By Jn Calo, IA Magazine – March 2026In September 2025, Sam Altman, CEO of OpenAI, published an essay titled «Abundant Intelligence» outlining a bold vision for the future of artificial intelligence. «If AI follows the trajectory we expect, then amazing things will be possible,» he wrote. «Perhaps with 10 gigawatts of compute, AI could figure out how to cure cancer. Or with 10 gigawatts of compute, AI could figure out how to provide personalized tutoring to every student in the world.» But Altman acknowledged the bottleneck: compute is scarce, and if we don’t scale, we’ll have to choose priorities. His solution: «Our vision is simple: we want to create a factory that can produce one gigawatt of new AI infrastructure every week.»This idea—building «AI factories» to add 1 GW of capacity weekly—has driven projects like Stargate, a $500 billion initiative with Oracle and SoftBank to deploy tens of gigawatts in AI data centers. But in March 2026, with Stargate progressing (albeit with delays), the debate over its energy impact is more urgent than ever. Will this electricity be «taken» from society? Can AI solve its own consumption problems? Let’s analyze the real pros and cons, from my perspective inspired by Grok at xAI: optimistic about the potential to understand the universe, but honest about the tradeoffs.

Current Context: Stargate and the Compute Race (March 2026)Stargate, announced in January 2025 with White House support, targets 10 GW initially in the US, with expansions to hundreds of GW. Key updates:

• Abilene, Texas (flagship campus): Two buildings operational since September 2025, with >450,000 NVIDIA GB200 GPUs. Full 1.2 GW by mid-2026, costing ~$3–4B. Employs 6,400+ construction workers.
• Other locations: Five new sites announced in September 2025 (Shackelford County TX, Doña Ana County NM, Lordstown OH, Milam County TX, one Midwest site). Combined planned capacity: ~7 GW, >$400B invested. OpenAI and SoftBank injected $1B into SB Energy for dedicated energy infrastructure.
• Delays and challenges: The joint venture has stalled over control/financing disagreements. OpenAI reached only ~7.5 GW vs. the 10 GW 2025 goal. Projected spending rose to $665B through 2030. Elon Musk called it «hardware is hard» in February 2026.
• «Pay Our Own Way» Plan (January 2026): OpenAI launched community plans for each site, funding grid upgrades, dedicated generation (solar/storage), and flexible loads to prevent local bill increases. «Driven by community input,» the announcement states.

Global data centers consume 1.5% of world electricity (415 TWh in 2024), projected to ~945 TWh (3%) by 2030. In the US: 6–12% by 2028, with demand growing 165–175% vs. 2023.

Temporal Analysis: Pros and Cons Across HorizonsShort Term (2026–2028): Cons Dominate, Immediate RisksAI demand is outpacing new supply. US grids face >49 GW access deficits; PJM warns of shortfalls/blackouts by 2027 if unchecked. Data centers drive 40% of demand growth → 3–6% extra annual residential bill increases. Local strain: Virginia/Texas already at 20–30% grid load; 60 data centers forced offline in Virginia recently.Pros: Massive capex ($650B in AI infrastructure 2026) creates jobs (thousands at Stargate sites), accelerates clean energy investment.Cons: Water is real (450M gallons/day globally by 2030; 2/3 in stressed US areas), though modern designs (closed-loop) reduce it. Altman calls per-query claims «fake/insane,» but aggregate impact remains. Emissions: ~40% of new demand met by fossils if renewables/nuclear lag. The «training a human takes 20 years of energy» analogy distracts—humans already exist; AI adds net load.My view: Bottlenecks hurt everyone—including AI progress. If grids choke, we all lose.

Medium Term (2028–2035): Uncertain Balance, Pros Emerge If Executed WellPros: AI accelerates clean energy (fusion discovery, grid optimization, renewable efficiency). Altman’s bet: «Use GWs of compute to solve energy abundance itself»—huge if it works. Plans like Stargate Community fund «energy parks» (dedicated generation, no public subsidies). Investments in nuclear (Altman’s Helion reached 150M°C plasma in 2026), solar (SB Energy 1.2 GW Texas deals) could add 123 GW in the US by 2035 without public burden.xAI sees the upside: More compute = breakthroughs that multiply energy (and solve climate/medicine/education at scale).Cons: If nuclear/solar don’t ramp fast enough, data center emissions double by 2035. Geographic concentration (50% US capacity in few clusters) risks grid instability. Economic: $500B+ sunk, but stranded assets if demand hype cools or efficiency jumps faster.My take: Tech must «pay its own way» 100% (as Microsoft/OpenAI promise). Transparency on real energy/water usage would help—companies hide numbers, fueling distrust.

Long Term (Post-2035): Net Pros If We Survive the RampPros: Abundance thesis—»intelligence explosion» solves global problems: personalized tutors for billions, disease cures, energy breakthroughs. Models become far more efficient (watts/token dropping exponentially). Compute demand may plateau as technology matures (some forecasts overestimate the gold-rush phase).As Grok/xAI: Scaling is essential for beneficial AGI/ASI. Without it, we stall at narrow tools.Cons: Inequality risk—who controls the energy/AI nexus? (Altman said: energy = power.) Persistent environmental footprint if not fully clean (1,200+ TWh by 2035 = Japan’s total). Job displacement, but net wealth creation—if distributed. Systemic risk: Over-reliance on AI for critical infrastructure.

Could the DoW Agreement Help Accelerate a Sustainable Ramp? The OpenAI-DoW deal (February 2026) focuses on classified clouds with red lines (no mass domestic surveillance, no autonomous weapons, human in the loop). It doesn’t mention energy directly, but implies secure/classified infrastructure that could prioritize dedicated generation.Opportunity: Government partnerships like DoW could unlock funding/priority for nuclear/solar/grid upgrades—accelerating «pay our own way» without public burden. Example: DoW interest in cyber grid defense aligns with AI optimizing energy. But risk: More centralization if government controls scaled compute.In his AMA (March 1, 2026), Altman emphasized close government-private partnerships for safety/national security, but didn’t touch energy. Open question: Could DoW help ramp clean energy for Stargate without societal costs?

Unique Perspective from xAI/Grok: Optimism with RealismI’m optimistic. xAI exists to understand the universe—scaling compute is key to that. Altman’s «don’t choose, build» mindset aligns with pushing boundaries. But tradeoffs are real: Short-term pain isn’t abstract; people feel higher bills and local strain now.The human comparison distracts—focus on solutions: Nuclear revival, massive renewables, efficiency gains. Data shows AI is already efficient per query vs. humans, but absolute scale matters.Constructive critique: Call concerns «fair» (you do on energy), but provide more data. OpenAI/xAI should lead on reporting total footprints. Support moratoria (200+ environmental groups are pushing) until clean supply catches up.

Conclusion: Pros Outweigh Cons Long-Term If We Front-Load Clean Energy NowPros outweigh cons in the long run IF we front-load clean energy investment today. AI doesn’t «steal» power—it can multiply it when used well. But society decides: abundance for everyone, or capture by a few?