Shares in Japanese tech investor SoftBank have taken a knock, after it revealed it has sold its stake in chipmaker Nvidia.
SoftBank surprised investors yesterday by revealing it sold its shares in Nvidia last month, raising $5.8bn, to fund its other investments in artificial intelligence pioneers, such as ChatGPT parent OpenAI.
Now, me, I think it’s obvious we’re in a bubble here. The problem with bubbles always being not whether they’ll burst but when – therefore, how long to hold on before cashing in?
So, Softbank does cash in – and for a damn good price look like – and Softbank’s shares fall?
Hmm.
Perhaps because they’re putting even more into OpenAI and the market doesn’t like that?
Selling NVDA at $200 certainly not a bad idea, it was sub-$100 last summer. However, putting that money into openAI?!
However, correlation is not causation. More likely that the drop is just part of a wider drop. The Nikkei generally went a bit mental the last few months and has taken a bit of a breather. That Softbank should do similar is unsurprising regardless of some relatively small changes in allocation.
Meanwhile, AEP takes his usual contrarian stance:
Ignore the doomsters, the economy is not in crisis
https://www.telegraph.co.uk/gift/35c396ab7912c29a
If AEP says everything is rosy, then we are definitely doomed.
Indeed!
He seems to be saying that the UK will be thriving as long as its socialist government does not do any socialist things.
Can’t argue with that.
There’s an old stock market maxim; It Is Never Wrong To Take A Profit. The rational behind it. Stock market prices are future looking. They’re pure opinion. And opinions change. Prices can go down as well as up. There comes a point on any price curve where the likelihood of price going down is greater than them going up. The art is in correctly assessing that point. The longer you stay in the higher the risk. Taking the profit is the lower risk strategy.
I had a lucky buy on Kenvue a week or so ago. Bought it, the following day, Kimberley-Clark put in a bid, shot up 15% in a day. I sold. Because you never know if those things fall through.
Why did you buy them in the first place? Has the bid changed your opinion? There’s an old market saw – up on anticipation, down on realisation. (Or vice versa) Maybe you buy them back later at a lower price
Honestly, I thought they were just a little beaten down because of some nervousness about lawsuits. I was thinking like hold for a year or so. When I got 15% in a day I figured I’d take the profit. That was a lot of my “underpriced” value.
There’s two investment strategies. Hold the stock on the basis of YOUR view of the future profitability of the company. Hold the stock on your opinion of OTHER PEOPLE’s opinions of the future profitability of the company. Most investment strategies are a mixture of the two.
So, Softbank does cash in – and for a damn good price look like – and Softbank’s shares fall?
How much sunk cost fallacy is involved in this? Softbank share price is opinion. The opinion seems to be that the underlying investments have risen in the past & therefore should continue to rise. Therefore Softbank should have continued to hold them. Softbank may have looked at those investments from the point of view; would it make those investments now at these prices? And decided no, too risky. So therefore the correct decision is to sell them. Looking at Softbank’s track record, I’d be inclined to go with their judgement.
Now you say all that, it’s got me thinking that this is Softbank moving investments for the endgame of the AI Bubble.
Softbank, Microsoft, Sequoia Capital et al have been pouring money into OpenAI and Anthropic. Basically subsidising growth and operational costs. They can’t do this forever. So what you do is pump money in, build hype and then at peak, you dump it on some rubes that can’t figure out that profit matters.
So, the events go: 1) OpenAI floats 2) Softbank makes a profit 3) everyone now holding OpenAI shares has loss-making shit which they suddenly realise and so the price tanks 4) this impacts on purchases of Oracle and Nvidia. Moving out of Nvidia at this point and into OpenAI makes a lot of sense to capture that bit of profit near the end, and before NVDA crashes.
Softbank owning a larger chunk of OpenAI will pay off in about 6-9 months, when the floatation happens.
“Erik Brynjolfsson, the head of Stanford’s Digital Economy Lab and a leading AI guru, says the productivity effect is a “J-curve”. It dips first as society learns how to use it. You have to redesign systems and retrain employees. Companies spend billions investing before they see any return. “Once you figure that out, things really take off,” he said.”
The problem is that Bayesian Inference suits either low value. low risk tasks (playing you a song you might like on Spotify) or it requires human monitoring.
Because it’s a “might be” rather than an “is” type of calculation. So you can’t leave the software to just run. You have to check it. And you always will. That means you don’t save a lot of humans.
And honestly, I think we’ve mined most of the value of it. A self-driving car that goes from 97 to 99% makes no saving. You still need a human in the loop.
Interesting points. I completely agree with the J curve argument, also the need for a human to still be in the loop. For all the hype we’re a long way from true autonomy.
On self driving cars, wake me up when they can drive me home from the pub. Actually, don’t wake me up, I’ll be asleep in the back seat. The car will beep the horn when it’s time to go inside, right?
It’s a fantasy. These robotaxis are all running with people monitoring them at quite a high ratio. It doesn’t add up.
I agree, and I don’t expect to see real autonomous vehicles in my lifetime. But that’s my point. Can it get me home with no intervention? If not, it isn’t truly autonomous.
I do, on the basis that they’re basically just killbots that don’t kill people on purpose.
There’s immense incentives to develop a killbot right now, and you can couch the development as “assisting the operator” in dealing with poor signal, high latency, and so on. For killbots there’s very few regulatory barriers, and you get to test them in the most adversarial conditions possible.
The intelligent drone teams that share targeting, sensor information, mesh network with each other, cooperate on hitting multiple targets, find their own routes through forests, identify targets and (currently) ask if their master would like those targets blown up, well that’s all just hardware and software to produce cars that cooperate with each other to all accelerate when the light turns green, come up with a plan in realtime to not hit a pram and not hit each other, deliver a pub’s worth of passengers to their various homes at kicking-out time and so on.
Because it’s a “might be” rather than an “is” type of calculation. So you can’t leave the software to just run. You have to check it. And you always will. That means you don’t save a lot of humans.
I’d agree with you except that’s how us humans run. The real question is does the software fault more often than we do? I’d say it probably depends on the application. I suspect that a properly implemented self driving car system would fault rather less often in autoroute conditions or city centres. Humans are not particularly good at handling multiple simultaneous data inputs. We actually do it by dealing with one input at a time & switching between inputs & can get distracted into not monitoring one or more inputs. We’re also very bad as assessing relative velocities & vectors.Go do a flying training course & you’ll be told how bad we are at them. They hammer it in. The real problem’s extending self driving beyond those two areas would be prohibitively costly. To make it work you really want data sharing between vehicles via a central control & extending that to all roads would be nigh on impossible.