Search ST Engineering

ST Engineering
ST Engineering

2023 was anything but normal, especially for the startup world. In a report by PitchBook on Annual US Venture Capital (VC) Valuations, VC valuations in 2023 continued to slide from their peaks in 2021 and early 2022, with median early-stage (down ~17% YoY) and late-stage valuations (down ~16% YoY) fell to their three- and five-year lows respectively. Only pre-seed and seed deals bucking the trends (up 33% in the past 2 years). Not only did 2023 see a dismally low number of IPOs, but the median valuation of startups that managed to make a public debut also declined to the lowest level in a decade. Although M&A activity dropped too, acquisition valuations increased year-over-year.

While the IPO market remained in a slump, this was buoyed by extraordinary hype over artificial intelligence (AI). Meanwhile, conflicts have started in 2023 and are continuing into 2024. The decoupling of U.S.-China technology ecosystems continues to accelerate with the recent investigation by a U.S. Congressional Panel concluding that 5 U.S. blue-chip VC firms (GGV Capital, GSR Ventures, Qualcomm Ventures, Sequoia Capital and Walden International) have invested more than US$3b in Chinese AI and semiconductor industries and calling for more restrictions on the U.S. financial industry’s ties to China.

This year, ST Engineering Ventures (STEV) looked into its crystal ball to guess what 2024 might hold. Against the backdrop of 2023, the STEV team believes that 2024 is shaping up to be an exciting year for technology, albeit with certain bottlenecks.

AI & Robotics Trends

While 2023 was the year of introducing AI into the consciousness of the masses, 2024 will be the year of data and model specialisations. Companies like OpenAI will require more and more data (including video streams from Youtube and TikTok) to train their foundational models, but other companies such as Mistral are focusing on expert models where curated data is important. This means that the future will be one where AI companies advocate for the democratisation of data, while policymakers push back on the grounds of privacy. With China having tighter control over their regulations, we might see Chinese large language models (LLMs) catching up to the Western world as they could have easier access to data. Also, while Western AI startups continue to focus on LLM development, Chinese Big Tech companies and startups have taken on a platform approach, where they have and are designing platforms comprising AI Agents that can work across different LLM models — both Western and Chinese.

What is sure, however, is that training costs will continue to drop: a combination of decreasing hardware costs, more efficient software, and cloud-based compute resources for AI training (China, the UK) means that smaller companies will also be able to train their own AI models. With the democratisation of data, individuals could even create a digital clone of themselves! In China, to overcome the restrictions on the import of graphics processing units from the West, we are observing the Chinese government centralising AI and energy compute infrastructure that will allow AI training as a service to their beleaguered AI startups, allowing them to stay in the game.

More realistically, we see a trend of companies finetuning foundational models with their curated data for specific applications. These companies will open-source their models to get traction quickly but also collect and train on user data in that process. The most successful ones will have curated data as their defensible moat, which can then be used to build closed-source models for monetisation.

The key to being successful, then, is in the deployment method — we predict a rise of AI agents integrated into the software we currently use, and in new interfaces that will replace existing ones:

  • While developers have had Github Copilot to aid them in programming for a few years now, there will be agents that can simply generate code instead of only aiding developers (e.g., dev for front-end, Poolside for general software development, Diffblue for Java unit test writing).
  • Elsewhere, search engines will strive to stay relevant by incorporating Generative AI but have a risk of getting replaced by AI agents that can provide more accurate answers for specific topics.
  • Creatives will find ways to reinvent themselves in the face of generative AI producing realistic photos, posters, banners, and now even videos. This could be in the form of rejecting AI by offering extreme bespoke services touting a genuine human touch, or giving in to the AI trend and becoming experts at AI prompts to deliver design work with greater productivity.
  • AI agents will be brought into the physical world to ensure greater frequency of use — AI companions such as the Rabbit R1, Samsung’s Ring, Rewind Pendant, and Humane’s AI Pin — but these will be exercises in understanding how to integrate AI into everyday devices, and they will eventually inform AI integrations such as Samsung’s Galaxy AI.
  • Robotics is another area VC is pouring funding into, particularly in the areas of Humanoids and Embodied AI. This is where beyond selected US startups, e.g. Tesla’s Optimus, Boston Dynamics and Figure AI, Chinese tech companies may have an edge due to their end-to-end manufacturing capabilities and less restrictive trade unions that may protest automation. Several Chinese VCs are of the view that AI will be able to display 80% of human characteristics by 2026, and are working towards a timeline of 2028 to develop a humanoid that will give AI its physical form and allow AI to enter our physical world.

AI Bottlenecks & Competition

In today’s world, progress in AI has been determined by the availability of talent and compute. Consequently, AI powerhouses have emerged largely in the U.S. and China, where there is access to a strong talent pool and capital to purchase compute resources. With data availability taking centre stage and the cost of training dropping, we expect that bottlenecks in AI progress will be in the availability of land for critical infrastructure, access to cheap energy, and availability of data. All of these are determined by geography, so we expect countries to pit their policies against each other in a bid to woo and develop champions in AI.

Diving into the details, Europe, which has a fragmented market, will continue to develop small, specialised models such as Mistral, Unlikely AI, and Poro given the limited capital and data per country. Meanwhile, we expect Southeast Asia, with its large population, increasing internet penetration and less robust data privacy laws, will become the next battleground for AI giants. It would be important to pay attention to Chinese companies as they are starting to look outwards to other countries, and Singapore is a key landing ground for them to go international or globalise.

On the innovation front, however, we are seeing increased focus by investors on defence tech. This could be the year where innovation swings back to the defence sector, after years of commercial technology superiority. More specifically, the defence-tech industry will have technologies focused on low cost, quick development, and easy manufacturability, as learnt from the ongoing conflicts. Some technologies we would look out for include counter-drone systems, navigation in GPS-denied environments, and decision-making software powered by generative AI.

Financial Markets

Looking towards financial markets, we expect investors to continue to be cautious and rational, focusing on fundamentals and profitability. Capital-intensive business models the likes of Uber and Airbnb which relied on network effects to fuel a gig economy will have trouble getting funding, while deep-tech companies must have a convincing business model beyond having great tech.

On the back of a jubilant funding ecosystem in 2021, we expect a continued trend of down rounds and failed startups. We are cautiously optimistic as interest rates may start going down with inflation starting to get reined in. Should this continue, we expect a recovery of the IPO market, especially if tech companies such as Databricks and Stripe have successful listings this year. This would help turn the tide on down rounds as investors become more willing to pay premium valuations, but would not change their cautiousness.

On the Web 3.0 front, we are thinking that there will be both a recovery of cryptocurrencies to an all-time high and another crypto-bank or exchange going bust. Interestingly, Bitcoin has already recovered to its highest since late-2021. The role of cryptocurrency in funding conflicts is undeniable (Israel — Hamas, Russia — Ukraine), and the ongoing conflicts will only increase scrutiny on cryptocurrencies. An optimistic view of cryptocurrencies coupled with a possibility of kneejerk reactions to cryptocurrency regulations might lead easily to yet another crypto winter.

In conclusion, as we embark on 2024, the reverberations of last year’s events continue to ripple through societies, economies, and ecosystems worldwide. Whether catalysing profound shifts in technology, e.g. AI and robotics, or reaffirming existing geopolitical trajectories, the repercussions of last year’s happenings serve as poignant reminders of the interconnectedness of our global community and the enduring impact of our collective choices. As we navigate the uncertainties of the future, we need to heed the lessons learned and strive for resilience, empathy, and cooperation in the face of the challenges ahead.