn’t accept single data points. And they understood that industry would always prefer fragmented information—researchers in one place knowing about groundwater contamination, workers in another knowing about miscarriages, union organizers knowing about leaking tanks, public health advocates knowing about chemical exposure—to an integrated picture that forced uncomfortable questions.
The industry’s risks ripple outward. Here’s the chain we observed from day one:
What those investigations in California revealed was the chemical reality of chip production. The process wasn’t just electronics—it was relentless chemistry. Each stage brief, each chemical selected for specific reactivity, each requiring precise ventilation and handling.
But here’s what wasn’t on the table yet: reproductive hazards, which emerged only later as researchers linked glycol ether exposure to miscarriages and subfertility in semiconductor workers at multiple companies.14 That knowledge came from workers who tracked their own pregnancies, from occupational health advocates who listened, from epidemiologists willing to ask hard questions about why women in certain fab areas had elevated rates of pregnancy loss.
Part Four: The California Blueprint Applied to North Carolina
When we brought California’s lessons back to North Carolina, we faced resistance that was simultaneously principled and self-interested. Economic development mattered. Jobs mattered. But so did the fact that North Carolina had already had its first toxic waste groundwater contamination from a semiconductor operation—IBM’s Research Triangle Park facility.15 And so did the future plants: Raychem in Fuquay-Varina, Data General in Apex, Hewlett-Packard in Wake Forest. The state didn’t have approved hazardous waste disposal sites. The water table was vulnerable. The infrastructure wasn’t ready.
Our argument wasn’t that the industry should be kept out. It was that the state should learn from what had already happened in California—not repeat it.
None of it was adopted. The “heat of legislative battle,” as the contemporary account noted, made touchy topics too risky. Better to fast-track the money, express confidence in industry self-regulation, and assume that if problems arose, they could be addressed later.
They couldn’t. Or rather, they were addressed only after people got sick. That’s the pattern SVTC had documented in California: causation first known through worker illness, then investigated, then reluctantly acknowledged by regulators, then addressed through litigation and legislation. The state could have shortened that timeline. It didn’t.
The semiconductor industry did come to North Carolina, grew rapidly, created jobs and wealth. It also generated contamination that took decades to understand and address—groundwater plumes discovered years later, worker illnesses that took litigation to establish causation, environmental damage that became apparent only when someone looked carefully.
California, meanwhile, became the proving ground for what could have happened here if precaution had mattered. SVTC’s work in the 1980s and 1990s forced the industry to acknowledge reproductive hazards, to phase out certain chemicals, to adopt better waste management practices.16
The tragedy isn’t that the industry came to North Carolina. It’s that it came without the hard-won lessons from California, without the institutional memory of what goes wrong when innovation outpaces precaution, without the organized pressure from workers and communities insisting that jobs and health were both necessary.
Part Five: Why the Cycle Repeats
I’ve spent forty years watching this cycle repeat because the incentives that drive it are structural, not accidental. A community facing economic decline gets offered jobs—real jobs, with decent wages, that don’t require moving away.
The economic development officials, the politicians, the community leaders all have strong incentives to say yes quickly and ask questions carefully. The companies offering the jobs have strong incentives to move fast, to sidestep precautionary requirements, to present themselves as responsible actors who will solve any problems that emerge.
The workers and communities who will bear the health and environmental costs have less immediate incentive to organize resistance, especially if the jobs are coming to a region that’s been hollowed out.
This isn’t a moral failing on their part—it’s a rational response to the choice they’re presented. Jobs now versus health risks later is a gamble many people feel they can’t afford not to take.
What SVTC understood was that this calculus could be reframed, not erased. They didn’t argue that people shouldn’t have jobs. They argued that jobs shouldn’t require poisoning the water or sickening the workers. That seemed like a simple enough proposition. It still does.
The most damaging part of the semiconductor industry’s history isn’t the contamination itself, which is eventually remediable. It’s the forty years of regulatory lag, of workers getting sick before causation was established, of communities being told their water was fine when it wasn’t.
That lag happens because the regulatory framework fragments responsibility across agencies, allows dispersed contamination that doesn’t trigger any single violation, and treats each new industrial expansion as a separate problem rather than as a systemic transformation of regional water and energy systems.17
Part Six: The Same Blueprint, Scaled Up
Now we’re watching the cycle repeat. Data centers and AI chip production are expanding with a speed and scale that exceeds even the semiconductor boom of the 1980s. The rhetoric is familiar: transformative technology, economic opportunity, jobs that will anchor communities.
The infrastructure questions are often framed as technical problems to be solved, not ethical ones to be confronted.
But the material reality is becoming visible to anyone looking. Data centers consume enormous quantities of water—estimates range from 3 to 5 million gallons per facility per day—for cooling servers that generate unprecedented heat loads.18
AI chip production uses the same chemical processes described earlier, but at exponentially greater scale as manufacturers race to satisfy demand. Intel’s foundry plans, TSMC’s expansions, Samsung’s new fabs—each requires millions of gallons of ultrapure water daily, dozens of toxic chemicals managed in concentrated form, waste streams that existing infrastructure wasn’t designed to absorb.
In places like Arizona, where Intel has proposed massive new fabrication capacity, aquifer depletion is already critical.19 In Taiwan, where TSMC dominates chip production, water stress during drought periods has threatened manufacturing.
In North Carolina—yes, North Carolina again—new data center proposals are facing resistance in rural communities that have watched their groundwater become unreliable and seen semiconductor-era contamination persist for forty years.20
The mistakes being repeated are precise repetitions. The economic argument crowds out the resource argument. Data centers will “bring jobs” and “tax revenue”—statements that are true but incomplete, because they exclude the question of what gets consumed and contaminated in the process.
The technological argument suggests that problems can be engineered away—that recycled water systems, better scrubbing, improved waste management will solve the problem, when the real problem is scale itself.
You can make a single chip fab’s environmental impact manageable through engineering. You can’t make fifty new fabs scattered across a water-stressed region manageable through engineering alone.
And the same regulatory frameworks remain inadequate. EPA water discharge limits, state air quality standards, federal hazardous waste regulations—all of these exist. But they were written for a different scale of industrial activity and fragment responsibility in ways that allow cumulative harm to escape accountability.
In 2025 we don’t fully understand the long-term health effects of some chemicals being used in the newest generation of semiconductor and data center equipment. We’re not systematically tracking worker health in the new fab capacity coming online. We’re not conducting integrated environmental monitoring that would catch dispersed contamination before it becomes catastrophic.
Part Seven: What Precaution Would Mean Now
If we took seriously the lessons of California’s experience and North Carolina’s missed opportunity, here’s what would be different:
These core measures would make development slower, more expensive, and more accountable — but far safer 21 22 23.
However, it requires alternative sources of pressure:
It requires researchers willing to do epidemiology that might find problems.
It requires lawyers willing to take on litigation that corporations will fight viciously.
It requires activists willing to do unglamorous work documenting contamination and health outcomes year after year.
It requires the media willing to treat occupational disease as a legitimate story.
It requires communities and advocates to recognize that the supposed choice between economic development and environmental protection is a false binary —one constructed to shut down precaution before it can be raised.
Part Eight: Will We Use the Archive?
This is the window where precaution could matter—where we could insist on the lessons that cost California forty years and North Carolina similar decades to learn:
Where we could require comprehensive environmental assessment before permission is granted.
Where we could build worker health monitoring into the industry from the start rather than fighting for it retroactively.
Where we could establish that tech companies have responsibility for the full impact of what they produce, not just the revenue.
The stakes are higher now than they were in 1981. The CHIPS and Science Act of 2022 authorized $52 billion in federal subsidies for domestic semiconductor manufacturing, explicitly framed as a matter of national security and economic competitiveness.24
The scale of proposed expansion dwarfs anything North Carolina faced in the early 1980s. Intel’s Ohio facilities, TSMC’s Arizona expansion, Samsung’s Texas operations—each represents billions in public investment and exponentially greater resource demands than the first wave of semiconductor development.
Yet the same pattern is repeating. Communities are being asked to absorb environmental and health costs in exchange for jobs and tax revenue, with minimal precautionary frameworks in place.
The difference this time is that some communities are organized. CHIPS Communities United—a coalition that emerged in response to the CHIPS Act—has been raising precisely the issues that SVTC raised forty years ago: water depletion, chemical exposure, worker health, environmental justice, the adequacy of existing regulatory frameworks.25
This isn’t abstract philosophy. It’s practical risk management.
The precautionary principle says: given what we know, slow down, assess capacity, build safeguards, require transparency, establish monitoring systems before the damage is done.26
Will it happen? The incentives suggest not:
Data center and chip fab development moves faster than regulatory frameworks can adapt.
Companies will innovate around restrictions rather than accept delays.
Communities will compete for jobs, undercutting each other’s environmental requirements.
Federal industrial policy prioritizes speed and capacity over precaution—the CHIPS Act’s timelines and performance requirements push projects forward rapidly, with environmental and health considerations treated as compliance hurdles rather than fundamental design constraints.
But the archive exists now. SVTC’s work is documented. The California contamination is mapped. The epidemiology on reproductive hazards is published. The testimony from workers who got sick is on record.
The question is whether that evidence will be used, or whether it will be filed away like the recommendations from North Carolina in 1981, waiting for the next community to experience what we could have prevented.
The work I and many other community, academic and labor allies did in North Carolina in the early 1980s mattered not because it stopped the semiconductor industry from coming—it didn’t—but because it created a record of what should have happened instead.
That record was built with guidance from California activists who understood something fundamental: that the job of environmental and occupational health advocacy isn’t to prevent all industrial development, but to insist that development be honest about its costs, accountable for its consequences, and structured to protect the people it depends on.
That principle doesn’t change as industries change. It’s as relevant to data centers in 2025 as it was to semiconductors in 1981. The question is whether enough people will remember it, and whether enough people will insist on it before the water is gone and the workers are sick and forty years have passed and we’re explaining to the next generation how we saw this coming and let it happen anyway.
In July 2025, just weeks before Chip and I connected, the White House issued executive orders directing the EPA to “develop or modify regulations” under the Clean Air Act, Clean Water Act, and CERCLA—the very laws that were supposed to prevent what Chip documents in his essay.
The stated goal? Accelerate AI data center and semiconductor facility construction by reducing “regulatory barriers.” Projects committing $500 million or requiring over 100 megawatts qualify for this expedited treatment.
This essay reminds us that decisions echo for decades and become cautionary tales when they could have been design principles. Here are three small but meaningful ways to practice that:
Ask local officials the kinds of questions no permit application volunteers:
What will this facility consume? What will it release? Who absorbs the risk?Strengthen the civic and community institutions that translate technical complexity into public understanding.
Even modest support keeps the outside view alive.Use the influence you already have–inside your company, your university, your field–to elevate precaution as a norm rather than an afterthought.
Small internal shifts accumulate into cultural change.
The blueprint exists. So does the archive of what happens when we ignore it. The AI inflection point is not just a technological juncture; it’s a test of how seriously we take our responsibility to each other.
And the outcome, as always, will be shaped by the people who choose to look from both the inside and the outside at once.
Endnotes
Joseph T. Hughes, Jr., “Healthy Future for North Carolina: Microelectronics 1981,” N.C. Insight, 1981.
Senate Bill 443, Chapter 704 of the 1981 Session Laws.
Microelectronics Center of North Carolina promotional materials, c. 1981.
Hughes, “Healthy Future for North Carolina,” 1981.
Ibid. IBM’s groundwater contamination was ongoing for three years before public disclosure.
B. Dooley, “EHPnet: Silicon Valley Toxics Coalition,” Environmental Health Perspectives 110, no. 5 (2002): A273. Available at: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1240840/
SVTC Timeline & History (archived at icrt.co).
SVTC’s strategy was accountability, not prevention—distinguishing their work from simple anti-industry positions.
Locus Technologies, “Superfund to Supermarket: Brownfields Redevelopment in Silicon Valley,” 2016.
M. B. Schenker et al., “Adverse Reproductive Outcomes Associated with Occupational Exposure Among Semiconductor Manufacturers,” American Journal of Industrial Medicine 28, no. 5 (1995): 639-650; A. Correa et al., “Ethylene Glycol Ethers and Risks of Spontaneous Abortion and Subfertility,” American Journal of Epidemiology 143, no. 7 (1996): 707-717.
George Herbert, “Statement Regarding Requested Budget Appropriations for Microelectronics Center of North Carolina” (testimony before the Joint Appropriations Committee, May 28, 1981).
California Department of Health Services, “Tabulation of Reports of Illnesses in California Reported by Physicians,” 1976.
Ibid. California’s mandatory physician-reporting system was the only such system in the country at that time.
Schenker et al., 1995; Correa et al., 1996.
Hughes, “Healthy Future for North Carolina,” 1981.
SVTC investigations documented in Poison PCs (2001), Toxic TVs (2002), and the Solar Scorecard (2009).
The regulatory framework’s fragmentation allows dispersed contamination that doesn’t trigger single-agency violations—a structural problem SVTC identified in the 1980s that persists today.
International Energy Agency, “Electricity 2024: Analysis and forecast to 2026,” 2024. Data centers consumed approximately 560 million cubic meters of water in 2023, projected to reach 1.2 billion cubic meters by 2030. See also: S&P Global, “Beneath the surface: Water stress in data centers,” September 18, 2025. AI chip production requires the same chemical processes as conventional semiconductors but at exponentially greater volumes. https://www.spglobal.com/sustainable1/en/insights/special-editorial/beneath-the-surface-water-stress-in-data-centers
In Arizona, the state receives just 13.6 inches of rainfall annually, making it the fourth driest state nationwide. CNBC, “Intel, TSMC are building water-dependent chip plants in Arizona,” June 4, 2021. https://www.cnbc.com/2021/06/04/why-intel-tsmc-are-building-water-dependent-chip-plants-in-arizona.html In Taiwan, TSMC’s water consumption increased 71% between 2015 and 2019. Fortune, “Taiwan’s drought is exposing just how much water chipmakers like TSMC use,” June 12, 2021. https://fortune.com/2021/06/12/chip-shortage-taiwan-drought-tsmc-water-usage/
Intel and TSMC semiconductor expansion faces critical water stress in multiple regions. Josh Lepawsky (Memorial University geography professor) analysis found that six of seven planned US semiconductor facilities are in watersheds at high or extremely high risk of water stress. Area Development, “Semiconductors’ Fragile Relationship With Water May Be Tested,” November 27, 2024. https://www.areadevelopment.com/advanced-manufacturing/q3-2024/semiconductors-fragile-relationship-with-water-may-be-tested.shtml
Worker health monitoring from industry inception could accelerate identification of new hazards—a principle that emerged from SVTC’s work.
Schenker et al., 1995; Correa et al., 1996.
Precautionary scaling assesses whether institutional and environmental capacity exists to manage technology at proposed scale—acknowledging that individual facilities might be manageable but cumulative expansion might exceed regional carrying capacity.
CHIPS and Science Act of 2022, Pub. L. No. 117-167, 136 Stat. 1366 (2022). The Act authorized $52.7 billion for semiconductor manufacturing, research, and workforce development, with $39 billion dedicated to manufacturing incentives and $11 billion to research and development.
CHIPS Communities United formed in response to the CHIPS Act to coordinate advocacy across regions facing new semiconductor development. The coalition has raised concerns about water use, hazardous waste, worker safety, and environmental justice impacts, particularly in communities already facing environmental burdens. CCU has pushed for stronger community engagement requirements, environmental impact assessments, and worker protections in CHIPS Act implementation.
The precautionary principle, as articulated in the 1998 Wingspread Statement, holds that “when an activity raises threats of harm to human health or the environment, precautionary measures should be taken even if some cause and effect relationships are not fully established scientifically.” This principle directly contradicts the regulatory approach that dominated North Carolina’s semiconductor recruitment in the 1980s, which assumed safety until harm was proven—a standard that guaranteed workers and communities would bear the burden of exposure during the years or decades required to establish causation. CCU is applying the precautionary principle that should have guided North Carolina’s decisions in 1981—the idea that when an activity raises threats of harm to human health or the environment, precautionary measures should be taken even if some cause-and-effect relationships are not fully established scientifically.








Amazing article, Dee and Chip! And how cool that you connected on your shared background in semiconductor manufacturing so many years later thanks to the AI Vanguard Society pilot :)
Thank you for the restack, @Ted Smith!