You can also listen to this podcast on iono.fm here.
SIMON BROWN: I’m chatting now with Viv Govender of Rand Swiss. Viv, always appreciate the time. You and I haven’t chatted AI in far too long. I want to touch on models first. We’ve got the Chat-GPT5.4, Nano Banana 2 from Gemini, which is Alphabet. We’ve got the Anthropic Claude Opus 4.6.
I’m reminded of a conversation maybe two years ago where we were wondering how good these models would get. I would wager that they’ve become really good. Yes, there are still flaws, they still make stuff up, but by and large we are seeing massive improvements.
Read: What AI can do (and what it can’t)
Google Chrome’s AI agent can now browse web on behalf of users
Apple plans AI-powered web search tool for Siri to rival OpenAI, perplexity
VIV GOVENDER: Yes, most certainly. Not to insult anybody out there, but if you are in the position that the models have kind of stalled, it’s more a case of the capacity, rather, to determine how good the models are.
That seems to be the case because we’ve had cases in which people like mathematicians have come out and … started to use AI to aid them in discovering new mathematics, new physics, et cetera.
We’ve seen the emergence of … which apparently, has this, levels like Opus … above Opus which apparently is too powerful to even be released to the public because it can look at any piece of software and determine new hacks, new what they call ‘exploits’ to the software.
Read: Alibaba, Tencent shares lose $66bn as AI vision falls flat
Simon’s weekly wrap: AI models, massive capex and the M&A lifeline
Avoiding the AI arms race and buying Netflix
And then it can even take one piece of software, find multiple exploits. Each of them may not be on its own powerful, but they combine and make like a combo punch that can give you access to certain sensitive data or allow you to do things that you are not supposed to do, et cetera. That’s just basically in the last four weeks.
We’re still waiting for … spun from Gemini OpenAI the new Gemini, the new Grok to come out. And quite frankly, the progress of these is as fast as has been predicted, if not slightly faster.
People who think it has hit a wall or any kind of barrier – that’s not true. In fact, it’s getting to a point where we just don’t have the tests anymore to determine how good these things are.
Read: Meet the new AI coworker who won’t stop snitching to your boss
It turns out AI trading bots behave just like humans …
How South Africans are actually using artificial intelligence
SIMON BROWN: That’s actually a great point, and I will attest. Opus 4.6 has just boggled my brain. And to be clear, I was quite happy with my AI before, but the new one was really great. To be clear, I’m using Anthropic. They currently get my $20 a month. I’m using Opus 4.6. Their problem is frankly capacity.
They don’t have enough compute power, and I can see it in my [computer] throttling.
Even just two months ago, I was running wild with Claude and now there are restrictions left, right and centre. This is proving a giant challenge.
VIV GOVENDER: Yes. Again, your $20 a month is not where you’re making their money.
Simon Brown: No, no, no.
VIV GOVENDER: Because what’s happened is with the emergence, especially of things like the really deep thought, the recursive kind of thought this thing goes through, as well as agents. Genesis one … at least [with] the same amount of money you’re spending on an engineer are these tokens, you pretty much are wasting the engineer.
Read: Winklevoss twins attempt pivot after Gemini loses money and employees
We talked about people spending thousands of dollars a day on these tokens. Some of the bigger companies out there [spend] hundreds of millions of dollars a year and, in fact, at current projections I wouldn’t be surprised if we approach $80-100 billion in recurring revenue for Anthropic by the end of this year. Okay, it’s about $30 billion at the moment.
This is the fastest growth we’ve ever seen in a company of this size. And with Anthropic, the thing is that they don’t have the money that OpenAI does; they don’t have the money Gemini does. They have some special source.
Whether or not it’s their philosophy of safety first, their principal philosophy, which attracts more workers to come and work for them, we’ve seen, for instance with Elon Musk, Grok… his philosophy. He has … created by his workers. And therefore you’ve seen basically every single founder of Grok leave. The same thing with OpenAI. People have left that.
Read: OpenAI doubles annualised revenue to $3.4bn, The Information reports
OpenAI goes from stock market saviour to burden as AI risks mount
Musk sues Apple, OpenAI, saying they hurt AI competition
But Anthropic seems to be able to retain workers, and therefore that’s why they’re doing so well with much, much less money … the guys out there.
But the way the system is working at the moment is there’s actually an interesting podcast by a guy called Rakesh Patel, in which he was basically questioning Diya Modi – this was a couple of months ago – and he said, ‘Isn’t it obvious that you’ve underspent on computer?’ Diya Modi was saying, ‘Yes, but if I overspend, it’s much, much worse. It’s like basically, I’ll go bankrupt.’ … everything.
But in this case, it’s definitely underspent and it just means that the chip companies out there, the memory companies – even though they’ve been hurt by some recent developments by Google – are probably still a pretty good bet at the moment.
SIMON BROWN: I take your point on that. The data out overnight is that Anthropic is doing a raise, and it looks like they’re going to clock the $800 billion valuation – which, as you say, is insane fast growth.
Nvidia – year to date there’s been a war and the like. Nvidia is up some 5%; it is doing fine. Meta, although they’re not really chips, have AI. They are flat for the year. Alphabet, Amazon are both in the chip game. Aside from other businesses, the Mag Seven, apart from Microsoft and Tesla, are actually doing quite fine over the course of the year, notwithstanding that we’ve had, what, six weeks of war in Iran.
Read/Listen: Tencent seizes momentum in China’s AI race against Alibaba
Meta is cutting several hundred jobs amid record AI spending
Mag 7 generating massive free cash flow
AI frenzy draws hordes to private markets in industry gold rush
VIV GOVENDER: And Tesla is up 5% today. I just looked at it. And Microsoft – apparently, what’s hurting them is that they basically kind of chickened out on the AI spend. That’s probably what they did – they pulled back and that was a mistake.
They were aggressive at the start, but they chickened out and they probably saw a bit of a wobble in Open AI’s prospects.
And we saw some of the big cloud companies, for instance, under pressure because what happened there was we saw that OpenAI may not have the money. And therefore people were thinking maybe these cloud companies are not going to get the money coming through.
But it has seemed that’s not the case. Part of it is, like I said, the … stuff has come out in the last, say, six months or so. Part of it has been the fact that the models are continuing to get better and better, the applications are getting better and better, and so people are still willing to spend the money.
Read: Tesla valuation looks unsustainable to Wall Street analysts
Microsoft signs $9.7bn AI cloud deal with IREN
But yes, I think Microsoft – the issue there was they were not as aggressive in AI and that’s what hurt them. It’s kind of the Apple situation.
SIMON BROWN: Yes, absolutely. And in fact, they proudly said a year ago or so, some time last year, that they were pulling back on compute. That is absolutely hurting them.
Viv Govender, Rand Swiss, as always, I appreciate the time.
Listen to the full MoneywebNOW podcast every weekday morning here.
#advancements #artificial #intelligence #models