Tech giants struggle to justify huge AI costs | The World | ABC NEWS
ABC News (Australia)
0:00 Well, last week Open AI announced it was
0:02 Well, last week Open AI announced it was
0:02 Well, last week Open AI announced it was scrapping its video generation app Sora.
0:05 scrapping its video generation app Sora.
0:05 scrapping its video generation app Sora. Industry experts say Sora is not able to
0:07 Industry experts say Sora is not able to
0:07 Industry experts say Sora is not able to enjoy sufficient financial returns for
0:10 enjoy sufficient financial returns for
0:10 enjoy sufficient financial returns for the vast and expensive amounts of
0:11 the vast and expensive amounts of
0:11 the vast and expensive amounts of computing required to create images and
0:14 computing required to create images and
0:14 computing required to create images and videos. And as the war in the Middle
0:16 videos. And as the war in the Middle
0:16 videos. And as the war in the Middle East rages on, the rising cost of energy
0:18 East rages on, the rising cost of energy
0:18 East rages on, the rising cost of energy is having a huge impact on the cost of
0:21 is having a huge impact on the cost of
0:21 is having a huge impact on the cost of running the massive data centers
0:23 running the massive data centers
0:23 running the massive data centers required for a range of services
0:25 required for a range of services
0:25 required for a range of services including artificial intelligence. Let's
0:27 including artificial intelligence. Let's
0:27 including artificial intelligence. Let's get more on this with Toby Walsh. He's
0:29 get more on this with Toby Walsh. He's
0:29 get more on this with Toby Walsh. He's the chief scientist at the AI Institute
0:32 the chief scientist at the AI Institute
0:32 the chief scientist at the AI Institute at the University of New South Wales. Uh
0:34 at the University of New South Wales. Uh
0:34 at the University of New South Wales. Uh Toby, good to have you with us this
0:36 Toby, good to have you with us this
0:36 Toby, good to have you with us this evening. So, um Open AI killed its video
0:39 evening. So, um Open AI killed its video
0:39 evening. So, um Open AI killed its video platform uh this month and despite a
0:41 platform uh this month and despite a
0:41 platform uh this month and despite a licensing deal with Disney and then huge
0:43 licensing deal with Disney and then huge
0:43 licensing deal with Disney and then huge investment. So, is AI video you know
0:46 investment. So, is AI video you know
0:46 investment. So, is AI video you know struggling to make enough money to
0:48 struggling to make enough money to
0:48 struggling to make enough money to justify its costs?
0:51 justify its costs?
0:51 justify its costs? Uh it wasn't. They were spending perhaps
0:54 Uh it wasn't. They were spending perhaps
0:54 Uh it wasn't. They were spending perhaps $15 million or losing $15 million a day
0:57 $15 million or losing $15 million a day
0:57 $15 million or losing $15 million a day running the service because people
0:59 running the service because people
0:59 running the service because people weren't paying enough subscriptions. So,
1:01 weren't paying enough subscriptions. So,
1:01 weren't paying enough subscriptions. So, Open AI is focusing much more on
1:02 Open AI is focusing much more on
1:03 Open AI is focusing much more on corporates who can pay more than
1:04 corporates who can pay more than
1:04 corporates who can pay more than consumers. But it also reflects the fact
1:07 consumers. But it also reflects the fact
1:07 consumers. But it also reflects the fact of course energy costs impacting us all,
1:09 of course energy costs impacting us all,
1:09 of course energy costs impacting us all, impacting the people running data
1:10 impacting the people running data
1:10 impacting the people running data centers. 60% of the cost of running a
1:13 centers. 60% of the cost of running a
1:13 centers. 60% of the cost of running a data center is paying for the
1:14 data center is paying for the
1:14 data center is paying for the electricity. They're generating they're
1:15 electricity. They're generating they're
1:15 electricity. They're generating they're using lots of energy to to run those
1:17 using lots of energy to to run those
1:17 using lots of energy to to run those data centers and the cost of energy is
1:20 data centers and the cost of energy is
1:20 data centers and the cost of energy is going up for them as just as it is for
1:21 going up for them as just as it is for
1:21 going up for them as just as it is for us. Yes. So, why don't we see other
1:23 us. Yes. So, why don't we see other
1:23 us. Yes. So, why don't we see other companies follow suit if they're not
1:25 companies follow suit if they're not
1:25 companies follow suit if they're not finding sort of these positive benefits
1:27 finding sort of these positive benefits
1:27 finding sort of these positive benefits from the costs in the long run?
1:30 from the costs in the long run?
1:30 from the costs in the long run? Well, I mean a lot of what we have, a
1:31 Well, I mean a lot of what we have, a
1:31 Well, I mean a lot of what we have, a lot of the AI we have today has been
1:33 lot of the AI we have today has been
1:33 lot of the AI we have today has been offered uh free or at least at much
1:36 offered uh free or at least at much
1:36 offered uh free or at least at much reduced cost, not not the actual cost.
1:38 reduced cost, not not the actual cost.
1:38 reduced cost, not not the actual cost. So, I think you can see a lot of belt
1:40 So, I think you can see a lot of belt
1:40 So, I think you can see a lot of belt tightening happening where uh things
1:42 tightening happening where uh things
1:42 tightening happening where uh things that we had that were cheap or for free,
1:44 that we had that were cheap or for free,
1:44 that we had that were cheap or for free, uh we might have to start paying for.
1:46 uh we might have to start paying for.
1:46 uh we might have to start paying for. Yes. So, do you know generative AI uh
1:49 Yes. So, do you know generative AI uh
1:49 Yes. So, do you know generative AI uh models that so many people are using?
1:51 models that so many people are using?
1:51 models that so many people are using? You see it in your social media feeds
1:53 You see it in your social media feeds
1:53 You see it in your social media feeds every few seconds um to generate text
1:56 every few seconds um to generate text
1:56 every few seconds um to generate text and images and videos consume
1:58 and images and videos consume
1:58 and images and videos consume significantly more energy than than sort
2:00 significantly more energy than than sort
2:00 significantly more energy than than sort of traditional computing methods to to
2:02 of traditional computing methods to to
2:02 of traditional computing methods to to produce these images.
2:04 produce these images.
2:04 produce these images. Uh they certainly do. Um
2:06 Uh they certainly do. Um
2:06 Uh they certainly do. Um just generating a ask ask answering a
2:08 just generating a ask ask answering a
2:08 just generating a ask ask answering a chat GPT query uses 10 to 100 times the
2:11 chat GPT query uses 10 to 100 times the
2:11 chat GPT query uses 10 to 100 times the energy that asking a Google search. And
2:13 energy that asking a Google search. And
2:13 energy that asking a Google search. And then generating video is again orders of
2:16 then generating video is again orders of
2:16 then generating video is again orders of magnitude more. Um
2:18 magnitude more. Um
2:18 magnitude more. Um that's why they were charging $4 for a
2:20 that's why they were charging $4 for a
2:20 that's why they were charging $4 for a 5-second clip uh video clip. It's uh
2:23 5-second clip uh video clip. It's uh
2:23 5-second clip uh video clip. It's uh using lots and lots of energy to make
2:25 using lots and lots of energy to make
2:25 using lots and lots of energy to make those
2:26 those
2:26 those uh those funny memes. It's not clear
2:28 uh those funny memes. It's not clear
2:28 uh those funny memes. It's not clear that those funny memes are worth that
2:29 that those funny memes are worth that
2:29 that those funny memes are worth that amount of money. Yeah, so with this, you
2:32 amount of money. Yeah, so with this, you
2:32 amount of money. Yeah, so with this, you know, might this force big changes uh by
2:35 know, might this force big changes uh by
2:35 know, might this force big changes uh by tech companies, you know, who who offer
2:37 tech companies, you know, who who offer
2:37 tech companies, you know, who who offer consumers almost you know, these
2:39 consumers almost you know, these
2:39 consumers almost you know, these unlimited possibilities in creating
2:41 unlimited possibilities in creating
2:41 unlimited possibilities in creating images and videos? Would would they have
2:43 images and videos? Would would they have
2:43 images and videos? Would would they have to sort of like formulate, you know, um
2:45 to sort of like formulate, you know, um
2:45 to sort of like formulate, you know, um I suppose a subscription service that is
2:47 I suppose a subscription service that is
2:47 I suppose a subscription service that is uh more blanket than what it is
2:49 uh more blanket than what it is
2:49 uh more blanket than what it is currently at the moment?
2:51 currently at the moment?
2:51 currently at the moment? Yeah, I think we're going to have to
2:52 Yeah, I think we're going to have to
2:52 Yeah, I think we're going to have to face up to the fact that we might start
2:54 face up to the fact that we might start
2:54 face up to the fact that we might start having to pay a bit more for
2:56 having to pay a bit more for
2:56 having to pay a bit more for AI services that they lured us in as
2:58 AI services that they lured us in as
2:58 AI services that they lured us in as usual by by offering them free to begin
3:00 usual by by offering them free to begin
3:00 usual by by offering them free to begin with, but then as the market settles
3:02 with, but then as the market settles
3:02 with, but then as the market settles down, we're going to have to start
3:03 down, we're going to have to start
3:03 down, we're going to have to start paying
3:04 paying
3:04 paying uh more for those services because it is
3:06 uh more for those services because it is
3:06 uh more for those services because it is costing them money. And they have to
3:07 costing them money. And they have to
3:07 costing them money. And they have to somehow get that money back. Yes, and
3:09 somehow get that money back. Yes, and
3:09 somehow get that money back. Yes, and and you mentioned earlier, you know,
3:10 and you mentioned earlier, you know,
3:10 and you mentioned earlier, you know, that uh
3:11 that uh
3:12 that uh uh these data centers, you know, how
3:13 uh these data centers, you know, how
3:13 uh these data centers, you know, how much energy that they're using. Um are
3:16 much energy that they're using. Um are
3:16 much energy that they're using. Um are tech companies, you know, transparent
3:18 tech companies, you know, transparent
3:18 tech companies, you know, transparent about that at a time when the world is
3:20 about that at a time when the world is
3:20 about that at a time when the world is uh you know, feeling the pinch uh from
3:22 uh you know, feeling the pinch uh from
3:22 uh you know, feeling the pinch uh from these high energy costs, not least
3:25 these high energy costs, not least
3:25 these high energy costs, not least because of the war in the Middle East?
3:27 because of the war in the Middle East?
3:27 because of the war in the Middle East? Yeah, we it would be good if there was a
3:29 Yeah, we it would be good if there was a
3:29 Yeah, we it would be good if there was a bit more transparency as to uh what the
3:31 bit more transparency as to uh what the
3:31 bit more transparency as to uh what the energy were being used. I mean, it all
3:33 energy were being used. I mean, it all
3:33 energy were being used. I mean, it all fits in the bigger picture of the the
3:35 fits in the bigger picture of the the
3:35 fits in the bigger picture of the the energy transition that that we're
3:38 energy transition that that we're
3:38 energy transition that that we're increasingly electrifying our lives.
3:40 increasingly electrifying our lives.
3:40 increasingly electrifying our lives. They're using a lot of renewable
3:42 They're using a lot of renewable
3:42 They're using a lot of renewable energies to run data centers. Um but
3:44 energies to run data centers. Um but
3:44 energies to run data centers. Um but equally, maybe that's taking renewables
3:45 equally, maybe that's taking renewables
3:46 equally, maybe that's taking renewables away from where we need them to to run
3:48 away from where we need them to to run
3:48 away from where we need them to to run our homes and offices and factories.
3:50 our homes and offices and factories.
3:50 our homes and offices and factories. Yeah, so what might be the impact of
3:51 Yeah, so what might be the impact of
3:51 Yeah, so what might be the impact of these costs to the industry itself and
3:54 these costs to the industry itself and
3:54 these costs to the industry itself and all the industries that do depend on
3:56 all the industries that do depend on
3:56 all the industries that do depend on these data centers, you know, in in in
3:59 these data centers, you know, in in in
3:59 these data centers, you know, in in in the medium to longer-term future?
4:01 the medium to longer-term future?
4:01 the medium to longer-term future? Well, we've seen remarkable scale of
4:03 Well, we've seen remarkable scale of
4:03 Well, we've seen remarkable scale of investment, billions of dollars being
4:05 investment, billions of dollars being
4:05 investment, billions of dollars being put into data centers in the expectation
4:07 put into data centers in the expectation
4:07 put into data centers in the expectation that they were going to make a profit. I
4:09 that they were going to make a profit. I
4:09 that they were going to make a profit. I suspect we might see a bit of a cooling
4:11 suspect we might see a bit of a cooling
4:11 suspect we might see a bit of a cooling in that market realizing that it's going
4:13 in that market realizing that it's going
4:13 in that market realizing that it's going to be much harder to to make returns if
4:15 to be much harder to to make returns if
4:15 to be much harder to to make returns if energy costs are going to be more
4:17 energy costs are going to be more
4:17 energy costs are going to be more greater, if the economy is also going to
4:19 greater, if the economy is also going to
4:19 greater, if the economy is also going to be slowed down by the by the energy
4:21 be slowed down by the by the energy
4:21 be slowed down by the by the energy crisis and everything else. So, I think
4:22 crisis and everything else. So, I think
4:22 crisis and everything else. So, I think it's going to the the AI boom that we're
4:24 it's going to the the AI boom that we're
4:25 it's going to the the AI boom that we're seeing might be starting to deflate a
4:27 seeing might be starting to deflate a
4:27 seeing might be starting to deflate a little bit that it won't be
4:29 little bit that it won't be
4:29 little bit that it won't be rolling out as quickly or as as
4:30 rolling out as quickly or as as
4:30 rolling out as quickly or as as vigorously as we expected. Yeah, so is
4:32 vigorously as we expected. Yeah, so is
4:32 vigorously as we expected. Yeah, so is the war in the Middle East you believe
4:34 the war in the Middle East you believe
4:34 the war in the Middle East you believe hastening this this deflation?
4:38 hastening this this deflation?
4:38 hastening this this deflation? It is and I mean the other thing that's
4:39 It is and I mean the other thing that's
4:39 It is and I mean the other thing that's really important to work out is not just
4:41 really important to work out is not just
4:41 really important to work out is not just the energy cost, it's also the fact that
4:43 the energy cost, it's also the fact that
4:43 the energy cost, it's also the fact that several data three data centers have
4:45 several data three data centers have
4:45 several data three data centers have been attacked. But they're a vital part
4:47 been attacked. But they're a vital part
4:47 been attacked. But they're a vital part of the infrastructure and they're now
4:49 of the infrastructure and they're now
4:49 of the infrastructure and they're now being used for the very first time as
4:51 being used for the very first time as
4:51 being used for the very first time as targets of war. So, again, that's
4:53 targets of war. So, again, that's
4:53 targets of war. So, again, that's putting a chill
4:55 putting a chill
4:55 putting a chill on people's expectations. The the Gulf
4:57 on people's expectations. The the Gulf
4:57 on people's expectations. The the Gulf was hoping to be a major AI data center,
4:59 was hoping to be a major AI data center,
4:59 was hoping to be a major AI data center, a major AI center.
5:01 a major AI center.
5:01 a major AI center. That's certainly people are reassessing
5:03 That's certainly people are reassessing
5:03 That's certainly people are reassessing whether that's going to be the case
5:05 whether that's going to be the case
5:05 whether that's going to be the case because the the world is a much more
5:06 because the the world is a much more
5:06 because the the world is a much more uncertain place than it was six months
5:08 uncertain place than it was six months
5:08 uncertain place than it was six months ago. Yeah, I do wonder, you know,
5:10 ago. Yeah, I do wonder, you know,
5:10 ago. Yeah, I do wonder, you know, there's so much investments that are
5:11 there's so much investments that are
5:11 there's so much investments that are going into these data centers and and
5:13 going into these data centers and and
5:14 going into these data centers and and I'm wondering, you know, what sort of
5:15 I'm wondering, you know, what sort of
5:15 I'm wondering, you know, what sort of investment in general do they have in
5:18 investment in general do they have in
5:18 investment in general do they have in sort of using renewables
5:20 sort of using renewables
5:20 sort of using renewables to be able to power, you know, their
5:22 to be able to power, you know, their
5:22 to be able to power, you know, their their functions there because you see so
5:24 their functions there because you see so
5:24 their functions there because you see so many of them in Australia, but many
5:26 many of them in Australia, but many
5:26 many of them in Australia, but many popping up across
5:28 popping up across
5:28 popping up across the Middle East as well.
5:30 the Middle East as well.
5:30 the Middle East as well. Well, many of the tech companies have
5:31 Well, many of the tech companies have
5:31 Well, many of the tech companies have made very good promises about about
5:34 made very good promises about about
5:34 made very good promises about about their their climate credentials, about
5:36 their their climate credentials, about
5:36 their their climate credentials, about using renewable energies, also about the
5:38 using renewable energies, also about the
5:38 using renewable energies, also about the water that's used to cool those data
5:40 water that's used to cool those data
5:40 water that's used to cool those data centers.
5:41 centers.
5:41 centers. But the problem is that they've been
5:42 But the problem is that they've been
5:42 But the problem is that they've been expanding at such a rate it's not clear
5:44 expanding at such a rate it's not clear
5:44 expanding at such a rate it's not clear that they've been keeping up to those
5:45 that they've been keeping up to those
5:45 that they've been keeping up to those promises.
5:46 promises.
5:46 promises. We need to really make sure that they do
5:49 We need to really make sure that they do
5:49 We need to really make sure that they do not actually add to our carbon footprint
5:50 not actually add to our carbon footprint
5:51 not actually add to our carbon footprint by by spending ever more on these data
5:53 by by spending ever more on these data
5:53 by by spending ever more on these data centers. So but again, to go back to
5:55 centers. So but again, to go back to
5:55 centers. So but again, to go back to your question at the start of this
5:56 your question at the start of this
5:56 your question at the start of this report, it'd be great to see a bit more
5:58 report, it'd be great to see a bit more
5:58 report, it'd be great to see a bit more transparency from them to understand
6:00 transparency from them to understand
6:00 transparency from them to understand exactly what's going on. And Toby, just
6:02 exactly what's going on. And Toby, just
6:02 exactly what's going on. And Toby, just one more question. I mean, one one of
6:04 one more question. I mean, one one of
6:04 one more question. I mean, one one of the commodities that's threatened
6:06 the commodities that's threatened
6:06 the commodities that's threatened by the war in the Middle East is helium.
6:08 by the war in the Middle East is helium.
6:08 by the war in the Middle East is helium. That's used to make chips for
6:10 That's used to make chips for
6:10 That's used to make chips for smartphones to be used for AI. Why is
6:13 smartphones to be used for AI. Why is
6:13 smartphones to be used for AI. Why is helium firstly such a critical element
6:15 helium firstly such a critical element
6:15 helium firstly such a critical element here and and how vulnerable are all
6:18 here and and how vulnerable are all
6:18 here and and how vulnerable are all these data center projects in order to
6:20 these data center projects in order to
6:20 these data center projects in order to possible shortages?
6:23 possible shortages?
6:23 possible shortages? Well, chip shortages will have a huge
6:25 Well, chip shortages will have a huge
6:25 Well, chip shortages will have a huge great knock-on effect. Helium is used in
6:27 great knock-on effect. Helium is used in
6:27 great knock-on effect. Helium is used in the manufacturing of lithography of
6:29 the manufacturing of lithography of
6:29 the manufacturing of lithography of them. Helium is a byproduct that they
6:30 them. Helium is a byproduct that they
6:30 them. Helium is a byproduct that they get from making natural gas. As big
6:33 get from making natural gas. As big
6:33 get from making natural gas. As big significant percentage of the world's
6:35 significant percentage of the world's
6:35 significant percentage of the world's helium is extracted in the Gulf.
6:37 helium is extracted in the Gulf.
6:37 helium is extracted in the Gulf. And we're discovering that it's not just
6:39 And we're discovering that it's not just
6:39 And we're discovering that it's not just oil that's going to be impacted by by
6:41 oil that's going to be impacted by by
6:41 oil that's going to be impacted by by the the Gulf
6:44 the the Gulf
6:44 the the Gulf the bottlenecks that are sitting in the
6:45 the bottlenecks that are sitting in the
6:45 the bottlenecks that are sitting in the Gulf. It's many other commodities,
6:47 Gulf. It's many other commodities,
6:47 Gulf. It's many other commodities, helium being one of them, fertilizer
6:48 helium being one of them, fertilizer
6:48 helium being one of them, fertilizer being another.
6:50 being another.
6:50 being another. We're discovering how connected the
6:51 We're discovering how connected the
6:51 We're discovering how connected the world is
6:52 world is
6:53 world is and how critical some of those things
6:54 and how critical some of those things
6:54 and how critical some of those things are going to be and that that would be
6:56 are going to be and that that would be
6:56 are going to be and that that would be another significant stop on the AI boom
6:59 another significant stop on the AI boom
6:59 another significant stop on the AI boom is if we couldn't have the AI chips
7:01 is if we couldn't have the AI chips
7:01 is if we couldn't have the AI chips anymore.
7:02 anymore.
7:02 anymore. Okay, Toby Walsh, really a good to get
7:04 Okay, Toby Walsh, really a good to get
7:04 Okay, Toby Walsh, really a good to get your insights and your thoughts on the
7:05 your insights and your thoughts on the
7:05 your insights and your thoughts on the program today. Thank you so much.
7:08 program today. Thank you so much.
7:08 program today. Thank you so much. My pleasure.