How Giant Will Invest Behind the AI Wave
The new dawn
The epoch of Artificial Intelligence (AI) is upon us, heralding a new era of possibility. Like many powerful and novel technologies, this one looks and feels like magic. Sora, the new product launched by OpenAI this month, can create full Pixar-level quality videos from text. The company’s commercial trajectory is equally impressive, with its annual revenues surpassing $2bn annually just a few months after the launch of ChatGPT. That makes it one of tech’s fastest-growing platforms in history. Through exponential gains in data and computing power, Large Language Models (LLMs) will reshape industries, societies, and human capabilities, catalysing a Cambrian explosion of innovation. When my co-founder Tommy interviewed the co-founder of DeepMind, Mustafa Suleyman, for Giant Ideas, Mustafa called this upcoming era one of “radical abundance.”
AI will also change investment strategies across the globe. At Giant, while we invest in the secular, evergreen trends of climate, health, and inclusive capitalism, our companies inevitably leverage the emerging technologies of the day. Typically, they do so by harnessing these technologies to solve a specific problem within one of our three verticals. In this article, I will outline how these recent advances in AI feed into Giant’s strategy and discuss how we plan to invest behind these waves of change.
Source: Sora
OpenAI’s iPhone moment
OpenAI’s systems are based on deep learning and reinforcement learning techniques, and their AI models are typically large-scale neural networks, such as the GPT (Generative Pre-trained Transformer) series. They are often called Large Language Learning models (LLMs) because they have been trained on the vast swathes of language data available online. The step-change in OpenAI's AI models stems from its newly found scale.
OpenAI's models are among the largest AI models ever created. For example, GPT-3 has 175 billion parameters, allowing it to generate text with high coherence and contextuality. The platform's scale underpins its generalisation capabilities, allowing it to perform well on tasks such as writing, translation, or answering questions - all without task-specific training. OpenAI has made its models accessible to developers and researchers through APIs, enabling them to integrate state-of-the-art AI capabilities into their applications easily. The generalisability and accessibility of OpenAI’s platform have opened up the technology for many industries and applications, leading some to call it AI’s “iPhone” moment.
Beyond OpenAI
At Giant, we’ve been asking ourselves which areas and tasks the horizontal platforms, such as Google’s DeepMind and OpenAI, will not be able to go after. Where will these behemoths’ capabilities asymptote, and where will they continue to compound? And where can startups get distribution or develop some unfair advantage?
Some investors are excited about generative applications, such as Sora, which facilitate creative content production. The general consensus of public market investors is that AI returns will accrue to the underlying hardware (chips/NVIDIA) and plumbing (the LLM platforms/Meta’s LLaMA), similar to the first Telecom boom. Indeed, this year’s stock market gains have been driven almost entirely by a few companies emblematic of this narrative. It’s not clear, however, how the LLM platforms, such as OpenAI, have sustainable moats unless emergent properties approximating Artificial General Intelligence coalesce from them.
Source: Sandi Besen
If the main moat is scale, then any company with sufficient capital and the requisite talent can train their models on the publicly available data on the Internet and create a similar end-product. Sam Altman, CEO of OpenAI, himself has suggested that there are diminishing returns to the models' size and the amount of data and computing power involved. In a recent interview, Altman said, “We're at the end of the era where it's going to be these, like, giant, giant models. We'll make them better in other ways.” It’s unclear where Altman thinks the moat will come from, but perhaps it will be the quality of the algorithms, talent, developer ecosystem, self-built novel applications, or an emergent artificial general intelligence.
However, without distinct moats, these companies may raise lots of capital and gain traction but fail to achieve pricing power and sustainable unit economics, particularly given the cost of training these models. This would be reminiscent of the “quick commerce” boom and bust during COVID-19. Building a sustainable competitive advantage is hard when capital is the weapon of choice. A16Z famously quipped that “the battle between every startup and incumbent comes down to whether the startup gets distribution before the incumbent gets innovation.” In this case, the incumbents, Microsoft, Amazon, Facebook, Google, OpenAI, and NVIDIA, are leading the innovation, have the computing power, and are pioneers. It’s tough to take those incumbents head-on, and they will likely dominate the lower layers of the AI infrastructure stack.
The areas we are most excited about
So, how do these developments in AI and the extraordinary technological advancements affect our strategy and decisions at Giant? At a helicopter level, we are excited by AI’s potential to help reinvent the process of scientific discovery within climate, as well as improve industrial processes and efficiency, automate repetitive and bureaucratic processes within healthcare, augment humans in their workflows, and bring personalised experiences to the masses across our three verticals.
Discovery: AI applied to biology has already led to the discovery and development of novel therapeutics within pharmaceutical healthcare, such as Recursion. Emerging platforms could radically transform the discovery and design processes for the critical materials needed for the transition - for example, the end-to-end design of novel metals and molecular design for applications ranging from carbon capture and storage to semiconductors.
Automation & Augmentation: AI is automating clerical tasks across industries. Early pioneers, such as Ada and UIPath, are followed by a legion of others, which will automate everything from scheduling healthcare appointments to financial auditing. Automation is the low-hanging fruit. Augmentation, rather than automation, occurs when a human uses AI to extend their capability but ultimately remains involved, so it’s not fully automated. For instance, developers and coders working hand in hand with AI co-pilots are seeing widespread adoption. Or consumers taking advice on their financial needs, a type of “money on autopilot.”
Personalisation: Whether in education, finance, or health, AI radically reduces the cost of personalisation. Within education, this will lead to personalised support at a much lower cost than tutors and a move away from today's current “one-size-fits-most” education system. Similarly, we expect to see huge advances in the personalisation of medicine and wellness, which will integrate with and learn from existing health data to act proactively and support health outcomes. This will broaden access to high-quality physical and mental healthcare, bringing it to the masses and democratising what has historically been a luxury.
At Giant, we’re more interested in startups applying AI with either initial data or a distribution advantage or those who can deploy an alternative machine learning approach that is more appropriate and elegant for that specific use case. Inversely, if the data required for the model is in the public domain, if there is no clear distribution advantage, and if large LLMs best power the product, then the horizontal platforms, such as OpenAI, will capture the value.
What we mean by a data advantage is access to valuable training data that is not easily accessible or publicly available. An example might be the siloed data of a hospital group or NHS trust, which, if unlocked, could provide a start-up with unique data for training their models. A distribution advantage arises when a company has a way of getting its product into the hands of users in a unique, cost-effective manner not available to others. OpenAI, for instance, currently has a unique distribution advantage because they have partnered with Microsoft to power their existing suite, so as a result, they are getting access to all of Microsoft's end users.
How we’ve invested so far
If we think about the birth of the internet as an analogous parable, it was the horizontal infrastructure companies that led the early value-creation, such as Intel & Cisco, but over time, value accrued to the application layer, and those companies that integrated the internet’s qualities with other emergent technologies. The likes of Apple and Google built upon and integrated the defining infrastructure technologies of the era, such as routers and the Internet, to create novel products that altered society. At Giant, we are mostly focused on the application layer of AI. AI applied to specific problems, complemented by unique data sets, with sustainable distribution advantage, is how entrepreneurs will build defensible value creation.
We’ve made multiple investments from our Seed and Early Growth funds, which bring our perspective on defensibility to life. Each also falls within our buckets of discovery, automation + augmentation, and personalisation.
For example, using AI for automation and augmentation, we invested in Matta, a company that builds AI co-pilots for advanced manufacturing that learn from the best operators in the factory, helping them do what they do best and automating the rest. We also invested in Byterat and Invert, companies that collect and aggregate data and optimise processes with AI for the Electric Battery and Biomanufacturing industries, respectively. They harness unique datasets and domain-specific expertise to create a valuable product for the end user, their “co-pilots” improving the end process and, in the case of Invert, improving yield and lowering the cost for biomanufacturing (applied to both industrial bio and biopharma). Both companies are developing a data advantage by collecting and ingesting unique, hard-to-collect datasets from their end users, namely battery labs and biopharma companies.
AI can also be used for essential personalisation of services. Our portfolio company RxDiet uses AI to personalise their service, partnering with insurance companies in the US to deliver the most affordable, healthy, and culturally appropriate food to insured members via its unique machine-learning platform. We’ve also backed a NYU spin-out building a foundational model for cancer - starting with breast cancer - which uses AI driven diagnostics for enhanced patient care. Sugar - a personalised super app for longevity - is being built by the co-founders of Gorillas, one of the few companies to exit for over a billion in Europe.
In the theme of discovery, we have invested in a company leveraging AI and physics-based methods to accelerate the discovery and design of molecular glues, and we are excited about companies using novel machine-learning approaches that combine generative AI and simulation to design novel molecular materials that could tackle climate change.
In each case, the companies have a distinct data or distribution advantage, often via unique and undisclosed partnerships, or are applying a different machine learning approach that is more appropriate for the problem set.
How AI impacts global equality and opportunity
Finally, as we contemplate the implications of AI on our investment themes, we must recognise its potential impact on inclusive capitalism at a societal level. Some anticipate AI will create new jobs and alleviate some of the drudgery in blue-collar work, while others express concerns about its disruptive effect on employment. This age-old dialectic of technological innovation has raged since the Luddites stormed the flour mills. The truth is, no one knows if this time is different, but it is clear that AI will enable entrepreneurs to build huge companies with far smaller workforces. Large swathes of coding, marketing, and sales functions can and will be automated with AI. Where full automation isn’t possible, people will be augmented by AI, delivering productivity far above what was previously possible. I believe AI will increase global productivity but not generate a distributed form of wealth creation.
To prevent a situation of further consolidation of power and wealth, it’s imperative that we build and support open-source LLM models and explainability, ensuring transparency and mitigating the risks of an opaque economy running on probabilistic mechanisms. Explainability is the capacity to express why an AI system reached a particular decision, recommendation, or prediction. Open-source models, available to all, and advancements in explainability serve as bulwarks against the opacity of probabilistic mechanisms. Society should decide which parts of our existence can be directed by probabilistic models and which elements must demand explainability. For instance, why someone is denied a bank loan should be explainable, and in the future, it would be immensely helpful to understand how AI discovers novel drugs. So, while we will mostly focus on the application layer, we monitor opportunities in open-source and explainability lower down in the technology stack.
In conclusion, the trajectory of AI represents a profound intersection of technological innovation and societal evolution. As catalytic investors, we search for the forefront of this transformative wave, guided by a commitment to harnessing the power of AI for the betterment of humanity, doing what we can to ensure a trajectory of AI aligns with societal values and aspirations. Through strategic investments in companies that pioneer domain-specific applications within our themes and champion openness and transparency, we endeavour to navigate the complexities of this landscape with diligence and foresight.
Written by Cameron McLain, Co-Founder and Managing Partner at Giant Ventures