🔮 My outlook for 2023
Deep tech, governments and capital; AI & big tech; Google's code red; Twitter
Happy New Year!
Every year, I share my outlook for technology, markets, and political economy in the 12 months that lay ahead of us. This year, I invited my colleagueto help with the assessment, and you can read our Q&A below. I’ll be talking these through (and some more topics) with paying annual members of Exponential View in mid-January. Please watch your inboxes for the details. If you aren’t an annual member, simply subscribe.
Last week, we published my grading of 2022 predictions, which you can read here.
Capital & catalysis
Chantal: How should we think about the tightening of the financial markets and the drying up of the supply of capital?
Azeem: It depends on the timeframe that we look at. Markets are always there. Valuations go up and down. The sentiment of the market and what it wants to relatively invest in varies, but over the cycle it evens out and it will back technology. Let me tell you about how capital is helping me tell the story of what might happen in 2023.
We’re in a really interesting moment because we’ve had a post-covid shock with economies going into recession, at a time when there is such deep strategic competitive advantage for nations and firms to be very good at tech.
It can be healthy when you move from a period of really excessive exuberance to a period where capital is a bit tighter. Money is expensive today: the Fed has raised rates by 4.25% in 2022; the Bank of England by nearly as much. Expensive money brings discipline and ends the bezzle. Froth is squeezed out. Fraudsters get crushed. Fakery is unmasked. Investors might be more risk averse - so founders need to work harder to make their case.
Expensive money and tanking tech markets chill venture investment. Fewer, smaller new deals. Startups need to plan for longevity, not growth. Yet the alarm bells being rung in the venture capital industry in 2022 need to be put in some kind of context. First of all, 2020 by the numbers was a stomping banner year for investing in public and private markets. By similar metrics so was 2021. The easy money flowing from loose monetary policy suppressed yields on all types of less risky assets. This brought new promiscuous capital, hunting for higher yields, into private tech investing. The result, some quarters later, was the rapid birthing of unicorns. The turn in 2022 (which will continue into next year) has seen a speedy decline in the creation of unicorns and levels of venture funding drop. So 2022 has been a worse year than the previous two but still well above, say, 2017. We need to be looking at the overall trend line: and that trend line is up.
I think the second thing is that a lot of investing was happening through momentum, almost a kind of FOMO-style investing, in spite of the fact that many businesses were clearly going to struggle to scale. Take, for example, 10-minute delivery. We were seeing enormous funding rounds in businesses like Gorillas. But these were hardly tech businesses.
It’s not the case that the tide is lowering every boat. Those firms that go straight to the bottom, should be there. Those that will succeed will be companies with a meaningful offering that goes beyond selling $2 for $1, a strategy adopted by many hyper-growth firms, like Bird, the scooter business.
What I’m concerned about is what will happen in the realm of deep tech. Deep tech has generally been underinvested by private capital. That’s mostly because it pays back over a longer period than a fund’s existence. But it is also because the milestones and risks are harder to evaluate. Can a deep-tech startup creating new materials really reduces its cost by a factor of a thousand in the next three years? Will it be able to scale if it does? How well will processes move from test-tube to industrial scale? These are hard questions which need to be figured out one case at a time. The ecosystem will need much more experience before the methods by which we assess, diligence and derisk these opportunities are well understood and more prevalent.
Take LanzaTech (CEO, Jennifer Holmgren, was a guest on the Exponential View podcast), which was founded 17 years ago. The problem is that the typical venture fund lives for 10 years and they have to make their money within that time. This makes it harder for deep tech companies to have access to capital.
This is where governments could catalyse funding by creating the right kind of incentives, programs, or signalling. For example, they could create a signal that virtual power plants and distributed microgrids will be part of our resilient clear energy grid network within the next 10 years. The idea is to clearly mark that a new industrial sector is needed - and that the state will support the creation of that market. That clarifies the opportunity for entrepreneurs and investors alike.
Read Azeem’s recent commentary on catalytic government in Wired.
Chantal: What other kind of catalytic action are you seeing?
Azeem: Climate is an area in which we’re seeing increasing catalytic activity. I can’t understate the importance of the EU’s work in climate, and what Biden is achieving with the Inflation Reduction Act with a massive signal and stimulus to the climate space.
But there’s still a real issue around managing the transition from an economy that depends on fossil fuels to one that doesn’t. We have lots of solar, but we’re going to need storage of many types, base load production and interconnectedness. The transition will involve some uncomfortable choices, certainly around natural gas or other hydrocarbons. It’s a complicated situation, especially as many countries are looking for a degree of autonomy, call it energy, food and technology security. It isn’t necessarily an autarky but just a sense that they can have some certainty on the provision of those three goods, regardless of circumstance.
Provenance of technologies and materials is starting to matter, as well as import and export restrictions. For example, in the US, a car maker or energy storage provider may not be eligible for a subsidy under the IRA if a certain proportion of lithium in their batteries comes from countries without Free Trade Agreements with the US such as China. And in all of these things, the question is how rapidly entrepreneurs and businesses adjust. For example, Tesla already made a move towards a cobalt-free lithium-ion battery. That’s because cobalt is a difficult-to-extract and ethically challenged mineral, which made it become a reputational brand risk.
The story of catalytic governments is direction-setting and de-risking, but it’s ultimately a collaboration with the private sector.
Chantal: You mentioned that funding is beginning to be tricky for deep tech. Who is going to fill the gap, and how will the sector adapt?
Azeem: The biggest issue around deep tech is about being able to align the amount of capital these companies need during their lifetime when it takes a lot of time and cycle times are slow. For example, if you’re genetically engineering algae to produce some industrial product, algae need time to grow. We’ll need to come up with a mechanism to allow deep tech firms to maintain their levels of funding.
I don’t know what that mechanism would look like, but this is where strong signals from government can be helpful to businesses and investors alike. This uncertainty has translated into a very high cost of capital for anyone building projects in that space.
The involvement of corporate capital also becomes important. Corporates, especially after these many years of romper company profits, are cash rich.
Even a small slice of corporate balance sheets could more than make up for any funding gap that deep tech firms require. Corporate investors, according to Silicon Valley Bank data (see graph below), have been increasing their exposure to venture deals. Pleasingly, they over-index towards deep tech and climate tech deals, where their operating and technical expertise is even more in need. One question will be whether top leadership, in the face of a recession and the workforce challenges post-Covid life presents, will continue to nurture frontier technologies as an important part of corporate strategy.
It takes a few years for the modus operandi or the standard operating procedure to develop as it did with SaaS. SaaS arrived at the turn of the millennium. Yet the standard metric and milestones that informed founders and investors took about a decade or so to standardise. For an investment ecosystem to mature: not something we have a lot of in the climate crisis.
AI & the competitive landscape
Chantal: What are you expecting from AI next year, and are AI companies becoming the next Big Tech?
Azeem: The hallmarks of an important technology appear to be there. The maturation of these generative tools, both in text and video and image, has sparked lots of useful experimentation and people building products quickly, and the creation of ecosystems around them.
Language and image-based AI tools are also making their way into products very quickly which, in turn, will help corporate adoption. Firms won’t want to be left behind. This will spur another wave of adoption - which has started to flat-line in the past couple of years, according to McKinsey.
But will AI firms, like OpenAI or Stability, turn out to become a new generation of big tech firms? OpenAI is on track to make $1bn by 2024, which sounds quick. But in reality, 2024 will be the outfit’s tenth anniversary. Google exceeded a billion in revenue in its fifth year. Facebook got there in its sixth. (And a billion bucks was more back then and the Internet economy was far smaller.)
And the idea of “big tech” is not merely about middling revenue in the low billions. It is also about structural advantage parlayed into strategic gain: power, in other words. Today’s big tech firms started to look big tech-ish by 2013/2014 but the phrase “big tech” really comes into its own a few years after that. The AI players are still a way away from that kind of power. But we are much more aware of how that kind of power can emerge—perhaps at an exponential rate.
Might AI-driven chatbots represent a major threat to the incumbent big tech firms? Meta’s main threat is its own boss, not AI. Amazon doesn’t seem wildly at risk: its core businesses, retail and cloud, are infrastructural. Apple’s devices simply get better and more appealing. But chatbots could seemingly disrupt search. ChatGPT has already been declared a “code red” by Google’s bosses.
I’ve got a view on this. On the technology front, it isn’t immediately clear that OpenAI has some kind of unassailable lead with large language models. Google (and many other firms) have developed their own large language models. Google is making some commercially available via Google Cloud, it is using others to improve search quality.
The chatbot product itself feels understandable. It is an interactive, persistent query of a data set. The queries are in natural language and the data set is a fuzzy, probabilistic hierarchy of representations in the language model. That looks pretty similar to Google’s existing product structure. Google has already demonstrated that it can adopt types of user input. It is far better with natural language queries, ambiguity or multiple words than it was in 2002.The underlying index doesn't just rely on PageRank anymore.
Interactive chatbots seem like an extension of what Google already does, rather than a brave new world. The obvious contrast is with social media. Google’s obviously flubbed and failed horribly with Google Plus. Social wasn’t a technology problem, it was a completely different product. Chatbot-based information extraction is much closer to the firm’s existing business than social was. (And Google’s index has the benefit of being milliseconds old.)
Could OpenAI have some technique for fine-tuning its chatbots that are somehow unique? The overall approach, reinforcement learning with human feedback, is rather well-known. And researchers are analysing how large language models are obtaining their abilities. OpenAI doesn’t have a monopoly on the know-how.
Having a great search engine or chatbot isn’t enough. You need to get lots of users to it. Much of Google’s traffic comes from smartphones ($15bn a year to Apple alone to be the default search engine), web browsers (two-thirds market share via Chrome) and habit. It’s unclear to me how easy it would be — without network effects or some other inducement — that large numbers of people who rapidly switch for a marginally better product. We do have marginally better products for searching for information. For now, Twitter is a better search engine for the topical currents that aggregate on Twitter. Booking.com is better for searching for hotel rooms. Almost any camera retailer (or review site) is better for searching for camera gear than Google’s index. But in the main, Google still reigns.
But am I being “faster horse” with this analysis? Perhaps the real risk comes from how chatbots might create new markets in a “blue ocean”, to use the framing of Kim and Mauborgne. What if chatbots become prevalent, perhaps attached to an individual or attached to an application? Today we might iterate across multiple Google searches to cobble together a piece of code, summarise the arguments in a book, or lay out the pros and cons of different routers. We might, in the future, do that through interacting with a specific chatbot.
Queries that we would have sent to Google, simply dissipate. And with them, the ads attached to those queries. A single chatbot, doing what Siri never delivered on, could take a meaningful slice of my search activity. I might even be prepared to pay for such a tool. I already pay for two, perhaps three, tools that use large language models. But they haven’t yet dented my Google activity.
So for 2023 at least, Google is safe.
Chantal: Let’s turn our attention towards the incumbents. What’s next for Twitter?
Azeem: So I think Musk’s management of Twitter reminds me of someone playing roulette. Many of the decisions that Musk has made are not just product decisions but have a trollish aggression to them. He’s in some sense made Twitter into a space that’s less welcome for women, and other minorities. Trust and safety teams are being axed. He’s playing to one part of his crowd. This is especially worrying as we think about Twitter of the aspects of Twitter that represent a public space.
I’ve been on Twitter for a very long time, and I’ve even built a company on it. I’ve not seen changes on Twitter since Musk took over that improved my experience; I’m actually seeing it get worse. For example, the quality of interactions is worse. Lots of the people I used to follow are not there. Discovery is not as good. There's much more spam in my timelines.It’s much nastier, and Twitter Blue is not a great product. There are weird glitches when I used the app on my phone. This doesn’t seem like momentum in the right direction, yet. (This is why you can find me on Mastodon.)
Bear in mind that Musk has, put a lot of skin in the game. He may whine and complain a great deal but his money is where his mouth is. Even if his instincts for Twitter are perhaps skewiff (rapidly changing moderation policies to unban people like Andrew Tate, hardly seems like a 90-day priority, for example), his instinct to succeed may sharpen the choices he makes and the direction he sets. And his own Cultural Revolution, wiping away most traces of the ancien régime, may ultimately create the conditions for genuine product and service innovation. Although it all feels rather more Great Leap Forward right now.
I can’t call it easily. Because I’ve not seen changes that naturally improve Twitter as a product—and it is a product I know something about—I imagine the service will deflate slowly like a balloon. But perhaps it is just too valuable to most of us to die and will soldier on. Perhaps, I’m 180 degrees wrong and Musk will point the tiller and take it to some amber-tinted climes.
Chantal: If you could travel one year ahead from now and ask Azeem one question, what would you ask him?
Azeem: I think that the question that I would ask would be the extent to which the manifest division between the tech industry and the rest of society, the thinking of history, criticality, and human experience, will get bigger or smaller.
It was a very problematic division in 2013 or 2014. And it's partly what drove lots of moves towards trying to get more women into tech, more women into venture funds. It was then connected to the ethical and responsible technology and ethical AI movements. A resurgence of division has emerged in the last couple of months, for reasons I haven’t been able to pick up yet. We should close it!
Happy new year
I’ve got a couple of trips in January. I’ll be speaking at DLD in Munich from Jan 12 to 14th. Ping me if you will be attending. I’ll also be at Davos presenting a more detailed horizon scan for the coming years. I’m also talking inter alia on generative AI and a couple of other things. Ping me on the event’s internal messenger if you’re attending.
Now, I can’t leave without wishing you all a Happy New Year.
I hope it brings health and happiness to all of you.
An example of this is when the US government phased out subsidies for internet backbones in 1994. It triggered private-sector investment from which we have benefited. An example of governments getting this wrong may be the obsession that some have with hydrogen as a potential fuel for domestic heating or ground transport. Such endeavours to foment a hydrogen economy are a boondoggle: they depend on products that defy the fundamental physics of hydrogen. My bet is on the physics.
I know because between 2001 and 2002 I was head of marketing at a natural language search interface. Google’s search engine was really terrible at queries beyond a couple of words, something we did better at. It wasn’t enough.
Most of my Twitter lists are still working, though. I’m spending more time on LinkedIn, although discovery there sucks. Mastodon has some really useful communities that I follow on it. Most of those have migrated from Twitter.
Is it a good or a bad thing that I let AI reate a summary of this newsletter? Feeling mixed feelings. So much Newsletters to read, but should I really let AI make a decision on what are the key takeaways for such a high quality one?