| --> YouTube-Generated Transcript <-- |
| 0:01 | · | Amazon laid off 30,000 people this week |
| 0:03 | · | in a move that had been widely |
| 0:05 | · | telegraphed for months. But the real |
| 0:07 | · | story is not about AI and jobs. The |
| 0:11 | · | story I'm seeing over and over and over |
| 0:13 | · | again in the news media is, hey, AI is |
| 0:16 | · | automating all of these jobs. This is |
| 0:18 | · | why we're seeing this. We're going to |
| 0:20 | · | see more of these cuts. The media seems |
| 0:22 | · | really excited about that story and they |
| 0:24 | · | just want to keep telling it. And I got |
| 0:26 | · | to say it keeps not being in this case. |
| 0:30 | · | There's a very interesting reason why |
| 0:32 | · | it's incorrect. And I think it's |
| 0:34 | · | actually really important for us to |
| 0:35 | · | understand because it gets at a core |
| 0:37 | · | narrative that a lot of people have |
| 0:39 | · | about AI that is wrong. So what actually |
| 0:42 | · | happened here? Number one, you need to |
| 0:44 | · | understand where Amazon's business came |
| 0:46 | · | from. I spent half a decade at Amazon. |
| 0:47 | · | I'm very fluent in this. Amazon makes |
| 0:50 | · | their money on AWS, not on retail. The |
| 0:53 | · | whole store doesn't do anything, right? |
| 0:54 | · | that the margins on the store are |
| 0:56 | · | ridiculously low. They make no money. |
| 0:58 | · | They would not be a profitable company. |
| 0:59 | · | Jeff Bezos would not be wandering around |
| 1:01 | · | in a yacht. Amazon makes their money on |
| 1:03 | · | AWS, which means the entire street, all |
| 1:06 | · | of Wall Street checks AWS's growth |
| 1:09 | · | numbers every single year obsessively to |
| 1:12 | · | tell whether or not Amazon is doing |
| 1:15 | · | well. And the problem is that AWS's |
| 1:18 | · | growth numbers have been declining for |
| 1:21 | · | the last few years. And so they're down |
| 1:23 | · | to 18% growth year-over-year, which is a |
| 1:25 | · | deceleration in their last quarterly |
| 1:27 | · | report. And the street doesn't like it. |
| 1:30 | · | And part of the reason the street |
| 1:31 | · | doesn't like it is that in the meantime, |
| 1:34 | · | AWS's two main rivals, Google Cloud and |
| 1:38 | · | Microsoft Azure, have been catching up |
| 1:40 | · | and they have been doing so with AI. |
| 1:43 | · | Right? If you think about where to go |
| 1:45 | · | for AI, you think about Azure and you |
| 1:47 | · | think about Google Cloud. And I got to |
| 1:49 | · | be honest, AWS is a distant third. They |
| 1:53 | · | just aren't there. And so the challenge |
| 1:55 | · | for AWS is they need to show that they |
| 1:58 | · | are still serious players in the AI era. |
| 2:02 | · | To do that, what do you need? You need a |
| 2:04 | · | specific piece of hardware that Jensen |
| 2:06 | · | Fuang sells that nobody else has. It's |
| 2:08 | · | the Nvidia GPU. I well there's a few |
| 2:10 | · | others that have GPUs but by and large |
| 2:13 | · | it's all Jensen's GPUs and you got to |
| 2:15 | · | buy a bunch of them and they don't come |
| 2:18 | · | cheap like think of this as a computer |
| 2:20 | · | chip that is worth a car and you have to |
| 2:22 | · | buy thousands and thousands and tens and |
| 2:24 | · | thousands of them to get anywhere with |
| 2:26 | · | AI especially at scale especially if |
| 2:28 | · | you're serving corporations that is the |
| 2:31 | · | dilemma that AWS faces now look at it |
| 2:34 | · | from a corporate finance perspective you |
| 2:36 | · | have to do that without damaging your |
| 2:39 | · | margins in AWS because the whole reason |
| 2:42 | · | Amazon is worth anything as a company is |
| 2:45 | · | because of AWS's margin. So you don't |
| 2:48 | · | dare damage AWS's margins to get this |
| 2:51 | · | job done, but you have to buy a whole |
| 2:54 | · | bunch of GPUs. In finance terms, you |
| 2:57 | · | have to add a ton to your capital |
| 3:00 | · | expenditures, your capex. Well, if |
| 3:03 | · | you're going to do that and you want to |
| 3:04 | · | keep your margins consistent, you have |
| 3:06 | · | to cut other places that are in your |
| 3:08 | · | expenses category. You have to look at |
| 3:10 | · | other fixed expenses. And what is the |
| 3:12 | · | biggest fixed expense category that you |
| 3:14 | · | have? It is salaries. It's salaries. |
| 3:17 | · | That is what they're looking at. And so |
| 3:19 | · | when you have that tradeoff, what you |
| 3:21 | · | should really be looking at is Amazon |
| 3:24 | · | did not fire 30,000 people because AI |
| 3:28 | · | automation was already taking their |
| 3:30 | · | jobs. Amazon fired 30,000 people because |
| 3:34 | · | they needed the money today in order to |
| 3:37 | · | buy GPUs to desperately try to secure a |
| 3:41 | · | place in the future of AI cloud. That is |
| 3:44 | · | the actual story. Now, that story does |
| 3:47 | · | not sound as nice for Amazon as a future |
| 3:49 | · | forward story about how we're automating |
| 3:51 | · | with AI. So, the story Amazon's putting |
| 3:54 | · | out is, hey, we're automating with AI, |
| 3:57 | · | so we don't need these jobs anymore. No. |
| 3:59 | · | Especially as someone who worked there, |
| 4:01 | · | the interior workflows at Amazon, and |
| 4:04 | · | anyone will tell you this, this is not |
| 4:05 | · | proprietary. It's all duct tape and |
| 4:07 | · | bailing wire in there. Like, everyone |
| 4:09 | · | does a lot of manual stuff. And that's |
| 4:11 | · | been very intentional as Amazon has |
| 4:13 | · | grown because it helps us keep costs low |
| 4:15 | · | for customers. And so, there is not a |
| 4:17 | · | huge massive automation magical solution |
| 4:21 | · | that they have invented that allows them |
| 4:23 | · | to right now cut 30,000 jobs. It just |
| 4:25 | · | that's not how it works. The people who |
| 4:28 | · | remain are very stressed. They may have |
| 4:30 | · | ambitious projects to eventually bring |
| 4:32 | · | AI into those areas. But if I'm looking |
| 4:34 | · | at it at a very high level, what I see |
| 4:37 | · | is not an investment roadmap for AI |
| 4:40 | · | automation that is already paid off that |
| 4:43 | · | they're trying to show. No, what I see |
| 4:46 | · | is Amazon saying these are areas where |
| 4:48 | · | we can afford to take a risk on less |
| 4:51 | · | talent getting less done. In other |
| 4:52 | · | words, these are areas that we can |
| 4:54 | · | divest a little bit. And that's |
| 4:55 | · | interesting because that makes a whole |
| 4:57 | · | lot of sense when you look at where some |
| 4:59 | · | of these cuts happened. As an example, |
| 5:01 | · | MGM got hit, the Hollywood studio that |
| 5:04 | · | Amazon bought a few years ago. I got to |
| 5:06 | · | say, if you're asking yourself, is this |
| 5:08 | · | an area where we have already invested |
| 5:10 | · | our best efforts to automate AI right |
| 5:13 | · | away? MGM would not be at the top of my |
| 5:15 | · | list. I do not think that is the most |
| 5:17 | · | strategic place in Amazon where Amazon |
| 5:20 | · | would have poured vast resources to |
| 5:22 | · | invest in AI. No. But I do think it |
| 5:25 | · | makes a ton of sense as a place where |
| 5:28 | · | Amazon would say we can invest less for |
| 5:30 | · | a little bit and cut talent for a bit |
| 5:32 | · | because we desperately need to |
| 5:34 | · | reallocate cash over to the GPU side of |
| 5:38 | · | the business. Well, that makes a lot of |
| 5:40 | · | sense, doesn't it? And people aren't |
| 5:42 | · | reporting that. And here's the reason |
| 5:43 | · | why all of this matters. The narrative |
| 5:46 | · | out there is very simple. We are in a |
| 5:49 | · | bubble. Everywhere I look, we are in a |
| 5:50 | · | bubble. AI is a bubble. AI is a bubble. |
| 5:52 | · | But step back, take a breath, don't just |
| 5:55 | · | obsess over the news. If we are in a |
| 5:58 | · | bubble, why does Amazon have a 25% |
| 6:03 | · | essentially overage rate? Right? The the |
| 6:05 | · | the amount of demand they have for for |
| 6:07 | · | GPUs right now vastly exceeds the |
| 6:10 | · | available GPUs. Would that be true if we |
| 6:13 | · | were in a bubble? Demand is a sign that |
| 6:17 | · | we are not in a bubble. surging |
| 6:19 | · | corporate demand that Azure has trouble |
| 6:21 | · | meeting because Azure's expanded their |
| 6:23 | · | data center investments this year that |
| 6:25 | · | Google Cloud has trouble meeting that |
| 6:27 | · | Amazon has tons of trouble meeting is a |
| 6:30 | · | sign that we have built something with |
| 6:33 | · | AI that is valuable enough that |
| 6:36 | · | corporations are lining up like crazy to |
| 6:38 | · | buy it. If you have customers out the |
| 6:41 | · | door and cannot serve enough chips to |
| 6:44 | · | all of them, that is a sign that you are |
| 6:47 | · | not in a bubble. Ipso facto. By |
| 6:50 | · | definition, it is a sign that you are |
| 6:52 | · | not in a bubble. And I don't understand |
| 6:54 | · | why this is so hard for people. I don't |
| 6:57 | · | understand why journalists are so |
| 6:59 | · | interested in pedalling the narrative |
| 7:01 | · | that AI has already automated jobs and |
| 7:04 | · | that somehow we are simultaneously in a |
| 7:05 | · | bubble because both of those things |
| 7:07 | · | cannot be true at once. If we lived in a |
| 7:10 | · | world where AI had magically automated |
| 7:12 | · | away all jobs or whatever they're |
| 7:14 | · | claiming, well then it wouldn't be a |
| 7:16 | · | bubble because people would like |
| 7:18 | · | corporations would have demand for that, |
| 7:19 | · | right? They they would ask for that to |
| 7:21 | · | happen so they could expand their |
| 7:22 | · | footprint. maybe not even to fire people |
| 7:24 | · | but to add more like c talent and |
| 7:26 | · | capability right there'd be demand for |
| 7:27 | · | it if we were you know in a bubble we |
| 7:30 | · | would not have the kind of surging |
| 7:34 | · | interest in AI at every level small |
| 7:36 | · | medium business uh commercial scale |
| 7:38 | · | enterprise scale the consumer like all |
| 7:41 | · | of us as individuals that's not the |
| 7:43 | · | story we see and yet we are being asked |
| 7:46 | · | to believe by the media simultaneously |
| 7:49 | · | that AI is so big and scary it is |
| 7:51 | · | automating jobs and also though that |
| 7:53 | · | somehow we are magically in a bubble and |
| 7:54 | · | it's all fake. It's it's not both |
| 7:56 | · | people. It's not both. And in this case, |
| 7:58 | · | it's not either one of them because we |
| 8:00 | · | don't have the talent yet to build AI |
| 8:03 | · | systems that fully automate roles and we |
| 8:06 | · | like not for a while. Like it's not |
| 8:08 | · | there. And yet corporations love that |
| 8:10 | · | narrative because it makes them future |
| 8:12 | · | focused. It makes Wall Street happy |
| 8:13 | · | because Wall Street doesn't know what AI |
| 8:15 | · | is. And everybody like goes away happy |
| 8:17 | · | with the story. And nobody pays |
| 8:19 | · | attention to the contradictions here. |
| 8:20 | · | The the answer is very simple. They just |
| 8:22 | · | need to buy GPUs because corporations |
| 8:25 | · | need to buy cloud compute. And that is |
| 8:26 | · | what happened. And none of that makes |
| 8:28 | · | getting fired easier. I don't want to |
| 8:30 | · | sort of pretend that being able to |
| 8:31 | · | explain it makes it better. It it sucks |
| 8:33 | · | to be fired. I've been fired before. It |
| 8:35 | · | just it's terrible. And so for those of |
| 8:38 | · | you who have unwillingly left Amazon, |
| 8:40 | · | there are Amazon alumni. We're out here. |
| 8:43 | · | We look out for each other. We're doing |
| 8:44 | · | our best. And you will find a spot. And |
| 8:46 | · | I know that none of this makes it |
| 8:48 | · | better. But I do hope that we actually |
| 8:50 | · | can tell the truth about what is going |
| 8:51 | · | on instead of getting fooled by stories |
| 8:54 | · | that are not even not even coherent, |
| 8:56 | · | right? Like you can't have both an AI |
| 8:59 | · | bubble and AI automating all jobs. It |
| 9:01 | · | does not work. And yet that is the story |
| 9:02 | · | we're being sold. So there you go. |
| 9:04 | · | That's the truth. That's why it matters. |
| 9:06 | · | And that's why we need to pay attention |
| 9:07 | · | when people make layoff announcements |
| 9:08 | · | like that. Good luck. Don't conf. |
