We should expect more AI TV commercials, like this one, a clip from Toys R Us. Source: Toys R Us, Sora
This year was all about whether there is a bubble in artificial intelligence. OpenAI, ChatGPT’s developer, is the fastest company ever to reach an annualised revenue of US$13 billion, as well as the fastest to reach user numbers of 800 million. Headlines like the one below, from the Wall Street Journal, evidence the almost unprecedented level of value growth that has taken place in the three years since the first large language model was released commercially.
Source: WSJ
The bifurcation in market returns generated by the surge in AI interest has been one of the major themes of the year
But where did we start the year? OpenAI’s valuation at the beginning of 2025 was $157 billion and it was generating annualised revenues of $5.5 billion by December 2024 (CNBC). While impressive, these numbers are dwarfed by the latest reports of the company seeking $100 billion in funding at a valuation of $830 billion. The enormous amount of funding required for OpenAI to continue growing is reflected in the amount of cash the hyperscalers (Microsoft, Google, Amazon and Oracle) and Meta are spending on building out massive datacentres to handle these compute-intensive AI workloads. Capex estimates for these companies have moved higher over the course of the year as the applications and use cases for AI have become more material.
What we are seeing now are the first commercial applications of artificial intelligence and the possibilities enabled by this burgeoning technology. It’s these applications that sketch out the addressable market for AI and underpin continued investment in the tools driving its growth.
Entire ads can now be created with artificial intelligence
Source: Toys R Us, Sora
Every industry is affected. For advertising, AI enables designers and artists to create ads with as little as a prompt. Meta is leveraging AI to enhance advertiser campaigns and improve ad targeting. For software, AI bends down the cost curve for app development and democratises code writing via vibe coding tools like Lovable and Replit turning anyone into a developer.
AI goes beyond the world of text, image and server farms. There are physical embodiments of AI – autonomous vehicles, robots and drones that are already fully functional. Waymo, which is Google’s autonomous taxi service, is operational in seven US cities 24/7, while in San Francisco it has a 22% market share of all taxi rides.
All of this change, disruption and value creation – but particularly the large language models on which artificial intelligence is based – has been made possible by the singular, spectacular growth in capacity from semiconductors, which has accelerated literally month by month for 71 years since the first prototype of the modern microchip was invented.
This is not a drill.
No, it isn’t. But it is correct to ask whether all this AI value is correctly priced.
It is possible that there is a component of overvaluation in AI companies, but this is not a bubble in the generally-accepted-accounting-principles sense. We often talk about real bubbles being a dramatic mismatch between finance and usage of the utility in question. The question is around how much the usage and market capitalisation diverge at present. For example, the global financial crisis was a US financing boom for a property market which was not underpinned by the quality of loans that drove it.
2025 marked AI’s transition from speculation to a structural industrial reality. Despite volatility from DeepSeek’s apparent low cost development shock and “Liberation Day” tariffs, AI-associated companies proved resilient. While concerns remain, the sheer scale of adoption across industries suggests that the current value growth is underpinned by genuine utility rather than just financial speculation.
2025: year in review
2025 began with the new Trump administration and news from Chinese start up DeepSeek releasing its R1 model purportedly developed at a cost of ~US$300,000, undermining the prevailing belief that high end models cost hundreds of millions to develop. The share prices of AI companies fell in what was described as an AI Sputnik moment, but recovered quickly once details surrounding its true cost came to light.
A policy driven market
On the policy side, the February announcement of tariffs on Mexico, Canada and China sparked concerns that Trump’s protectionism was more than a bluff. These worries came to a head on 2 April with the announcement of sweeping “Liberation Day” tariffs on some of the US’s closest trading partners, kicking off a trade war with China.
Poor macro compounded the effect. CPI figures in February indicated that price levels were still running too hot for the Federal Reserve to cut rates and the US labour market was holding strong against rate rises. Coupled with inflationary tariffs, the market pushed out rate cuts which hurt “higher duration” AI stocks and many others besides.
As these data points roiled global markets, Loftus Peak reduced positions in companies exposed to cyclical recovery to take advantage of depressed prices in some of our most favoured names that we had previously sold due to high valuations including AI-related companies. These included Nvidia, Crowdstrike, Arista Networks and Samsara – most of which contributed strongly to performance in the year.
Markets snapped back as initial tariff announcements were “walked back” by the Trump administration and the tone from Fed officials turned dovish. The first rate cut in September helped to lift markets to new highs.
Streamers outperformed again in 2025
In addition to exposure to secular (non-cyclical) growth trends, a major consideration during the tumult of April was how exposed our holdings were to the Trump administration. As a result, we increased our exposure to relatively insulated streaming holdings like Netflix and Roku, which were both major contributors for the year.
Netflix flexed its pricing power, continued to grow subscribers and developed its advertising technology to sustain its mid-teens revenue growth rate while expanding margins – a recipe for strong stock price performance. The streamer also pushed harder into the last bastion of linear television, live sports, leveraging its distribution to bring NFL, boxing and WWE to Netflix subscribers globally. The biggest news of the year came in the past few weeks with the announcement of Netflix’s acquisition of Warner Bros Discovery for just shy of US$83 billion. While there is a legitimate view that the Warner Bros Discovery asset, which includes HBO, is worth more inside of Netflix, the Netflix story just became a lot more complex, with greater execution risk, exposure to legacy revenue streams and substantial regulatory hurdles that make us cautious.
China’s technological dominance
The rise of China was one of the themes that Loftus Peak was founded on, given its profound disruptive potential. Loftus Peak had steered clear of these companies over the past few years on concerns surrounding the regulatory and governance landscape, but in 2025 initiated positions in a handful of Chinese companies exposed to secular themes. Chinese battery maker CATL was one of Loftus Peak’s top contributors this year. CATL’s best-in-class technology is used by European, Asian and US automakers and the company is expanding into the growing energy storage system market to be leveraged in datacentres in combination with renewable energy sources to underwrite energy security. China’s performance in critical hard technologies is not just a commercial endeavour, it’s a matter of national security for the Chinese government.
OpenAI vs. Google
The dynamic around the AI trade started to shift in the back half of the year. While the previous couple of years had been characterised as a tide that lifts all AI-exposed boats, the release of Gemini 3 and the performance of Google’s TPU (designed in partnership with Broadcom) turbocharged stock prices for those companies exposed to the Google AI ecosystem. The resolution of an antitrust case against Google also boosted sentiment. Google alone added US$1.6 trillion in value since 30 June.
Recent news reporting of OpenAI’s US$830 billion valuation and performance of GPT5.2 has clawed back some of the underperformance against Google but it’s still anyone’s race.
Software: disruptor or disrupted?
We often talk about the beneficiaries of AI in the context of the portfolios we manage. In the case of software there are legitimate concerns that AI disrupts the status quo for the Software as a Service (SaaS) business model. Part of the SaaS competitive advantage is the ability to add features into the platform and charge for them by moving users to higher pricing tiers. If the cost and skill curve for software bends downward due to AI, how does traditional SaaS continue to be a value proposition for customers? Compounding this is the fear that as AI does more work, there will be fewer employees required and therefore fewer software subscriptions. This is yet to come through in the earnings of software companies, but it is a concern.
We believe that the performance of coding tools isn’t at the stage where enterprises would feel comfortable building and running business critical software on them. Yet. But this will likely change and there will be winners and losers. As we’ve seen over the past few years the performance of AI models has increased exponentially and with it a new range of abilities unlocked. It’s these concerns that hampered some of the software companies we held in 2025, including ServiceNow, Salesforce and Adobe.
It’s not all doom and gloom for software companies. They have deep roots in enterprise workflows making them hard to turn off. Additionally, software companies are highly capable at leveraging AI for their own purposes. Loftus Peak has been highly selective with our software exposures through the year due to these concerns and sought companies with defensive market positions, large addressable markets (including cybersecurity and operational technology), and attractive valuations.
Looking ahead
Loftus Peak’s strategy openly acknowledges that the leaders of yesterday aren’t necessarily the leaders of tomorrow. Consumer habits and business models are constantly changing, often underpinned by technological change. Artificial intelligence is no different and we are looking forward to 2026 and the potential for new business models and beneficiaries to emerge. Meanwhile, outside of artificial intelligence, there is significant disruption occurring in other industries such as automotives, energy, streaming, software and life sciences, to name a few.
The team at Loftus Peak will continue to look ahead, understand and value the impact of disruption on all industries and risk-adjust our portfolios to best expose clients to those opportunities.
Thank you for your continued support and we wish you all a safe, happy holiday season and wonderful start to the new year.
Share this Post