AI is shifting from answering questions to taking action — executing, transacting, and coordinating on its own. That is pushing investors past the hype towards what actually makes AI usable: infrastructure, security, and data control. Digital assets are on a similar path, moving past hype into real market infrastructure like stablecoins and token settlement. Here to unpack where the smart money is looking is Anthony Georgiades, Founder and General Partner at Innovating Capital. Anthony, welcome to the show.
Thanks for having me.
You have talked about AI moving from answering to acting. What does that shift actually require to work safely in terms of identity, authorisation, and accountability?
When you use a chatbot, there is no real risk profile tied to a wrong answer. But as we move toward AI agents that have the authority to send money, alter a database, or communicate with a customer — that changes the risk profile completely. Enterprises are spending a lot of time needing to understand and know which agent is acting, what it is authorised to do, what data it is using, and whether its actions can be audited or reversed. That trust and control layer may become as important as, if not more important than, the underlying model infrastructure itself.
As an investor, you are looking beneath the model layer. What infrastructure or capability gaps are you most focused on?
Foundation models are enormously important — but they are extremely capital intensive and becoming fiercely competitive. We are interested in what you would call the picks-and-shovels aspect that makes AI deployable across a variety of enterprise stacks. Data infrastructure, orchestration, evaluation, agent identity — including a ChatGPT agent being able to talk to an Anthropic-based agent while sharing the same memory pool. We particularly like businesses where AI owns a viable workflow instead of merely adding a specific type of feature.
Where is the real monetisation coming from in AI and who is positioned to capture that value?
Monetisation has moved beyond selling access to a model. The strongest companies will charge for an outcome, a completed workflow, or a measurable productivity gain. If a product can save an enterprise $10 million, pricing it on tokens consumed leaves a lot of value on the table. The Anthropic S-1 and imminent public listing will be very interesting — it is really the first point of true price discovery for LLMs generally. Usage is not the same thing as value capture. As inference gets cheaper, some model economics will dramatically compress. But cheaper intelligence will also expand the number of economically viable use cases. Value can migrate upward into applications and downward into specialised infrastructure.
You have described digital assets as becoming market infrastructure rather than disappearing. What does that look like in practice?
Stablecoins are increasingly better understood as payment and settlement infrastructure rather than just a crypto product. They allow dollars or dollar-like value to move globally around the clock on programmatic rails. They have been the backing enabling even FDIC-insured banks to offer real-time instant payments 24/7. The days of waiting to send a remittance between traditional banking hours is gone. And a lot of that is powered by stablecoin or stablecoin-like infrastructure at banks like JP Morgan and Capital One. They are beginning to look less like speculative crypto and more like an internet-native settlement layer.
What is your outlook on IPO activity in AI and crypto infrastructure?
I do not think it will be a broad risk-on market. I think it will be a selective reopening. Public investors will reward AI and digital infrastructure companies that can demonstrate durable growth, improving margins, recurring demand, and genuine strategic importance. The biggest point is that usage is not the same as value capture. And as inference continues to get cheaper, there will be a clear playing field to discern what that value capture actually is.
Anthony, thank you so much for joining us.
