Mid-year review 2023 - AI & Robotics: Well, that escalated quickly

Table of Content

Portfolio Snapshot

AI & Robotics Overview

Computing & Automation Systems
Productivity AI
Artificial Intelligence Platforms
Data Management
Data Infrastructure
Robotics
Autonomous Mobility
Semiconductor Manufacturing
Machine Learning
Virtual Universes
Augmented Reality
3D Printing

Year’s Favorites 2023 – An update

Catalysts/Risks

Portfolio Snapshot

Overview

The AI wake-up call

The Big Bang happened…

In our previous outlook, we said that we expected 2023 to be a foundational year for AI; judging by the first six months, we can say it is definitely the case. The first quarter witnessed a massive hype cycle ignited by the launch of ChatGPT, which triggered a technological race from competitors all around the world not to be left behind. Concerns quickly emerged about a potential bubble, but they were put at bay by the earnings season. We have every reason to believe these are still the early innings, as the uptick in companies’ financials remains at a very early stage for now and the spending cycle has just begun. This spending uptick will propagate all along the supply chain, from chips to Data Management players, and we have positioned the portfolio accordingly to capture such a trend. 

… but the best is yet to come

ChatGPT is the poster child of AI applications but is far from being the only one. Behind its creator OpenAI, large players are either trying to replicate the technology (Alphabet) or to apply it at a larger scale (Microsoft Corp). But behind these juggernauts, the blatant teaching is that AI is on its way to penetrating every business, with smaller companies also building applications on top of external or in-house technology. Possibilities seem limitless, as the core technology powering generative AI (the infamous Transformer) appears extremely flexible and adaptable to a variety of use cases. The best is yet to come!

"Tis but a scratch"

As we previously said, this does not mean that the road will be without any bumps. Major breakthroughs usually bring their lot of anxiety and fear, and this is definitely the case here, especially when considering the cultural background of AI. Some parts of civil societies are already warning – rightfully – of potential abuses, and regulators will ultimately have to conduct their business. However, as we believe that the cat is proverbially out of the bag, and to draw a parallel with the American Wild West, this does not mean that the end of the road is ahead, but merely the end of the frontier.

Back to Table of Content

Chips are leading the way

Well done, Nvidia, well done

It is simply impossible not to mention Nvidia in this document, as it was indisputably the stock most impacted by the ChatGPT hype during 1Q23. As a consequence, many were afraid of 1Q results to be a day of reckoning. In the end, the company did not even set a single toe in purgatory, defying even most optimistic expectations, and today appears as the undisputed winner of the AI tsunami. Once again, we believe this is just the beginning: the infrastructure is far from being correctly dimensioned yet, as end-users and businesses are just starting to adjust, and it is likely that even Nvidia’s sky-high guidance may, actually, end up being conservative.

More than crumbs for other semiconductor players

Nvidia’s hegemony does not mean that other players will not see any traction. While it is true that GPUs currently are the go-to chip for Machine Learning (ML) workloads, specialized chips are progressively coming to life, with Google already being at its fifth generation of TPU chips. More generally, due to AI training and inference remaining computing-intensive, the entire data center supply chain is expected to benefit, from CPUs to networking and memory. There may be a lag compared to GPUs, but a slight uptick was already visible during the latest earnings season (e.g., Marvell ). More is indeed to come.

Various fortunes for non-AI players

Players not directly exposed to the AI supply chain have witnessed various fortunes, due to ongoing tough macroeconomic conditions. Consumer spending remains in a ditch, leading to inventory normalization taking more time than previously expected. Chinese players in the automation supply chain have been impacted by the deteriorating sentiment. The only exception seems to be those exposed to the Computer Vision (CV) segment, due to… an AI-induced bubble. There seems to be no way around it. But with the increasing probability of an economic slowdown, stimulus measures and investors turning their look toward the recovery cycle may be of help.

Back to Table of Content  

Hopes, but also some doubts

An opportunity rarely encountered

On an equally-weighted basis, Productivity AI has been the best-performing segment in our investable universe during the first semester. This performance is not a surprise: the quest for more automation has been a constant for companies since automation exists, as it enables productivity gains that fuel higher profits. With the advent of generative AI, players offering this kind of service face a once-in-a-lifetime opportunity, with capabilities becoming on par with expectations and customers lining up at the door. Now is the time to deliver, buoyed by uncertain economic perspectives pressuring companies for more efficiency.

Are the disruptors already disrupted? Not so fast!

Concerns have been voiced regarding Robotic Process Automation (RPA) players (e.g., UiPath Inc) being disrupted by generative technology even before reaching their prime. We tend to disagree, as we see both technologies being a perfect fit for one another. While we do acknowledge that generative AI may indeed lead to low-code interfaces becoming obsolete (there is indeed no simplest way to create an application than asking a bot to code it for you), they need to be integrated into some kind of framework to have an impact on business operations. This is where RPA players have an edge, as their platforms are designed to connect to many components of a business. We, therefore, see generative AI as an enabler and a tailwind rather than as a threat for this segment.

Is it too late for consumer AI assistants?

Once touted as the future, AI assistants in their current form appear to be marching toward their sunset. Their capabilities, although useful in some specific cases, are indeed far from what chatbots like ChatGPT can propose, and their business case remains uncertain, as users are concerned over the use of their data. A deep upgrade is therefore necessary, and may well illustrate the classic technology evolutionary landscape: ultimately, the top of the chain is not made of first-generation players. However, we do not rule out a rebound in the subsegment, especially with the advent of new Large Language Models (LLMs), but probably under a different form.

Back to Table of Content  

Now the toughest part begins

Time to show the cards

AI platforms, i.e., companies applying AI technology within a specific vertical, have naturally been strong beneficiaries from this year’s enthusiasm, if just for the free advertising. However, now that investors have hopped in, they will want to see some results in return. Execution will therefore be critical in the next few quarters, as we expect those with lofty valuations but not delivering to suffer a painful reality check.  And in this regard, one should not be mistaken: generative AI has been hyped to the extreme, but it is not the answer to every problem, nor is it easy to integrate into a mature product portfolio.

Strong developments

Examples of good execution over H1 have proven that the reward can be substantial. Defense-oriented Palantir Technologies Inc, after much work, has finally turned a profit on a non-adjusted basis, sending the stock on a parabolic trajectory. AI-assisted drug discovery player Schrodinger Inc enjoyed a similar fate after demonstrating that its platform was indeed a clear differentiator for its business operations. Companies clearing the hype threshold will be able to enjoy all the benefits from AI: major product differentiation, higher sales, and leverage on margins. All that an investor could dream of!

What about regulation?

That is, if regulators do not step in too hard. While we think that, over the longer term, regulation is desirable for the segment to fulfill all its potential, strong ethical backlash from some parts of the civil society may pressure regulators into over-dimensioning their action. This would particularly be the case regarding the access to, and use of, data. On top of that, the lobbying from some high-profile players might lead regulators to favor some parts of the market at the expense of others. One thing is certain: regulation is bound to happen, but it may prove difficult to anticipate properly a fast-moving sector.

Back to Table of Content  

Sitting on a gold mine

A self-perpetuating momentum  

Data being the 21st-century oil is not something new. However, the rise to prominence of LLMs led many to the realization about how important they actually were in enabling modern AI technologies. This, in turn, kickstarted a virtuous cycle for the segment: data management solutions would become a sought-after resource to enable AI applications, and complexifying AI applications would require an increasing amount of sophisticated data management solutions. There are good chances for this cycle to only be at its beginning, as the quality of data will ultimately be the main driver in differentiating between increasingly accurate AI models.

Doing more, doing differently

The amount of data being generated is exponential, as more and more people and devices become connected over the years. Data management systems, therefore, face a two-pronged problem. On one side, they must ensure that this data is usable, which is not always straightforward, especially when it comes to unstructured data; on top of that, this consistency must be maintained across the full data spectrum. On the other side, the data must be relevant for its final use, which can be the equivalent of finding a needle in a haystack. Ultimately, both these problems are taking a toll on the amount of resources required, driving up costs and leading to the apparition of new business models, notably consumption-based ones. The resulting transition periods are likely to increase uncertainty and risks.

At the forefront of ethical concerns

The use of data is becoming increasingly problematic. Polemics are emerging at an accelerating pace. Individuals and/or businesses are complaining that they did not consent for their data to be used in training datasets. Privacy concerns are central. Intellectual property disputes are on the rise. All this may lead to massive legal battles, which must already be anticipated. Data management players will have on one side to conduct their business, and on the other one to develop tools to provide fine-tuned traceability and auditing capabilities. This will, of course, have a cost, but it is the price to pay for a sustainable business.

Back to Table of Content  

Some slowdown, but still the strongest driver

Feeling the pressure, not the pain

There might not be any sign of doom for cloud spending, but the segment has definitely started to feel the impact of the macro slowdown. Total spending has indeed fallen below the 20% threshold for the first time ever, as companies around the world are optimizing IT spending to protect their bottom line. However, the recently published more than respectable 19% growth figure shows that the segment remains extremely dynamic, as the technology enables major operational efficiencies vs. on-premise infrastructure.

Playing the optimization

This more cautious spending has clearly boosted players in the monitoring segment, such as Datadog Inc. Already enjoying the tailwind provided by cybersecurity-induced concerns for their observability services, these players are ideally positioned to benefit from the optimization phase, as their product portfolio is precisely built to give their users a better view of who is doing what in order to optimize resource consumption. In the end, it is no surprise this subsegment outperformed the broader Data Infrastructure segment, despite the presence of some of the biggest names from the technology sector (Amazon.com Inc, Microsoft Corp) to skew the performance upward.

AI to buoy the segment

AI will mechanically benefit cloud players. Previous AI applications already relied on an internet collection for portable devices to let the heavy lifting work be done by central servers. The trend is not going to reverse soon, as generative AI is extremely demanding in terms of computing power: even running inference workloads of a ChatGPT-like model requires several dozens of high-end GPUs, and the requirements are even higher for training. Both economic and operational efficiencies, therefore, demand the use of cloud-based services, and in this light, it is striking to see that the main rationale for the partnership between OpenAI and Microsoft was the supply of a dedicated cloud infrastructure capable of supporting operating the entire ChatGPT workflow. As we said, with this application being the fastest-growing in history, the cloud is poised to benefit.

Back to Table of Content  

Eclipsed by AI, preparing its revenge?

More or less what we had expected

In December, we had a very positive view on warehouse automation, a positive one for healthcare robots, a negative one for personal robotics, and short-term concerns over industrial robots. Performances in warehouse automation and personal robots validated our thesis. However, the performance of medical robots didn't match our expectations, due to macro concerns weighing on sales forecasts; only the players with the biggest installed basis managed to do well, as they are considered more immune through higher recurring revenues. Industrial robots, on their side, did pretty well thanks to a solid momentum in China, which managed to overcome doubts over the state of the global economy; interestingly, this momentum also benefited non-Chinese players.

Logistics to remain the main driver

Warehouse automation was by far the best-performing segment in our Robotics universe. This is no surprise. As we said in our previous note, the sector offers the ideal solution to rising inflation in the logistics services segment. Robots are progressively becoming competitive with human workers on both cost and capabilities basis. In addition, macroeconomic uncertainties are powerful incentives for companies to try to boost their operating efficiencies, something robots are a perfect match for when deployed in the right conditions. Combined with the rollout of “recession-friendly” sales models, such as consumption-based models, we do not see the tide reversing any soon.

AI convergence

While generative AI has been touted for chatbot applications, its potential impact on robotics has been much more muted, probably because people are traumatized by would-be killer robots. We believe the technology may be a game-changer for robotics. It has the potential to transform dumb machines into truly collaborative robots, blowing current cobots’ capabilities out of the water. Such applications will not materialize overnight, but the first proof-of-concept projects are already taking form. As generative AI technologies make tremendous progress, especially on the multimodality front (i.e., the capability to deal with both text and images), operational applications may not be that far away, and may at least be foundational in some interesting equity stories we are closely monitoring.

Back to Table of Content  

Getting there, step by step

Steady progress, although maybe not that spectacular

After much fanfare in the mid-2010s, the Autonomous Mobility segment has made a U-turn and has mostly disappeared from public view. This does not mean that innovation has stopped. In the U.S., Waymo has doubled the size of its serviced area in Phoenix ahead of its 2024 expansion in Los Angeles, while Cruise crossed the 2mn-mile threshold barely 3 months after crossing the 1mn mark. In China, Baidu and Pony.ai are also progressing regularly, with tests having started in Beijing (among others) with full driverless rides. Baidu is also about to launch its 6^th generation driverless vehicles, with a cost of ~$37k. But Chinese players may face problems in sourcing some specific computing components due to U.S. sanctions.

Technology is still somewhat struggling

The sector has been a large benefactor of the fall in sensors’ prices (which, on the other hand, has been a bloodbath for lidar makers), allowing to cram more capable ones in vehicles and increasing redundancy. However, the software component of the equation appears to still not being ready for larger-scale deployment, with vehicles breaking unexpectedly or not being able to manage complex situations yet, at least from what public information from U.S. test deployments shows. Chinese players do not seem to encounter this kind of problem, but this could very well be linked to a lack of information, although Baidu told us they had never been involved in an accident for which they were considered liable. In any case, both sides remain safely deployed in well-defined zones allowing them to avoid the most complex situations, but the big leap towards more complex zones appears to be closing in along the exponential number of rides already recorded.

Safety first

The critical point for public acceptance of the technology remains safety, as this was the foundational promise of the technology in the first place. On this aspect, the track record appears to be mixed, with the latest casualty being a dog hit by a Waymo car in San Francisco. This explains the extra cautiousness of players, with vehicles frequently taking routes significantly longer to avoid tricky situations, such as rush-hour freeways. But such cautiousness will inevitably come in the way of economic returns at some point, especially in a context where rising interest rates may constrain funders in keeping up with the large capex required.

Back to Table of Content  

Mixed dynamics

Consumer still a headwind

Despite a rebound on a YTD basis driven by the positive sentiment on AI, the segment remains under pressure due to the ongoing mixed dynamics in the consumer segment. Indeed, consumer accounts for roughly half the semiconductor market, and products such as smartphones and PCs are still engaged on a downward path, with the latter’s shipments falling by 30% YoY in 1Q23. On top of that, inventory clearing appears to be slower than expected, putting additional pressure. Although a rebound is expected in 2H23, mixed macroeconomic perspectives and elongating upgrade cycles in a post-covid boom add some uncertainties.

Traction from AI chips

On a general basis, AI chips (i.e., mostly GPUs) belong to the most complex level of sophistication. Usually reaching a respectable size, they push to their limits existing manufacturing processes to extract as much performance as possible. Their manufacturing is, therefore, complex, meaning they command a higher manufacturing price benefiting the supply chain, especially for TSMC. However, this higher price gets diluted by relatively low volumes compared to the broader market, a situation that is unlikely to change before the arrival of more generic ASICs, and, most importantly, of specialized dedicated hardware for smartphones, which will only be possible when algorithms have matured a bit.  

China’s turmoil

Nothing has changed much on the sanctions front regarding China, but this does not mean that the issue has become less pressing. It increasingly appears that the country has no choice but to develop a local ecosystem, as demonstrated by local players frantically ordering as many GPUs as they possibly can before the latest sanctions take their full effect. Invading Taiwan is unlikely in the short term, contrary to what might have been feared not so long ago. We, therefore, expect local players to benefit from sustained support plans, as the needed step-up in the ecosystem will not happen overnight.

Back to Table of Content  

We’ve seen some rockets rising slower

Peak enthusiasm, or is it?

If a sector has benefited from the AI craze, apart from computing components, it is definitely the Machine Learning (ML) one: a company such as C3.ai Inc is up by nearly 300% YTD, despite serious accusations from a short-selling investor. Most of the companies we follow, even though too small or illiquid to integrate our investable universe, have approached the 100% threshold at some point during the semester. “Bubble” is on many lips – rightfully so. But there are also valid reasons for such enthusiasm, as ML players are critical in ensuring the larger deployment of AI technologies for people and businesses not necessarily having the skills to do so on their own. Execution will be critical to sustain such returns, but will likely reap huge benefits.

Practical hurdles to overcome

Training an AI model boils down to making correlations within a dataset. Accessing the data and ensuring its quality is therefore paramount, but may prove increasingly difficult - for privacy reasons in the first place. Due to negative feedback loops then, as generative AI will become increasingly present on the internet. But the biggest hurdle may actually come from the discrepancy between the exponential quantity of data requested to finetune models, and the actual quantity of data generated annually. There is, therefore, an opportunity for ML players operating in synthetic data, or managing to build adequate databases to feed much-in-need customers.

All Quiet on the IPO Front

2022 has been a blood bath when it comes to valuation multiples on the listed market, while it may not have been the case in private equity, at least to such an extent. As the recovery has been rather muted in 2023, the incentive for IPOs has been all relative, explaining, as we had feared, that the best names of the private ecosystem are comfortably staying out of the public markets. We do not see any meaningful evolution of this situation by the end of the year, but definitely long for the likes of Databricks or Dataiku to make their move.

Back to Table of Content  

Time to talk about realistic products

So long Metaverse, it was not even fun while it lasted

We had anticipated some kind of hibernation for Meta Platforms Inc’s Metaverse, as we believed the concept was not in a limited supply of shortcomings. But in the end, its creator simply decided to kill it, at least in its original form, and to reallocate the corresponding investments, which were taking a heavy toll on the company’s cash flows without any perspective for returns in a reasonable timeframe. We cannot say we are surprised, as we believe the launch of such a project was decided to put some distance between the companies and past major scandals in the first place. Hopefully, the new investment target will not suffer the same fate, as Meta has shifted its desire towards… AI!

What consequences?

The apparent loss of its main supporter might have dealt a fatal blow to any sector. We do not think it is the case here, as it may actually be a blessing. Meta will keep developing some technologies related to the sector, such as headsets, but will target them to a narrower audience, which may allow it to iron out the inherent kinks with much less fanfare than it would have been otherwise. More interestingly, the company’s focus on content creation may give birth to technological tools to create virtual universes from scratch in the future. The Metaverse may be dead in its current form, but we may well meet its heirs in some time.

Industrial universes keep on going

Meta was not the only player active in the segment. In the shadows of this failure, Nvidia has been progressing on its Omniverse project, i.e., a virtual universe for industrial applications. Maybe not as sexy, but definitely as useful, if not more. Nvidia’s universe approach is, in our view, progressively giving birth to an integrated framework that may transform the manufacturing sector. At the very least, it will help the company to fuel its transition towards software further, and perpetuate its domination in differentiating ecosystems as was the case with CUDA.

Back to Table of Content  

Apple has come, what now?

One launch, many questions

Apple Inc has finally unveiled its augmented reality device. However, its high price tag ($3’500) has generated concerns over the company’s ability to sell it in large volumes, leading to muted reactions from investors. In addition, many wondered if people would accept to use it in a social context while questioning its actual capabilities. While we do agree that the device’s success is no sure thing, one should remember that the first iPhone and the first Apple Watch were, in retrospect, far from being mature devices and did not yield their full potential. We’d like to leave the product a chance before taking a final judgment, as recent history has shown that Apple made not many mistakes.

Untapped potential

The demand for augmented reality applications appears limitless, especially when the enthusiasm around AI collides with already practical applications. While hardware appears to be catching up, or will have to in the wake of Apple’s headset, the software is not staying still, with the new generation of game engines (such as EPIC’s UE5 engine) blurring further the boundaries between simulation and reality. Applications in healthcare, engineering, content creation, and many more, are taking note.

Not the ultimate weapon

It can be tempting to see emerging technology as the solution to every previous problem. In this regard, augmented reality makes no exception and may fall into the same traps as its predecessors. The perfect example can be found in engineering, where the sustained use of digital twins and innovative modeling solutions was supposed to accelerate the development of the new Boeing trainer jet for the U.S. Air Force and cut development costs. In the end, major delays and cost overruns. While reality can be augmented, it cannot be fully replaced, at least for now, and previous tools keep their pertinence.

Back to Table of Content  

Not much room for enthusiasm

M&A as the main enabler

Over the first half of the year, the 3D printing segment has been one of the bottom performers in our AI & Robotics investable universe, despite some positive catalysts. Earnings were rather positive, yet did not manage to fuel a positive cycle. This only left M&A as a potential tailwind, with Nano Dimension and 3D Systems both willing to acquire the unwilling Stratasys. M&A is usually the territory of segments failing to generate growth otherwise. Although we do not believe 3D Printing to be mature enough to fall into this category, we remain convinced that the segment has yet to find a large-scale application that would bring it into another dimension.

Dreams of reaching the stars vanish

We had strong hopes for one of these applications to be rocketry. U.S. “new space” player Relativity had developed a rocket entirely 3D-printed, paving the way for a larger use in an unforgiving sector. Despite the failure of the maiden launch in March, which is not that uncommon for a rocket, we remained hopeful for the following launch as well as for the following larger generations of rockets based on the technology. These hopes were crushed when the company announced it would retire the rocket without even attempting a second launch, and that its successor would be manufactured using more traditional methods. Another proof that the business case for large-scale 3D printing applications is far from obvious, if it can even exist at all.

Dental lagging, as expected

As we had expected, the dental market is facing tough market conditions, at least for the leader 3D Systems, with the business unit being down ~46% on 1Q23 results. Considering the ongoing tough macroeconomic conditions, there are few chances for the market to significantly restart in the short term, taking with it a major driver. All in all, we currently fail to detect meaningful opportunities in the sector, despite loving the technology’s potential.

 Back to Table of Content  

Year's Favorites 2023 - An update

Ambarella 

(AMBA US)

We chose Ambarella due to its central position when it comes to Machine Vision. Our conviction is that the sector will largely benefit from the rollout of advanced driving assistance systems and from autonomous vehicles, that the company’s technology and commercial positioning is differentiating, and that this potential will be unleashed at the incoming end of the inventory transition.

Although the company is delivering on the technology front, its end-market, as for juggernaut Mobileye, appears rather clogged on China's weakness, leading to disappointing guidance and hampering our investment case. On top of that, the company added that due to this weakness, the inventory correction lasted longer than expected. On the positive side, the company announced a design win with Continental, which was precisely what we had expected in our investment case. We expect such wins to grow in number and, as they translate positively into the company’s financials, lead to a deserved rerating of the stock.

AutoStore

(AUTO NO)

We chose AutoStore due to its favorable positioning in the warehouse automation segment, which we expected to benefit from the combination of inflation and macro headwinds incentivizing companies to optimize their supply chain structure. We also expected the company to benefit from the launch of a consumption-based sales model, which would expand its addressable market in the SME segment.

In a nutshell, our investment case held rather well. The company delivered a strong set of 1Q23 results, with orders largely beating estimates and margins benefiting from positive pricing adjustments and favorable dynamics in the aluminum market, which the company uses a lot. The stock is largely outperforming indices at the present time on a YTD basis, and we expect the consequences of the consumption-based sales model, which only launched in February, to be a major catalyst for the end of the year.

Baidu

(9888 HK)

We chose Baidu due to our conviction that it was a national champion in AI. We expected the company to benefit from the reopening in China, and to capitalize on its integrated AI technology. More important, we expected Autonomous Driving technology to be a major catalyst.

Our investment case fundamentally held true, but maybe did not deliver the expected returns yet. The company did indeed benefit from the reopening of the Chinese economy, with digital advertising coming in above expectations. The company also strongly benefited from the AI hype, having announced a homegrown competitor to ChatGPT; however, despite the technology delivering, the actual product launch was not as smooth as could have been expected, leading to a classic “buy the rumor, sell the news” movement. Finally, the catalyst from autonomous driving remains an 2H perspective, but recent figures communicated by the company show very positive momentum in this regard. However, the company appears more than ever dependent on Nvidia’s chips, which could become hard to procure due to U.S. sanctions, and pushing investors to the sideline.

Back to Table of Content  

Catalysts

• Application maturation. The first iterations of advanced AI already had a major impact on users despite their unpolished status. Refinement and better integration into existing processes will enable a massive boost in their adoption, especially with the rapid progress in generative AI.

• Search for efficiency. Periods of tension and/or crisis are the most transformational ones, while periods of economic downturn trigger a search for higher efficiency. AI can match both criteria.

• Better accessibility. AI and automation used to require a high level of technical expertise to be deployed. The emergence of low/no-code platforms will change this fact and enable a smoother rollout. At the same time, the technology's democratization will fuel a higher user familiarity and acceptance of the tools. 

Risks

• Polemics and fear. The impact of advanced AI applications will be transformational, hence their rollout will generate friction and resistance against change. Major controversies are bound to happen and may generate some resistance, dragging on the theme.

• Restriction on data. Data is crucial to train AI models. However, people and regulators tend to restrict the use of personal data, which could lead to models not aligning with reality anymore.

• U.S. sanctions over China. While we do not believe such sanctions could trigger the invasion of Taiwan, they would hamper the development of the Chinese ecosystem in the short term while cutting U.S. players from a major source of revenue.