00:00Now one of the new terms we're hearing all the time around AI is something called AI tokens, but what
00:05are they?
00:06Well, tokens are tiny units of data that can be seen as the vocabulary of the algorithms, which do the
00:13processing in any given task.
00:15The more advanced AI models use far more tokens to process tasks, and that, of course, has a cost implication.
00:22This is where China has an advantage.
00:24Its cheaper energy and greater efficiency allows it to produce tokens for less than its competitors and therefore offer customers
00:31a lower price.
00:32That is important because NVIDIA boss Jensen Huang says the production and use of the digital units will drive the
00:39AI economy.
00:40Now, earlier this month, Chinese tech giant Alibaba announced the creation of its Alibaba token hub.
00:47It's a new business group.
00:48It aims to take advantage of this development.
00:51But U.S. groups are racing to catch up by building their own AI factories.
00:57Open AI, Anthropic, and Google have all reported strong revenue growth.
01:02Now, Rob Nyes is the founding partner of technology venture capital firm H3 Capital.
01:07Joining us now, hello, Rob.
01:08Please explain very simply how something that is essentially a way of understanding the patterns of AI is monetized.
01:18You're basically selling units of time or units of effort.
01:21I mean, if you think about what a token is, it's simply put a store of value.
01:24So whether it's a minute of compute power or a certain amount of data back, it's just a way of
01:29counting effectively.
01:30So it's really nothing new.
01:32It's just a way of counting them in a way between systems that seems mostly interchangeable.
01:37Why has China got the edge in producing AI tokens?
01:42I think it's like any commodity that when you have these markets, they become very efficient.
01:46And then it comes down to who has the lowest cost of goods sold.
01:49In the case of China, you have very low energy costs comparatively for the rest of the world.
01:53And you're able to bring up capacity fairly quickly.
01:55So I think in any commodity market, you see a race that people compete on price and people compete on
02:00quality.
02:01And there's not a middle over time.
02:03You know, we've always looked at how much data we get for what costs.
02:06But it seems the game has changed completely.
02:08And I'm wondering how much AI agents, which just goes so much further than our average chatbot, for example, is
02:15changing the game.
02:18It's pretty amazing because they can learn and they can gauge and test their own work to determine do they
02:23have to do it again or try harder.
02:24So it's like having a human reviewer in some cases where AI is monitoring the other parts of AI and
02:30saying,
02:30OK, do this or chunk out this task, these different steps.
02:33It's pretty phenomenal because it is the way a human mind typically works in terms of segmenting tasks.
02:38And, of course, all that just chews data.
02:40So we need so much more processing power.
02:43And that's why we've got these AI factories.
02:46How do they differ from the traditional?
02:47I mean, gosh, traditional is relative data centers.
02:52Sure.
02:52A regular data center is basically on or off.
02:55The computers are on, computers are off.
02:57There's not really much variability that historically computers, the CPU, takes up so much power in a given time.
03:02That's fixed.
03:03With AI data centers, because they use something called a GPU, which is what powers machine learning,
03:07those are a lot more variable where you can turn them on and off on the fly.
03:11And they're sort of clustered where if the local grid around the AI factory needs more capacity, they can pull
03:17down their load.
03:18If there's extra capacity, they can pull up their load.
03:20So it means, you know, in cases you can put data centers where otherwise they didn't work because there wasn't
03:25enough capacity
03:26or places like hydro plants where there's excess capacity but very peak loaded.
03:31I understand that tech businesses and startups are already implementing AI tokens to incentivize their employees.
03:38Could you explain how that's working?
03:41Effectively, it's becoming a digital currency.
03:44If you, like a gift card or, you know, any store of value, it presents value to the user.
03:48So in this case, it's the same as, you know, buying credits at Google or buying credits at Anthropic.
03:53It becomes, you know, a very convenient source of exchange.
03:56And so I think the interesting part is when these do become more financialized, that over time, you know,
04:00will people be selling these on a secondary market or arbitraging different times of day?
04:05I mean, this is it.
04:06Is it going to become sort of tradable?
04:08You know, are we going to have indices, that sort of thing?
04:11I think it's very likely.
04:12The hard part now is, you know, measuring quality.
04:14So on the low end, sort of for basic power, that's fairly commoditized, I think.
04:18It will be more commoditized.
04:19On the high end, the quality will matter.
04:21Like luxury goods, you'll have very high-end AI models that can do very specific things at a higher cost.
04:25But I do think, given that the producers control the supply of the tokens through the machines, the data centers,
04:31they do have the ability to constrain the sale.
04:34But I think everything leads to this path, that everything that can be financially broken down turns into a financial
04:41instrument.
04:41So it wouldn't surprise me at all if people find some way to trade these on the market.
04:45That is so very true.
04:46Thank you so much.
04:47Rob Nies is the founding partner of tech venture capital firm H3 Capital.
Comments