Cold open · the premise
Every question you ask a chatbot drinks a full 500 mL bottle of water. Picture a person doing that at your desk, one bottle per question, and you would have words for them.
That's the premise. It kills, because it feels true.
Now let's audit the premise. Bring your receipts and a poncho.
Act I
DRINK 💧
First, the fact-check the comedian never asked for: per question, today's frontier chatbots sip about a quarter of a milliliter — not a bottle. The bit is off by roughly 2,000×. Comedy has never owed anyone error bars.
*"Measured" with an asterisk you could park a truck under: Google counts direct cooling water only — not the water evaporated making the electricity, which the UC Riverside team found is most of it. The bottle is smaller than the bit says; it is also not the whole bottle.
How far does one 500 mL bottle actually go?
Prompts answered per bottle of water, by accounting method. Log scale — every gridline is a 10× jump.
prompts per bottle · log scale · hover a row for the fine print
napkin math: ~2.5B ChatGPT prompts/day × 0.32 mL ≈ 18.5 bottles/sec 🍼
So no — your one question is not a bottle. But two and a half billion questions a day is ~800,000 liters of direct water. Daily. The comedian was wrong about the sip and right about the drinking problem.
Act II
THINK 🧠
Why does thinking cost anything at all? Because of a 1945 floor plan. A von Neumann machine keeps memory over here and compute over there — so a transformer answers you by marching essentially every weight in the model across a bus, for every single token. Your chatbot is not thinking. It's commuting.
And that 37-second pause in the bit? On stage it's a man chugging water. In silicon it's DRAM round-trips. Either way: the pause is the expensive part.
Now the flip. An energy-based model doesn't march weights around to compute your answer. It learns an energy landscape where good answers sit in the valleys — then asks physics to roll downhill. On a GPU you have to simulate that rolling (expensive). On a thermodynamic sampling unit, the rolling is literal: the circuit's own thermal noise does the sampling, natively, in analog.
Thermal jiggle isn't a bug to cool away — it's the compute. GPUs spend real megawatts generating fake randomness; a TSU is made of real randomness and just… uses it.
Act III
BURN 🔥
Zoom out from your one little prompt and the numbers stop being cute. Data centres are on track to double their electricity use by 2030 — to roughly Japan's entire annual consumption — with AI-focused facilities tripling.
Data centre electricity, worldwide
TWh per year · IEA (2026). Single series; the 2030 point is the IEA base-case projection.
hover for values · projection striped, because the future hasn't been measured yet
So here's the actual headline: a wave of post-von-Neumann hardware wants to delete the commute instead of widening the road. The full tour, honesty gauges included:
Claimed energy-efficiency multipliers vs a GPU baseline
Log scale. One of these numbers is measured on real silicon; one is a prophecy with a foundry deal. The badges say which.
× more energy-efficient than GPU baseline · log scale · hover a row for provenance
Finale
The bit, re-run on better physics
Same heckler, same question, two machines. On the left, today. On the right, the claimed post-transformer future — an energy-based model relaxing on a thermodynamic chip.
Encore
The receipts 🧾
Every number above, its epistemic status, and where it came from. This table is the accessible twin of every chart on this page.
| Number | What it is | Status | Source |
|---|---|---|---|
| 0.24 Wh · 0.26 mL · 0.03 gCO₂e | Median Gemini text prompt: energy, direct cooling water, emissions ("less TV than 9 seconds") | self-reported measurement; excludes training, networks & indirect water | Google (Aug 2025), DCD |
| 0.34 Wh · 0.32 mL | Average ChatGPT query — "about one-fifteenth of a teaspoon" | vendor blog post, no methodology published | Altman (Jun 2025) |
| 500 mL per 10–50 answers | GPT-3-era water per medium-length responses, lifecycle incl. water behind electricity, varies by site & season | academic estimate | UC Riverside, "Making AI Less Thirsty" |
| 700,000 L | Freshwater evaporated training GPT-3, on-site estimate | academic estimate | same paper |
| ~1,900 / bottle | Gemini prompts per 500 mL bottle (500 ÷ 0.26), direct water only | napkin, from Google's figure | derivation shown |
| ~2.5 B prompts/day | ChatGPT daily volume, mid-2025 — powering the ticker (~18.5 bottles/sec, ~800 kL/day, ~850 MWh/day ≈ 29,000 US homes) | vendor figure + napkin chain | OpenAI via press |
| ~90% / ~10% | Share of AI runtime energy spent on data movement vs actual compute | measured, industry research | IBM Research |
| ~640 pJ vs 3.7 pJ | 32-bit DRAM fetch vs FP32 multiply (45 nm) — the ~170× commute tax | measured, canonical | Horowitz, ISSCC 2014 |
| 485 → ~950 TWh | Global data centre electricity 2025 → 2030 (≈3% of global demand; ≈ Japan's annual use; AI-focused DCs tripling; +17% in 2025) | IEA base case | IEA (2026) |
| 4.4% → 6.7–12% | US electricity share to data centres, 2023 → 2028 range | national-lab analysis | LBNL (Dec 2024) |
| 73× · 47× | IBM NorthPole vs comparison GPUs on 3B-param LLM inference: energy efficiency · speed | measured on real silicon | IBM Research |
| 1.15 B neurons · 15 TOPS/W · ~100× | Intel Hala Point (Loihi 2) neuromorphic system; advantage on sparse, event-driven workloads | vendor-measured | Intel (2024) |
| 10,000× | Extropic's claimed energy advantage for denoising thermodynamic models (EBMs) on TSUs vs GPU sampling | claim — simulation-backed, pre-production silicon | Extropic, VKTR |
| $75 M | Non-binding letter of intent, US Dept. of Commerce CHIPS R&D office → onshore TSU fabrication (X0 → Z1 → Z1.5) | announced Jul 29, 2026 | Extropic |
| ~35 seeds · ~1⁄300th | Chia-seed cost of one answer: one seed ≈ 1.2 mg ≈ 5.8 cal ≈ 6.8 mWh; 0.34 Wh ÷ 6.8 mWh ≈ 35; claimed EBM 0.034 mWh ≈ seed⁄300 | napkin, USDA calorie data | derivation shown |
Full sources
- Milan Patel — "I'm Really Good at Flirting," Don't Tell Comedy — the water-bottle bit this page adapts.
- Google Cloud — Measuring the environmental impact of AI inference · MIT Technology Review's read on it
- Li, Yang, Islam, Ren — Making AI Less "Thirsty" (UC Riverside)
- IEA — Key Questions on Energy and AI (2026) · LBNL — US Data Center Energy Usage Report
- IBM Research — the von Neumann bottleneck · Horowitz — Computing's Energy Problem (ISSCC 2014)
- Intel — Hala Point neuromorphic system
- Extropic — Thermodynamic computing chips in America ($75M DoC LOI) · VKTR on the 10,000× claim