YouTube2Text

How AI uses our drinking water - BBC World Service — Transcript

by BBC World Service · 986 words · 101 segments · language en · Watch on YouTube

Full transcript

  1. 0:001/15 of a teaspoon. That's how  much water the average single
  2. 0:05interaction with ChatGPT uses,  according to Sam Altman, the
  3. 0:09boss of OpenAI. So if you type,  can you help me solve this
  4. 0:13maths problem? That's a drop.  Or can I put lime instead of
  5. 0:18lemon in this recipe? That's a  drop. Or why is the sky blue?
  6. 0:22Help me write this email. Help  me improve my website code.
  7. 0:27Mr. Altman claims there are  1 billion messages sent to
  8. 0:30ChatGPT every day, and ChatGPT  is just one AI bot. Chuck in
  9. 0:36Gemini, DeepSeek, Claude and  others. It's clear that the AI
  10. 0:40revolution is a thirsty one.  Striking though it is, some
  11. 0:46experts are more than a little  sceptical of Sam Altman's
  12. 0:49estimate on water usage.
  13. 0:52At this point, there was just  not enough information for me
  14. 0:55to agree with or trust the  number. Their number was perhaps
  15. 1:01referring to some tiny models.  We're considering a medium
  16. 1:06sized large language model  that's the size of GPT-3.
  17. 1:10Basically, if you write an  email or ask some questions, if
  18. 1:14you have 10 to 50 queries,  you're going to be consuming
  19. 1:17roughly 500ml of water.
  20. 1:19This calculation includes  water used in cooling and
  21. 1:22electricity generation. The  BBC asked OpenAI for more
  22. 1:26details about Sam Altman's  estimate, but the company
  23. 1:29declined. Either way, it's clear  AI uses a lot of water. But
  24. 1:34why? Every time you send a  prompt to an AI, it has to run
  25. 1:39complex calculations to understand  and respond. This work is
  26. 1:44done by the most powerful and  specialised computer chips in
  27. 1:47the world, housed inside enormous  data centres. Even before
  28. 1:52users can send prompts, the  training process for the models
  29. 1:55uses the chips to carry out  intense work. And all that extra
  30. 2:01power means the hardware can  overheat and become damaged if
  31. 2:04not cooled properly. Most  data centres use air cooling
  32. 2:08systems, which was fine until I came along.
  33. 2:12But now, because these data  centres and the infrastructure
  34. 2:16that's going in is so much more  energy intensive, there are
  35. 2:21liquid cooling approaches that  are now being implemented.
  36. 2:24For liquid cooling the water  must be clean to prevent
  37. 2:28bacteria growing or clogs and  corrosion in the system, which
  38. 2:31means using mostly drinking  water. Here's how the most
  39. 2:34common liquid cooling process  works. It begins by piping
  40. 2:38coolant over the processing  chips within the servers. This
  41. 2:42cooling liquid absorbs the  heat and takes it away from the
  42. 2:45electrics to a heat exchange  unit. Water is used to reduce
  43. 2:49the temperature of the coolant. The coolant then recirculates back to cool the servers.
  44. 2:54Meanwhile, the now hot water is piped
  45. 2:56to cooling towers, where a combination of fans and water vapour dissipate the heat,
  46. 3:01cooling the water. Some of the water
  47. 3:04evaporates in that process, while the rest is recirculated through the cooling process
  48. 3:09several times before being
  49. 3:10discharged back into the nearby water source. Overall, up to 80% of the water evaporates.
  50. 3:17What it means is that this  type of water is gone, and that
  51. 3:20we are extracting water  from a water circuit that is
  52. 3:24necessary for irrigation, for  human consumption and hygiene.
  53. 3:30Communities around the world  concerned about data centres
  54. 3:33putting stress on water sources  and electricity grids are
  55. 3:36pushing back. Protests have  been held in Spain, India,
  56. 3:40Chile, Uruguay and parts of  the US. And it's not just the
  57. 3:44operations within the data centre that need water. Generating the electricity to run
  58. 3:49them requires a lot of water too, because power
  59. 3:51plants like coal, gas and nuclear heat water to create steam,
  60. 3:56which drives a turbine. The International Energy Agency
  61. 4:00has said electricity demand for AI optimised data centres is
  62. 4:04expected to increase by 400% by 2030 to 300 terawatt hours.
  63. 4:11That's roughly the electricity consumption of the whole of the
  64. 4:14UK for a year. And aside from electricity,
  65. 4:18water is also needed when manufacturing the semiconductor chips used to run AI.
  66. 4:24So water is both used directly  and indirectly in the whole
  67. 4:27supply and creation chain of  AI technologies. It is used for
  68. 4:32the refination of the critical  raw materials that are needed
  69. 4:37to create the hardware of AI.
  70. 4:39Getting accurate figures on how  much water it takes to build
  71. 4:42AI systems and run them is  difficult. Google, Meta and
  72. 4:47Microsoft release annual  figures showing that their data
  73. 4:50centres use billions of litres  of water every year from
  74. 4:53local sources, but none of  them indicate how much of it is
  75. 4:57due to AI. Most tech giants  recognise the impact it's
  76. 5:00having. Many, including Google,  Microsoft and Meta, have
  77. 5:04pledged to be water neutral by 2030.
  78. 5:10S5 We hope that can happen,
  79. 5:12there is a long way to go to get to those kind of numbers. Part of what
  80. 5:18we hope to see is, across the industry,
  81. 5:21a range of innovations that allow us to maybe minimise the use of water.
  82. 5:25Companies are trialing, for  example, ways to cool data
  83. 5:28centres without evaporating any  water at all. And to use the
  84. 5:32heat that's generated to  warm homes. There are also
  85. 5:35experiments to move data  centres away from communities
  86. 5:38entirely under the sea, to  the Arctic, or even off the
  87. 5:42planet.
  88. 5:43Can we actually put capacity  out in space? It's very, very
  89. 5:48early stage. So what, you know,  we at NTT are looking at is,
  90. 5:52can we launch satellites that  can at least do some more
  91. 5:56backup-oriented or other oriented tasks?
  92. 6:00Though skeptics point to the  many hurdles that need to be
  93. 6:02overcome, there is optimism,  too, about a more sustainable
  94. 6:06future.
  95. 6:07Let's remember that that GenAI  capability is still very,
  96. 6:10very young. It's moved  exponentially fast, but as an
  97. 6:14industry and as a use, it is  still young. Ideally, we can
  98. 6:18learn together as as a society  and as a world society, how
  99. 6:23do we minimise against the use  of water and energy? Because
  100. 6:26this is all, you know, a world resource when we
  101. 6:29talk about water.

About this transcript

This page contains the full transcript of How AI uses our drinking water - BBC World Service by BBC World Service, generated from the public captions YouTube serves with the video. The transcript has 986 words across 101 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.