From hype to harvest: investing in the AI boom

Executive summary

Artificial Intelligence (AI) is rapidly evolving from a research-driven frontier technology into a real economy disruptor. While enthusiasm is high, markets remain inefficient at pricing the complexity, cost, and global scope of this transformation. We view AI not as a monolithic trend but as a structurally rich, multi-dimensional investment opportunity that demands selectivity, a deep understanding of technological undercurrents, and a global lens. Our approach spans long-only and long/short strategies, enabling us to back structural winners and also short businesses exposed to displacement, commoditisation, or unsustainable fundamentals.

This paper outlines our views on the trajectory of AI as a general-purpose technology, where we see durable value creation and risks, and how we are positioned to capitalise on this paradigm shift.

AI: the next general-purpose technology

Artificial Intelligence (AI) has crossed a critical threshold.  No longer an experimental frontier, it now stands alongside electricity, computing, and the internet as a general-purpose technology. The impact is already measurable: AI tools are writing code (Microsoft CEO Satya Nadella announced that AI now writes between 20% and 30% of the company’s code¹), accelerating drug discovery (Nvidia-backed Recursion reduced target identification time from months to weeks²), and powering customer service across sectors (AI increases the speed of resolution in customer service by 57%³). The real breakthrough lies in large language models (LLMs) and multi-modal AI, which can ingest and generate text, images, audio, and video with human-level fluency transforming everything from legal analysis and insurance underwriting to marketing and design.

This technological leap is translating into widespread commercial deployment and capex growth. Global AI infrastructure spend could reach $800bn in direct AI equipment, and another $300bn in ancillary equipment/spend, leading to $1.1 trillion in total AI capex annually by 2027⁴.

Just as the internet redefined business models in the 2000s, we believe that AI is reshaping competitive dynamics across the economy.

The capex cycle is real, but monetization is uneven

We see accelerating AI infrastructure spend over the next decade. However, the market continues to underestimate the cost, time, and operational drag associated with this investment wave.

Microsoft, one of the most aggressive investors in AI infrastructure, underscores a broader challenge facing industry leaders as monetization pathways are still maturing. Despite its early-mover advantage and deep integration across cloud and productivity platforms, Microsoft has come under pressure as the earnings visibility around its AI investments remains uncertain, prompting downward revisions to future earnings. This disconnect between technological momentum and earnings traction is likely to persist across parts of the market, particularly for firms building at the infrastructure layer without immediate application-layer returns.

Open-source innovation is compressing margins faster than expected

A pivotal development is the emergence of DeepSeek, a Chinese foundation model developer offering GPT-4-class performance at a fraction of the cost:

• 94% lower training time
• 90% lower inference cost
• 30x cheaper pricing per million tokens than OpenAI

Most importantly, DeepSeek is open source, with commercial Application Programming Interface (API) access priced at a fraction of prevailing market rates. This radically lowers the barrier to entry for developers, enterprises, and startups seeking to integrate frontier AI without being locked into proprietary platforms.

What DeepSeek represents is not just an incremental improvement but a paradigm shift. First, it exposes how quickly the economics of scale in AI can change: the assumption that only the most capitalised Western firms can build top-tier models is being eroded. Second, it highlights the power of open-source ecosystems to drive global adoption, competition, and iteration-much like Linux, Android, and TensorFlow did in prior waves of computing. Finally, it underscores that technological advantage may be increasingly ephemeral, especially when innovation is rapidly disseminated, and infrastructure becomes commoditized. These cost and efficiency breakthroughs continue at a rapid pace from other Chinese tech companies as well. Since January 2025 further cost innovation from Baidu and Tencent has reduced inference costs by a further 60% in open-source models.

For investors, the implications are significant. Value in AI may accrue less to those who train the largest models, and more to those who control distribution, integrate capabilities deeply, or build differentiated applications on top of increasingly accessible model platforms.

Commercial use cases are scaling faster in consumer-facing verticals

Much of the mainstream narrative around AI investment focuses on foundational model developers and enterprise AI platforms. While these are important, we believe that some of the most immediate and economically scalable AI adoption is taking place in consumer-facing verticals, where feedback loops are faster, product cycles are shorter, and monetization paths are more direct.

In particular, we are observing concrete deployment and revenue generation in four high-impact domains:

1. Advertising and Marketing Automation

AI models are driving increasingly sophisticated content generation and audience targeting strategies. Tools that create dynamic advertising copy, localized imagery, and tailored video creatives at scale are already reducing customer acquisition costs for brands and agencies. Moreover, major platforms (e.g. Meta, Google) are embedding generative AI into ad-buying interfaces, enabling automated campaign optimization based on real-time performance data.

This translates into measurable productivity gains for advertisers and more efficient monetization for platforms-a rare case where AI boosts both operating margins and revenue simultaneously.

2. Ecommerce and Retail

In digital commerce, AI is being deployed to improve search relevance, recommendation engines, and customer support. Large retailers and marketplaces are integrating multi-modal AI to create voice- or image-based search experiences, real-time personal shopping assistants, and fully automated helpdesks.

This not only improves conversion rates and customer satisfaction but also reduces reliance on human agents, cutting costs in high-volume service environments. Importantly, these tools are already deployed at scale, making ecommerce one of the most mature AI verticals from an ROI standpoint.

3. Content Creation and Media

From graphic design and copywriting to music composition and video editing, AI tools are enabling creators to produce high-quality assets at a fraction of the time and cost. This trend is empowering independent creators while also being adopted by large studios, publishers, and platforms seeking scale.

Examples include:

• AI voice cloning for dubbing and localisation.
• Automated editing and animation for social content.
• AI-assisted scriptwriting and image generation.

As tools become more capable we expect a structural shift in media economics, where the marginal cost of content creation approaches zero and monetization becomes increasingly driven by distribution and IP ownership.

4. Gaming and Interactive Entertainment

The gaming industry is among the most enthusiastic early adopters of generative AI, with applications across:

• Procedural content generation (maps, missions, dialogue).
• Real-time character behaviour (non-playable characters driven by LLMs).
• Personalized gaming experiences shaped by player inputs.

The potential here is not just operational efficiency, but a creative expansion of the gaming experience, enabling studios to offer richer, more adaptive environments at lower production cost. Given gaming’s high user engagement and monetization rates, this may be the first vertical where AI materially increases lifetime value per user.

Key risks we monitor

While AI is a transformational trend, our investment discipline demands close attention to downside risks:

• Open-source disruption compressing software margins.
• Regulatory volatility around data use, algorithmic governance, and chip exports.
• Capex cycle overextension, leading to oversupply or investor disillusionment.
• Geopolitical fragmentation of AI ecosystems, creating divergent paths for US, China, and Europe.

The intelligence premium: capturing the real AI upside

The headlines are full of promises-but not all will be fulfilled. History reminds us that pioneering a technology doesn’t guarantee capturing its value.  Xerox invented many of the core ideas behind the personal computer in the 1970s-including the use of a screen, a mouse, and windows-but it was Apple and Microsoft that turned those inventions into world-changing businesses. Netscape built the first popular web browser, but it was companies like Google and Amazon that captured the internet’s economic value. Some firms will ride the AI wave to sustained outperformance-adapting business models, creating new markets, and embedding AI into their core value proposition. Others will sink under the weight of inflated expectations, unsustainable burn rates, or business models built on commoditized tools.

Our edge lies in foundational understanding of the technology and the interplay between innovation, capital intensity, and adoption pathways. In an environment where capital is chasing narratives, we concentrate on the underpinnings of long-term value creation: scalable capabilities, ecosystem control, and positioning within the technological stack. This depth of insight shapes how we allocate capital-not reactively, but with foresight. It is anchored in technical fluency and real-world execution experience.

The emergence of DeepSeek highlighted just how quickly assumptions can be overturned in AI. Just months earlier, upstream infrastructure stood out as the most compelling way to access the opportunity. But DeepSeek’s ability to compress costs reshaped what constitutes scarcity overnight. When technology moves this fast, advantage belongs to the agile. Our edge is the ability to reassess and reposition as value shifts.

One way we bring this insight into portfolios is through our high-conviction long-only strategy. It is built to capture AI’s structural impact by investing in companies integral to its deployment, acceleration, and monetisation. These are not always the most obvious names. Many of the most compelling investments lie not only in the firms building digital infrastructure, driving energy efficiency breakthroughs, or enabling AI-native applications, but also in the companies quietly using AI to transform their own economics by boosting productivity, enhancing margins, or unlocking new revenue streams. We combine this with a global perspective. Much of the world’s AI infrastructure and adoption is advancing outside the core of U.S. big tech, including in Asia, where we find both critical enablers and companies using AI to transform.

Alongside this, our global long/short equity strategy offers a differentiated and complementary way to access value across the full spectrum of the AI landscape. It is designed to capture the divergence between structural winners and losers. As capital floods into AI-linked names, many companies are rewarded more for their narrative than their fundamentals. These may include businesses whose products or services are becoming increasingly commoditised, eroding their pricing power and margin sustainability, or losing share to more technologically advanced competitors.

Both strategies reflect the same investment philosophy: the opportunity is real, but its rewards will not be evenly distributed. Success lies in knowing where the leverage is, and in staying ahead of how capital, compute, and competition evolve.

For further insights or to discuss our current AI positioning in more detail, please reach out to ir@fulcrumasset.com.

Fawaz Chaudhry, Partner, Head of Equities

Fawaz holds a Master’s in Electrical Engineering and Computer Science from the Massachusetts Institute of Technology (MIT), where his thesis focused on neural networks (a core component of Artificial Intelligence models) for complex real-world tasks. His work combined early machine learning architectures with robust engineering to solve a commercially relevant problem: designing a full AI system for offline recognition of unconstrained handwritten numerals on bank checks. This hands-on expertise in model design, error recovery, and deployment in messy data environments provides him with a rare combination of deep technical insight and market expertise.