EPISODE 1881

[EPISODE]

[0:00:12] GV: Hello, and welcome to SED News. I'm Gregor Vand.

[0:00:16] SF: And I'm Sean Falconer.

[0:00:17] GV: For anyone that's joining us for the first time on SED News, this is just a slightly different format to the SE Daily Podcast, where we take a spin through tech news, we go through the headlines, we have a main topic in the middle, and then we look at some hacker news highlights and some predictions ahead. As always, nice just to catch up, Sean. We're doing this a little bit earlier than our usual end-of-month recording, so it's coming a bit earlier to listen to this month. As usual, this is easily enough for us to cover. What have you been up to since we last caught up?

[0:00:50] SF: Oh, just heavy travel schedule this fall. Actually, our big annual conference for Confwin is next week in New Orleans, so I'm gearing up for that. I've been very busy getting ready for all the big product announcements and stuff that we're going to be doing there. It should be a fun week, but a very tiring week as well.

[0:01:06] GV: Yeah, nice.

[0:01:07] SF: How are you?

[0:01:08] GV: I've started a new role, but I'm not going to say which company yet. But yeah, it's fun so far. I was also at a - to the big cybersecurity conference here called GovWare. It doesn't sound very tech with gov in the name, but it is. Yeah, just the usual names and faces showing off their products. I wouldn't say I've seen a lot of change to last year. I feel like last year was the first, maybe cybersecurity, where AI was really being pushed. This year, I would say, it's just more of the same, but I guess that's what we're seeing, maybe across the board with a few of these bigger enterprise areas.

We should really talk about our predictions from last month, because yeah, let's go through what we did predict, or try to predict. Then we also missed one. Yeah, your prediction, I think, Sean, was to do Salesforce developing a model of their own. Any word on that?

[0:01:55] SF: No. I knew that was a little bit of an out there reach prediction at the time, with Dreamforce coming up. They made lots of big announcements at Dreamforce, but nothing in terms of launching their own model, at least at this point.

[0:02:06] GV: Yeah. My one was, if you did catch the last episode, there was a Hacker News about hosting a web server on a disposable vape. I thought maybe we'd see something that one up to that on Hacker News. Can't say I did. That just shows me that project was a bit an outlier for the last few months. We implicitly predicted something which was around AMD. We're going to get into it in our main topic, but why is AMD important?

[0:02:35] SF: Well, we were talking about Nvidia, Intel, everything is going on, and the chip world last time, and Intel is getting some attention. They got some funding, essentially, from the government, they get investment from some other big players in the market. Then we raise the question of, like, what's going on with AMD? We bantered about, well, we haven't really heard a lot about AMD. I wonder what's going on with them. We should follow up with that. Then a couple of days later, suddenly there's all this news about AMD investment in the AMD big deal that they signed with OpenAI, is taking up to 10% stake in the company. All this is going on. We really missed our opportunity to have an accurate prediction for the first time, where we probably could have said something about AMD making their move in the world of AI.

[0:03:20] GV: Yeah, for sure. As I say, we're going to get into that. The main topic today is really looking at the - when we say chip makers, we don't just mean fabs, but anyone who is able to ultimately create a chip. That includes Apple, for example. We're going to get into that. We've got a bit more detail on what's been going on there. I think last month we were postulating on a whole bunch of things. We've got a bit more info now on some of these deals and what's being on behind the scenes. That's going to be interesting to get into. We're going to hit the main headlines. These are things that, as it sounds, they hit mainstream news. Whether that could be Wall Street Journal, or Reuters, or Forbes, etc. Yeah. I think a couple of things have come up. Security AI has just been bought for 1.7 billion. What's that about, Sean?

[0:04:07] SF: Yeah. I saw this actually announced today, actually, since you're in the security world, are you familiar with security AI?

[0:04:13] GV: You know what? I'm not. That's an interesting - it could be where I am in the world. Maybe we just haven't seen them over here. Yeah. What are they about, Sean?

[0:04:22] SF: Yeah. I've never used their software. I've seen them around, especially in my prior role, when I worked for Skyflow, since we were in the privacy and security space. We would see them at a lot of the same conferences that we were at. They are focused on doing security, or privacy compliance, discovery of sensitive data across wherever you're storing data, and then being able to actually protect it. They just sold for 1.7 billion dollars, a sizable amount of money to a company. I think it's called VM Software, which are, they have a whole bunch of different parts of software, like data protection, backups, recovery solutions.

I think they're owned by a private equity. Big chuck going into this. Shows, I think, both of them, there's still tremendous amount of momentum in the world of security. Also, I think specifically with what's happening in AI right now.

[0:05:10] GV: Yeah, it is interesting. This is what I find, to be honest, sitting over here in Singapore, there are so many companies that just never cross my radar, unless someone calls them out, because their presence tends to be very US-centric. Yeah. Basically, unless you set up a sales office over here, Singapore tends to be thinking of the sales hub for at least Southeast Asia. It's interesting that just so many actually quite sizable companies just fly under the radar, pun not intended. Moving on to another well-known company on both sides of the Pacific, LangChain has raised 125 million series B at a 1.25 billion valuation. Yeah. What do you think that's signaling?

[0:05:50] SF: I think it's a good thing for people out there building AI companies, or interest in the space, because it is one of the bigger fundraises and valuations that I've seen for a company that's not one of the model companies, or running inference in the cloud, the core infra for building these AI systems, or the base foundation models. They're certainly an abstraction layer on top of that. They also have good open-source roots. I think this is good sign for open sources live and well.

Of course, they have ways that they're monetizing with hosting AI software and agents directly within their own cloud, as well as their observability and eval testing tools and stuff like that. I do come across LangChain a lot, but I think they were one of the early companies that really made a name for themselves in the space when the early large language models were really coming out. They've been able to adapt as well with the market change, because really, LangChain originally was a nice abstraction over top of all these different models, where the model APIs were not necessarily that easy to use at the beginning.

Then the model APIs have gotten significantly better and more and more have been owned by the model companies, and they've been able to adapt into things LangGraph and adding additional functionality and become more sticky player. All while the model companies like OpenAI, Anthropic, well, no so Anthropic, but OpenAI has also made inroads into offering their own competitive products.

[0:07:20] GV: Yeah, and we've got a episode coming up, probably comes out after this episode with Pydantic AI. For those that are familiar with Pydantic, they now have another product library, Pydantic AI. That's very much in the same space as this, but they also did raise money for that. That is a commercial interest, as much as an open-source interest. Look out for that, because certainly, LangChain comes up in that episode a bit, just compare, contrast the two approaches that are being taken to the same problem around basically, agents. Yeah, look out for that.

Moving on, we've got an announcement that a company called Periodic Labs has just come out of nowhere. Who are they? They are one of OpenAI's most respected researchers: Liam Fedus and his former Google Brain colleague, Ekin Dogus Cubuk. I probably completely botched those names, but yeah, this is interesting. 300 million seed round led by A16Z.

[0:08:19] SF: Mono seed round.

[0:08:20] GV: Yeah. I mean, yeah. I believe as these things happen these days, they being offered checks before they'd even formed a company. Investors were getting nervous because they had nothing actually to invest in. That's obviously a good place to be if you're about to found a company and you've already got that lined up. Yeah, what is Periodic Labs, or at least what are these saying they're going to be doing?

[0:08:42] SF: My understanding was they're trying to bring together AI with the physical world. I think one of the taglines is AI gets a wet lab. They're trying to essentially build out a digital scientist in some respects, like Databricks has their AI data scientists is trying to offload certain, albeit co-pilot, or offload certain responsibilities that data scientists might typically have to do on in this world. Snowflake, you have an AI analyst that offloads some of the analyst work. This is an AI scientist that can do leverage AI, essentially, to do some traditional science work in the real world.

[0:09:18] GV: Yeah, very interesting. I mean, I think also interesting that OpenAI, so one of the founders, one of the most respected researchers from OpenAI. Even though I believe, when he was leaving, he said, "I've got the blessing of OpenAI," and it hinted that they would be investing. That didn't materialize in the end, probably for the best on both sides. Perhaps, we're going to be getting into just a bit of OpenAI's financial web later on in relation to the chip makers. Certainly, finance for OpenAI is a bit of an open question right now.

Moving on, say we're recording this on the 22nd, 21st of October. We've just had a major AWS outage, which, yeah, I'm sure for most of us - well, I think again, I was in Asia. Monday it was my Monday. I killed the second half of my Monday. Yeah, what actually happened there, Sean?

[0:10:11] SF: Well, I mean, it makes me to half the Internet, but I actually got some things from my old company, some people that used to be on my team asking for help with the website being down, and Monday being a lot of teams based in India, it was Diwali, so a lot of people were off. He had this combination of factors, and I let them know, well, it's okay. You're in good company, because half the Internet's down. It's not just you.

What I understand there was, so basically, US is one started experiencing some elevated error rates sometime in the middle of the night. They started to investigate. It turned out some of these were related to DynamoDB service, which ends up having a lot of explosive damage across all of AWS, because a lot of things run on DynamoDB. Not just client services, but core services with AWS, you get this cascading effect. If Dynamo is down, it's not just going to be isolated. Dynamo is going to essentially impact all these other services. You ended up with people experiencing really slow response times from websites, things just timing out, and essentially, a ton of services were unavailable for the day.

[0:11:20] GV: Yeah, there's been murmurings that this is effectively related to a brain drain AWS. I saw some, I mean, again, I'm not going to say this is the accurate figure, but someone posted it that they'd laid off 40% of their infra engineers. Now it's AI trying to do self-healing, etc., etc. Do you think that's accurate? Is that something you're aware of? Or was there anyone else talking about this?

[0:11:46] SF: Yeah. I mean, I read Corey Quinn's article that he wrote about this for the register and basically, that was a position that he said that like, well, they've had a ton of layoffs and it was all, of course, good people have also left on their own accord, because they're like, "Oh, this isn't the AWS I remember." There's been a lot of public messages from some of those people who left criticizing some of the decisions that have been made at AWS. I don't know about the idea of them being replaced by AI and whether that was a factor, but I certainly think that if a lot of your experienced engineers are no longer there and then you have some outage like this, not only is it probably going to take longer to fix, but could have been not happened essentially from the get go, because they were experienced enough to recognize either the signs, or knew the risks that were being taken to lead to this situation.

[0:12:35] GV: Yeah. Corey Quinn, Mr. AWS. Definitely look him up if you're looking for some good analysis of all things AWS. Yeah, I saw a, again, I think the timing was real. I don't think this was just a marketing stunt, but David Heinemeier Hansson posted that, so his base camp and obviously big Ruby on Rails, founder of Ruby on Rails and many other things, they've been doing quite a push to move away from AWS. They created their own framework called MRSK. It sounds like Mersk, like the shipping line, so containers. They'd be moving to their entirely self-sufficient bare metal server framework.

That day was when they had called a town hall, where DHH was standing up on a little mini stage and projecting to the team that they were closing their AWS account that day. That was the final, final click off. This was literally just a few hours after the outage had started. Of course, he got exactly what he wanted. Lots of people saying how smart he was on this one. I think it's still going to be interesting just to see, because what are the alternatives to big cloud? Well, just another big cloud, basically. If it's not AWS, you're going to GCP, or maybe Azure. But if you're not there, then you have to take, yeah, a base camp approach of thinking about, if you have those requirements, DHH claimed they were paying 3.5 million a year to AWS. They're not a huge company base camp. They're very profitable, but they're not huge. They're somewhere between 50 to 100 employees. Yeah, 3.5 million is a big chunk of change. They just said, "We're paying that every year. We can invest that amount of money into something that costs way less and is way easier for us to run, basically." Interesting to watch that one.

[0:14:20] SF: Yeah. I don't know. I mean, and maybe that's true for base camp. Maybe they have the existing talent that can run those servers, and that's a good use of that talent and stuff like that. I think for a lot of companies, is that burden of work really what you want to take on? I definitely think this is an incentive to maybe think about how you protect against such issues as in the future of the business. Either not solely dependent on singular region, like US-East-1, or multi-cloud. There's a lot of different ways that you could potentially approach this problem that doesn't require you stacking and racking your own servers. That might be a little bit extreme as a result of this.

The other thing that I thought was interesting than this, and I haven't closely tracked this, but I remember in the early days of AWS, this is over 10 years ago, these outages used to happen all the time. It's very common that, like, "Oh, okay. Well, US West is down today." Then you haven't really seen that, or at least from my perspective, I haven't seen it on a regular basis for quite a long time. Obviously, this is a huge outage. It was like 15 hours. Even if you said that there was no major outages for whatever number of years has been, AWS and the cloud providers companies have become more and more dependent on them. They have a bigger, bigger footprint. There is a tremendous amount of collateral damage when something like this happens, because they end up taking down so many businesses that are ultimately going to be losing probably millions of dollars by not being operational for some period of time.

[0:15:46] GV: Yeah. I think just to wrap this up, I mean, the fact it was roughly one region, that's actually quite confusing, because you can be sitting across outside the world and bits of a product stop working, but not all the bits, because it depends where they are. Are they doing smart region on some things? Have they just said other things have to go through US-East-1? It was a confusing down period, because, yeah, bits of products were working and nobody really knew. Also, in a company Slack, some people are affected, some people are not. It causes a lot of confusion, as opposed to just saying, okay, well, this is just a outed, and this is not like a power cut in a city or something, where you just say, "Well, until it's back on, it's back on."

Moving on, last month, we did touch on the private equity buyout of Electronic Arts. Yeah, there was just a bit more color on that one. This was private equity, Guy Durbin. He's well known for putting together these massive deals in private equity. Turns out that he was the mastermind behind this. We didn't have that color on it before. Also, given it was the Saudi Arabian public investment fund, also since learned that the leader behind that he's very into gaming just generally. We know that that fund is a bit of a pet project, as much as it's called the public investment fund. It very much does follow how the leader of Saudi Arabia just likes to whatever he's interested in. I think this was a perfect deal, where they could get someone who has a lot of money at hand and is actually interested in gaming.

We have seen Battlefield 6 come out. It has got pretty good reviews. I think that's smoothed that one over. We were obviously questioning whether the hype was going to match reality and was affecting anything. Yeah, just a interesting extra color there. Finally, Apple they've actually allowed, so I'm sure people are vaguely aware that Apple had this class UI, which I think dropped maybe a couple months ago. I'm not an iPhone user. I'm not an iPhone user, so this didn't really affect me. I didn't really have to think about it too much, but I'm aware that a lot of people were not very happy about it. Then other people were saying, "No, no. This is the future. They're so smart. They're just getting us ready for AR experiences, where everything will be translucent in front of your face, and so on, so forth."

They have announced that they're allowing a rollback though of this. Have we seen anything like this from Apple before, or is this actually the first time we're seeing them admit that maybe a design decision wasn't well received?

[0:18:21] SF: I mean, I think it's pretty rare from what I can recall. I'm not an iPhone user either, but I have been in the past. But I think it's pretty rare that Apple admits that their design decisions aren't one size fits all, and we know best for the consumer. It seems like they're learning. Maybe the backlash was just so extensive in this case that they had to admit, or maybe it signals some larger cultural change, but they're trying to balance the aesthetics with accessibility, and they're giving users more control, instead of forcing them to have a uniform design, at least in this case. Whether that are transcended on the products or not, we'll have to see. But I haven't seen them do stuff like that previously.

Generally, it's like, hey, remember that port on your laptop that we said was the best thing ever? Well, we changed it for this new version of the laptop, because it's even better, and you got to buy all new adapters and new wires, and we know best. Just go do it.

[0:19:18] GV: Yeah. I must say, I was very happy when they brought back MagSafe to MacBook Pros, at least. It was just a very nice piece of technology. The problem that they, as you call out, they just kept flip-flopping between what the form factor of that was, as well as, oh, we're going to remove it, USB-C, and then put it back again. If they just evolved, rather than completely yo-yo'ed around. Yeah, with the liquid glass, they say that you can now have that you can now have clear or tinted. I'm afraid I don't really know what that means, because, again, I don't use an iPhone. But if you're an iPhone user, that's possible now.

Apparently, one of the only other times they've walked something back in this realm is Safari's address bar moved to the bottom of the iPhone screen at some stage in 2021. People got very angry about that, and they had an option to put it back to the top again. Even though the bar at the bottom makes sense, your thumbs are there, but sure.

[0:20:08] SF: Yeah. I think a clear background is tough, depending on what the background is, right? You have a busy picture, or something like that, and then you have a, I don't know, a clear notification on top of it. How are you going to be able to read the text? I think that's where it starts to become problematic.

[0:20:24] GV: Yeah. I mean, I think to play devil's advocate, I do get frustrated on, as I think most people do on their phone, icons all smashed together in this sort of, oh, I want them to look the same now. If I'm reaching for, say, Authy, Authy, and Pocketcasts, look very similar. Red, white, circular. It's kind of, you have to be quite diligent about organizing your space, or just muscle memory, knowing where that app is, and you know it's not on the top or the bottom, or something. I don't think the glass translucent approach fixes that either. It didn't look like it would fix that. If I saw screenshots of it, it just looked perhaps even more confusing, but I don't know.

Yeah. Just to wrap that one up, they did achieve record iPhone sales this past quarter. It doesn't seem to have actually affected product sales. It's just that they're at least accepting that a lot of users have pushed back on wanting to take on this new UI. Okay.

Moving on to our main topic. This is behind the scenes at the chip makers. What's really going on with a whole bunch of companies? We're going to be looking at Intel. We're going to be looking at AMD. We will go back to looking at Apple in this lens, Nvidia, of course, and then just also looking at where does OpenAI - they're not a chip maker, obviously, but they have a lot of fingers and pies of chip makers now. We just want to understand how are they affecting the landscape. We're just digging in on details of where everyone's at.

Intel, we did touch on them. I think we touched on them, maybe even for two past episodes, just they've been doing deals and having to get back up to financial strength. I did actually say, Oh, they're not too big to fail. They're just too important not to do well. I think you're much more on the money, Sean. They were actually getting to failure point, apparently, behind the scenes.

[0:22:15] SF: Yeah. It's crazy to think about. I mean, they just were so dominant in the '90s. Back then, it was a two-horse race, but they were just almost everything was Intel. Then, of course, the early Apple computers and laptops were Intel. Now, they've gone and you started developing their own chip technology and stuff. It just seems like more and more both moved away from Intel. Then, of course, with GPUs and AI, or they're playing some catch up there. They've been struggling. I think even more than certainly I was aware of.

[0:22:45] GV: Yeah. I mean, there was a comment from an ex-CEO of Intel, Craig Barrett. He believes that even today, even after deals that they've done, the US government is getting 10% and Nvidia investing 5 billion, which obviously, isn't actually that much money when you take into account both of these companies. He still thinks that they're going to need about 40 billion injected. I mean, obviously he's not part of the company anymore, but he probably has a pretty good idea and has a few back-channel information sources as well. He thinks about 40 billion still needs to be injected into Intel to get it back up to where it should be.

[0:23:24] SF: Do you know any insights into where that 40 billion goes to? I'm assuming 40 million, that's a lot of money. They got to be some physical manufacturing type of stuff that they had to be investing into. I mean, there's R&D and stuff like that. But, I mean, 40 million, a lot of money, just of like, hey, we're this far in the hole and we need to inject a lot of money just to, to your point, catch up.

[0:23:46] GV: Yeah. I mean, I think it's basically because they have been bleeding customers. Their cash flows are just down. Yeah. I mean, exactly where would that money go? Well, yeah, they really need to get back up to actually delivering chips that people want to buy. I mean, it ds really obvious, but they've - that seems to be in the biggest problem that they've basically, the quality of what they produce has just gone down and down. That was the most obvious, I think, that affected anyone in software is the Apple moving away from their chips. I mean, I had at least two MacBook Pros on Intel Silicon, and they were terrible. They were so unreliable, and I had used to have to get the whole logic boards replaced. It was just crazy.

[0:24:30] SF: I mean, that's partly what really hurt Blackberry back in the day when they were scrambling to try to catch up to the iPhone. When the iPhone came out and then they were just delivering a really subpar product. Even people who were loyalists to Blackberry like this is just not even functional. I can't use this thing.

[0:24:47] GV: Yeah. I mean, to maybe you or I, we would have maybe figured out, oh, it's Intel. It's the problem here. Our perception of Apple doesn't go down massively. We're like, so long as they can get a better product out, it's Apple. We like apple. I think a lot of people at that time started just associating Apple laptops as being a bit subpar. The fan was always on. It was supposed to be the quietest, fastest laptop, and it was anything but.

[0:25:13] GV: I mean, that's part of the motivation for Google to develop the pixel, even though they don't really mass produce them to that extent. Part of it was they wanted to own, essentially, the entire phone so they could show that you can make a really high-end Android phone. It's not just because it's Android; it's inherently wrong. The reason you end up with an Android phone that doesn't function very well is because maybe some part of the hardware, or something like that, that was manufactured was not done in a quality level that should have been done up

[0:25:42] GV: Yeah, exactly. Just to round out what's been going on. Yeah, Pat Gelsinger, who is CEO, he was brought back. He was part of the company, and then he was brought back in, and everyone thought it was going to be the turnaround of Intel, and it was the opposite. He focused on the Chips Act, which was basically lobbying the US government around more chips needed to be made in the US, as opposed to made overseas. Not to say that wasn't important, but it sounds like he maybe spent a bit too much time and focus on that, as opposed to the company itself, because he - I think it sounds like, he was a bit complacent around where Intel was, and this is during the time that Apple moved away. He spun up a bunch of expensive R&D projects, and they've all got canned now under Lip-Bu Tan, the new CEO.

Now, they're just Intel; they're having to sell off assets, basically. These are one of the first things Lip-Bu Tan was sell off 51% of the Altera programmable chip business that got 3.5 billion. Again, that doesn't sound like a lot, but I guess, you just look at all the drips and drabs of billions that they're accumulating, and sure, adds up to 10, 15, maybe 20. It just feels strange that they're having to actually do this stuff, like sell off assets and their Intel, but hey, here we are. Very interesting.

[0:27:03] SF: Yeah, absolutely.

[0:27:04] GV: AMD, though, this is, well, I see we missed this last month, but the point was last month, I think quite deliberately, there hadn't been a lot of chatter about AMD, probably, because they knew what they're about to announce. Yeah, they've announced that they're doing a big deal with OpenAI. What are the terms of that deal?

[0:27:22] SF: Yeah. It looks like OpenAI is seeking up to 10% stake in the company. As a result of that, their MD stock grew 25%. It's actually crazy how much some association with OpenAI, or one of the leaders in the AI space, dramatically impacts your public stock price, because those companies are perceived to be, probably, the future leaders of addressable market of companies out there in the space. We have the Fang companies, and the next generation could be the OpenAIs and the Anthropics of the world, essentially. Having any name association with these companies suddenly has a dramatic impact on your stock price.

Then, on top of that, they also announced a deal with Oracle Cloud. They just signed a deal for 50,000 AMD AI chips. There's a lot of recent news that came out with AMD. We had talked last time, as we mentioned, what's going on with AMD. They seem really quiet and then suddenly, they just came out with a whole bunch of stuff at once.

[0:28:21] GV: Yeah. There was an interesting live interview with Lisa Su, the CEO of AMD, and I think it was Greg Brockman at OpenAI, I think he's head of product. It was an interesting dynamic because it's such a huge deal, and they're having to talk about what this was even about, and Bloomberg is quite good at that. Being like, "Come on. Give us real details here. This is all a bit hand wavy." They basically said that they're doing a - this is 6 gigawatts of AI compute and inference over the term of the contract, which is about, well, I think it's at least three years, for example. But they think that 1 gigawatt is only going to come online in the second half of next year. Obviously, to get to six is, yeah, that's a bit of an interesting stretch there.

It was interesting. Lisa Su actually used the phrase, "We're tied to each other now," on Bloomberg, and they even commented on that. They were like, "That's quite a strong statement." Yeah. This isn't just, oh, we're investing a little bit here and there. AMD, I mean, that's the CEO saying we are tied to each other now. OpenAI is the part of our future. She said, Partnerships like this take years to get comfortable. Yeah, sounds they really, as you say, Sean, the stock price went up quite a bit. Clearly, the CEO sees OpenAI as a huge part of their future.

[0:29:40] SF: Yeah. I mean, it's valuable from AMD's perspective, where this isn't just OpenAI purchasing AMD chips. It's actually a true partnership where they're taking a stake in the company, too. That way, in some sense, the switching cost is higher, because OpenAI actually has stake in the AMD game at this point, where they have they own shares, they own common stock. If AMD does well, OpenAI does well. But this also goes back to, I think, one of the things we talked about last time, of just the rinsing of money that's happening in the space right now. It's like, OpenAI gives Invidia $20 and then it gives $20 back to OpenAI. Then they give that money over to AMD, and AMD gives that money back. There's a lot of this repeated transactions that are happening.

[0:30:25] GV: Yeah, exactly. I mean, there are these circular financing concerns across the whole industry.

[0:30:31] SF: Along the strip.

[0:30:32] GV: Yeah. Yeah, Nvidia invested in OpenAI, but then they use it to buy chips, and that's just going around and circling in quite a few of these companies. We'll touch on a little bit more of that just at the end of this segment, just in terms of any other circular things to maybe watch out for. Just as a wrap up, AMD, it was interesting. They did clarify that this inference, this compute that's being rolled out, gigawatt, and then in theory, up to six. This is not coming out of just AMD data centers. This is AMD chips being placed in wherever they can get them, basically. They said, Oh, Oracle are probably going to take a bunch, and OpenAI gets compute from Oracle already, or Oracle data centers, rather.

Yeah, this is again, that adds to the web if you like. This isn't AMD able to say, "Well, we can provide with all our own infra and assets, this compute." Still a lot of interdependence, I think, is what's going on here. There has been some collaboration between AMD and Intel. I mean, seems more on the technology side. I mean, just taking one example, they just announced it's called check tag. I'm not obviously a processing engineer, but check tag, it says, is a set of new and enhanced x86 instructions to detect memory safety violations, such as buffer overflows and misuses of freed memory. I mean, I think, yeah, we're all vaguely familiar with what that means.

Interesting that they are actually collaborating; these two are collaborating. I maybe pick up that Nvidia is not invited to the party here, and they need to. But yeah, I think interesting how who could be perceived as competitors are actually collaborating on the behind the scenes tech as well.

[0:32:16] SF: I would suspect that chip manufacturers they want to get - I mean, if you look at the - you see the explosion of growth and value that Nvidia has had. I'm sure Intel and AMD, from their perspective, was like, "Why isn't that us?" They're making the moves to make that happen.

[0:32:32] GV: Yeah. Just looking at Apple through this lens, I saw that an interesting analysis is very, I don't want to say this was some deep analysis someone had done, but I think it was a good observation, which was just that they thought that Apple is playing the long game here. Just launched the M5 chips. I believe I'm getting an M5. Today, actually is being delivered. I'm quite excited about that. It's a around the idea of running models locally. Purposely first, not needing to rely on expensive and environmentally unfriendly massive compute in data centers.

Everyone's been saying, well, where is Apple in the AI race, if you want to give it that name. It seems like they're really going for you can run models on our devices, and our devices will be best in class at local AI. Is that what you see as well, Sean? Do you see it differently?

[0:33:27] SF: Yeah. I mean, it seems like that's the market they're going after, which probably makes sense from an Apple perspective. I think they've always been very privacy focused company, at least, both from the way that they build software and hardware, but also from a marketing perspective. If they can try to own the market around locally running LLMS, and they feel they can get to a place where they have, that's certainly going to be a market. Even if you don't get to a place where you're running GPT, whatever, 10 from several years from now. You're still going to be running, probably fairly powerful models locally on these devices. If that's future and Apple can go in that, then I would say they're making probably good strategic bets in the long term to do that.

[0:34:11] GV: Yeah. They've got aiming for developers with Apple's Tensor APIs. This looks interesting, because I mean, I've got any strategies when it comes to running AI. This is the one that interests me the most, because I don't like the idea that we have to rely on these far away, expensive policing data centers. Yeah. It'd be nice that we can actually do a lot of it locally. Yeah. Obviously, from a privacy perspective as well.

[0:34:36] SF: Yeah. There's certain use cases that you would want to do locally, because a translation, for example, if you want to have be able to do real time translation, you don't want to have to pay for the network costs involved in that. Even if it's a hundred milliseconds, or something. There's going to be lag in being able to create an experience like that. You probably have to run that locally.

[0:34:57] GV: Yeah, for sure. Nvidia, what's going on there at the moment is "still up here to be doing well." Stock prices generally hovering in a good place. Price to earnings, for example, is still, I think, around 50 or something, which is not crazy. I know it sounds probably 20 years ago, that would be crazy, but it's in today's world. That's not crazy. That's the same as Shopify, for example.

They have a target on their back, basically, that lots of custom chip alternatives are emerging. There's just such a huge price premium on Nvidia chips. I think this is when there's a huge incentive for other people to come and undercut them and create these custom alternatives that fit very specific cases for much lower price point.

Jensen Wang he's downplayed this threat. He says, Nvidia provides complete service systems, including GPUs, CPUs, and networking products, and not just individual chips. You can't really - this isn't apples to apples. Yeah, I don't know. My take on this is I think there's going to be some decrease to Nvidia's market share. Maybe that seems obvious, but I think there has to be, and people will be able to create chips that just fit particular use cases better, is my hot take, I think. Yeah.

[0:36:19] SF: Yeah. I don't know. I mean, I think at the same time that there's potentially threats to Nvidia to siphon off some of these workloads and value, there's also this exploding market that's happening at the same time. Which is going to, maybe their growth slows down as a consequence of that, but I don't know that their growth totally goes away, or anything like that. I think they're still going to have a lot of growth. I mean, I think they're trying to bet on their entire CUDA ecosystem and developer tooling to go beyond just being a chip provider, but they're providing all this other stuff. This is indispensable training, inference as scale. They have a lot of different Lego blocks that companies can OEM as well, that want to be able to offer native inference within their, whatever their product is, or even agentic experiences and stuff. You can just pluck that stuff off of the Nvidia shelf and start using it, which is something that you're going to get from some of the pure chip manufacturing companies.

[0:37:19] GV: To wrap up this, we were just going to take a very brief look again at OpenAI and how are they influencing and affecting the space as well. We've talked about a bunch of deals they've done. I think the headline here is that they've done a reported 1 trillion dollars worth of AI deals in 2025 so far. I mean, I say so far, but every month here is like a year in the past times. That includes a 300 billion deal with Oracle for computer infrastructure. Over five years, a 22 billion deal with CoreWeave for data centers, within Nvidia GPUs.

They're just spraying cash, and I'm going to say spraying cash. I mean, we're talking signing deals. It's just very unclear how they're going to reach any revenue that can cover these deals, and that just seems to be the concern, really. Because these deals have propped up public company prices, stock prices, instead of, like, if a card starts to fall somewhere, how will this affect things? Because it seems like everything is glued together by OpenAI at this point of what are they doing? What is the next product? What next, dealer are they doing? Who's involved? If you're not involved, then you're nowhere. It's a little concerning just from a bubble perspective.

[0:38:38] SF: It does have some bubbly features, I would say. Clearly, not all of these things are going to be able to survive, and there's so much attached value to it. Like we talked about last time, OpenAI, I forget when exactly it was, but maybe a year ago. They almost died when Sam Altman was pushed out. There was a 48-hour period where -

[0:38:58] GV: That seems like so long ago.

[0:38:59] SF: Yeah, it does seem like a long time ago. But that was a very real situation for 24 or 48 hours there, where OpenAI, their future was pretty questionable. Obviously, they come back roaring stronger than ever, but it does show how fragile some of this stuff can be.

[0:39:17] GV: For sure. Yeah, I hope that's been helpful to listeners just understanding where all the chip makers today, and I guess, the whys as well, like why some of them are where they are today. Like Intel, we keep hearing that they're not doing well, but there are some reasons we touched on there. Yeah, we did get to finally talk about AMD having literally live, because we sort of like, "Oh, yeah. We've not really heard much about AMD lately." Of course, something quite big came out.

On to, yeah, the favorite part of the show, Hacker News highlights. As usual, I was trying to find developer off the deep end. I'll get that one in a second. First one was actually, it's related to the AWS outage. It was the fact that someone posted that their postman, which is an API platform, where you can send mock requests to what should be pretty local first experience. Turns out that when AWS was down, a lot of postman was down, which I think is a little bit confusing, when you're supposed to be able to run local requests. As postman, the product has evolved; it seems a lot goes through a central server these days. Yeah, and just full disclaimer, I got well fed up of Postman. I just found that the UI kept changing around, and it's super intuitive to set up groups of requests and this kind of thing. I just did a random search for what was an alternative, and this thing, Bruno, popped up. I've been really happy with Bruno. It's very just way paired back. It's open source. Yeah, I definitely felt this person's pain when they were like, Why, when AWS is down is my post, is Postman down as well.

[0:40:56] SF: Yeah. Actually, mine is related to that, which was, is this post of just use curl, and it's a website justuse.org/curl. The user, I think it's Emeric, or something like that. Essentially, the gist of it, and they really attack Postman, and it's a little aggressive, but it's funny. But I think the takeaway is let's not overthink things. Curl just works. It's on your computer. You don't need a whole service for this thing. I'm actually a fan of Postman, but I think that's still the sentiment of sometimes we can over engineer and overthink these things if we - and this is basic as like, hey, I just want to hit this API endpoint with some payload and see what happens. You can certainly do that with curl. You don't need a whole tool for it. There's lots and lots of examples of that where we go for something that's a heavier weight system to do something relatively simple.

[0:41:48] GV: Yeah. I love this post, quote. You know what's better than downloading Postman? Not downloading Postman. I think that's a reference to the social network movie; there's a line of something like that in there. I agree. I agree that there's a ton that you should be able to just get on with curl and not worry about a product like a postman. I have found having a product wrapper around it, quite helpful for, especially things like, I do a lot of email testing. Sending tests, email requests to a mail server that has to involve a ton of headers, as email does.

I'm not really sure that would be practical for me to do that all-in curl. It just to be able to actually see quite clearly what's being sent, and so on and so forth. Yeah, I'm not trying to plug Bruno massively here, but yeah, do try Bruno. I've just had a nice experience with it.

Yeah, the second one this got a huge kudos upvote, 1,700 at the time of me checking it yesterday, I think. It was called bypassing Kindle DRM. I think, I guess, this probably got just resonated with a lot of people, because people do believe if you buy a book, it's yours. There shouldn't be this lock in. Does seem you download it from the Kindle store and EPUB, which is an open format. Yeah, you try and run that EPUB on another device, like a Kobo, and you can't do it. This person just got very aggravated by this. Submitted by a user Pixel Melt. I think that's the user, because the blog is blog.pixelmelt.dev.

Yeah, this was really interesting. I'll try and summarize it very briefly, as best I can, how they actually got passed the DRM. They discovered that every page request on a Kindle is a randomized alphabet that gets deciphered, every single page. You can't just figure out how they've encoded it and then decode it. It's randomized on every page turn. They then discovered that they had, so then they've also got a whole in basic terms for just say, 26 characters in the alphabet. Obviously, there's more than that for books in general. It means you've got five potential things that could be an A and five that could be a B, and they've done these funny ligatures through the letters to make OCR very difficult.

Basically, what this person ended up doing was getting all the SVG renderings that could possibly be on a Kindle, then did a bunch of character matching, and then re-rendered the whole book in a new font and then recompiled as an EPUB. Yeah, it turns out that removing DRM is actually a lot about SVG and font rendering, because that's how they've decided to approach this DRM. Yeah, it feels crazy. I'm definitely, I'm on the side of Pixel Melt. If you buy an EPUB, just because you got up from Amazon, why on earth should it only be usable on a Kindle and not anything else? I don't agree with that, but yeah.

[0:44:49] SF: Especially, where the books are generally mildly less expensive than buying a paperback, but not that much. But you're still, you're buying a digital copy of this thing, which is roughly akin to buying a physical copy of it. I have a physical copy of it. I can do whatever the heck I want with the book. I should presumably be able to have a similar amount of freedom with a digital copy.

[0:45:13] GV: Yeah, it seems crazy. It's like, you're buying a physical book and every time you want to open it, you need this special key from the bookshop.

[0:45:20] SF: You got to use a decipher coder, or something like that.

[0:45:22] GV: Exactly. There's no, apart from obviously, money; there's no logical reason to have that DRM on it. As usual, Hacker News is always giving us some interesting insights, as well as, yeah, developers who find the time somewhere to do these deep dives on things like that. Looking ahead as usual, predictions, we've just talked about - the name of the game here is we're usually way off with our predictions, but that's what makes it fun to me, to just compare what might have happened that's vaguely similar. I'll go first this time.

My prediction is related back to the main topic of where are the chipmakers and the OpenAI web. I predict we'll see some concrete reporting on a crack in that web, something, just a lot of speculation right now. People are saying, well, there's all these deals. I think we might see something from one of the news outlets who've actually got a bit of information on something concrete on that one. That's my prediction. What are you thinking that we'll see over - well, might be for the next six weeks if we get back to normal scheduling?

[0:46:26] SF: Yeah, this is probably one of my least out there predictions. I think that after everything that happened with the AWS outage, I think we'll see some announcements of companies making some significant investment into be going multi-cloud.

[0:46:41] GV: Interesting. Yeah, that's a good - because companies do do that, but probably not enough. Yeah, the big enterprises do multi-cloud, because they need obviously that amount of redundancy. Yeah, you're right. We probably don't see enough of it. Yeah, cool. That's an interesting one. Let's see what comes up there.

As always, thank you for listening to another SED news. We'll be back in about four to six weeks as we get back to the end of the month schedule. Yeah, nice to see you, Sean, and catch up again then.

[0:47:09] SF: Yup. Thanks, everyone.

[END]