In October 2024, OpenAI raised $6.6 billion, valuing the company at $157 billion. Fast forward six months, and OpenAI is in talks of raising $40 billion, at a valuation of $340 billion. This week, Anthropic announced a $3.5 billion round, valuing the company at $61.5 billion. In December 2024, Elon Musks Grok raised $6 billion series C, valuing the company at $75 billion less than a year from inception. Ilya Sutskever’s SSI is currently raising money at $20 billion valuation (pre-launch) and the list goes on.
As a result of their fast growth, both Anthropic and OpenAI became two of the most valuable private companies in a short period of time. But is it sustainable?
The launch of Deepseek’s R1 model on the 20th of January 2025, was dubbed by many as the ‘Sputnik moment’ in the AI race because it challenged the undisputed US dominance in the foundational model space up until that point. Deepseek, a Chinese company without (seemingly) access to the latest Nvidia chips, claimed have spent only $6M to train and fine tune a reasoning model that reached similar performance benchmarks as OpenAI’s o1 model. The costs later turned to likely be higher in reality, but the possibility of disrupting the dominance of US foundational models and the infrastructure they rely on (Nvidia) rattled the market and triggered a huge tech sell off and Nvidia’s loss of $650 billion in market cap in a single day. Since then other Chinese companies including Alibaba and Bytedance have launched a number of models challenging the US companies, most are open-source and free or much cheaper than their US equivalents.
Until now, the majority of the value in generative AI has been captured by the infrastructure layer: foundational models (LLMs), the cloud providers and the hardware infrastructure/ data centres. However now, the AI landscape is undergoing a significant shift, prompting a crucial question: Where will the future value of AI accrue?
The thought is that going forward the value is about to shift from the infrastructure to the application layer and AI agents. There’s precedent to this belief: during the first dot com bubble, browsers went public with huge fanfare (see Netscape), but very quickly became commoditised. It turns out the value of the Internet was not in the tools used to access web pages, but rather in the services used (and paid) by businesses and consumers. Will that also be the case in AI?
The Wave of AI Commoditisation
The underlying trends suggests a move towards the commoditisation of foundation models. This means that the core AI models themselves are becoming increasingly accessible and affordable, levelling the playing field for businesses of all sizes. I’m not sure how quickly this will happen, if at all, but consider the following arguments supporting this trend:
- Decreasing Costs: Training AI models is becoming cheaper and more efficient. DeepSeek’s R1 model, for example, reportedly matches the performance of OpenAI’s flagship systems at a fraction of the cost.
- Increased Accessibility/ the rise of open source models: Many foundational models are now open source or available through cost-effective APIs, making AI integration easier for startups and smaller enterprises. Case in point: the CEO of Figure, the humanoid robotics company publ
- Shift in Focus: The focus is shifting from the capabilities of the underlying models to how effectively they are integrated into broader systems.
Every challenge can be an opportunity
The commoditisation of LLMs could mean opportunity for startups: going forward, the bar for building AI-native products is much lower, enabling smaller companies to integrate AI and fostering innovation. As the value shifts from the models themselves to how they are implemented, the competitive advantage now lies in building reliable, autonomous AI agents and embedding AI models into intelligent systems that can handle real-world tasks.
Companies that build domain-specific solutions or layer proprietary data onto commoditised models will have an edge, spurring new applications across industries and ultimately leading to market expansion through cost reduction. This technological shift may require established players to adapt to avoid being outpaced by more agile competitors. The emphasis is shifting towards holistic systems that harness a model’s outputs to automate tasks and drive real-time decision-making.
One of the challenges of investing in AI is the wave of commoditisation. What seems novel at the time of making an investment decision can be fully commoditised a year later (or sooner in some cases). That made it difficult for some VCs to place bets – that’s not about to change. Below are three examples of potential winners.
Vertical AI
Even with the commoditisation of foundational models, vertical AI is a huge opportunity because it addresses complex, industry-specific demands with tailored functionalities, deep integrations, and specialised workflows.
While foundational AI players focus on technology, vertical AI applies those tools to meet industry requirements. The competitive edge lies in product and distribution, building defensible positions through industry alignment and effective go-to-market strategies. Vertical AI markets tend to follow a “winner-takes-most” dynamic, emphasising the importance of scaling rapidly and capturing market share early. Instead of focusing on easily commoditised applications, vertical AI should tackle end-to-end workflows and integrate tightly with existing systems. Companies can differentiate themselves by targeting overlooked categories with complex requirements and nuanced needs.
Examples from the Remagine Ventures portfolio: we invested in Trulux, develops AI-driven authentication and traceability solutions designed for the luxury retail market as well as Storywise, an AI platform transforming the antiquated book publishing industry.
Intelligent Systems and AI Agents
AI agents are well-positioned for success in a landscape of commoditizsing foundation models because they shift the focus from the models themselves to the intelligent systems leveraging them. This evolution moves from AI as a “co-pilot” to an “auto-pilot,” where AI executes tasks autonomously. AI agents offer action-based value by innovating and packaging tasks into services, particularly benefiting SMBs through access to affordable, sophisticated services via AI agent marketplaces.
These marketplaces, which turn software into services, are expected to dominate due to network effects and the ability to provide pre-vetted, on-demand AI services. Success will be further driven by building trust and explainability in AI agents, enabling them to make intelligent decisions, and creating AI-first organisations capable of wide-ranging actions. I recommend reading Gigi Levy’s post on the NFX blog.
In addition, the infrastructure to build these agents is getting increasingly more robust in a quick time frame. Madrona’s Jon Turow has a great primer on this.
Examples from the Remagine Ventures portfolio: we invested in Aigency (stealth) which is developing agent to agent communication for the transactional web as well as Keewano, an AI co-pilot for mobile analytics, starting with gaming and consumer apps.
AI Apps
AI apps are a good opportunity because AI-native products are scaling faster and engaging users more deeply. The introduction of more advanced models and capabilities have led to a surge in usage. There is increasing fragmentation across different AI model providers, allowing for specialization and offering creators more tailored options. Text-to-web app platforms are accessible to both technical and non-technical users, changing who can create with AI, not just who can use it. Many apps with lower usage have proven to be effective at monetisation through mobile subscriptions.
A16Z just published their third instalment of the top 100 AI consumer apps – take a look at the landscape below.
In summary
It’s too early to declare the commoditisation of foundation models, and it’s safe to say, that even though these companies might be overvalued, companies like OpenAI, Anthropic, Grok and other LLMs have seen some of the fastest waves of consumer adoption we’ve seen in recent times.
The AI battle between the US and China will continue to play a role (at least during Trump’s presidency) and may actually lead to a quicker adoption of open source models. What is certain is that for early stage investors, there’s a big opportunity now, and that the next couple of years will continue to be a rollercoaster ride in terms of new improved models, and of course, commoditisation.
