AI markets have developed rapidly since their nascency in 2022/2023. However, it now appears certain, at least in the short term, that we have clarity on who the key players and likely winners will be. This does not mean that other competitors won’t emerge over time, or that these current winners won’t be acquired by even larger incumbents or simply fail. After all, Yahoo did not win search, and MySpace did not win social networking. What is clear is that the key players at the foundational layer are now established, alongside some breakout companies in specific verticals.
At the foundational model layer, the narrative to date has mainly centered on scaling laws. These scaling laws describe the predictable power-law relationship between a model’s performance (often measured by test loss) and the resources used to train it, which are primarily model size, dataset size, and compute. These laws show that performance improves as these three factors are increased together. In practice, these models are driven by scale; scale of capital and scale of data. To win in this market, vast amounts of capital are required. For example, OpenAI recently projected its cash burn will total $115 billion through 2029, with over $8 billion expected in 2025 alone, up roughly $1.5 billion from what was previously expected (more than just a rounding error!). This unsustainable financial demands of staying competitive. Due to these scaling laws, clear winners have emerged in this space: Anthropic, Google, Meta (via Llama), Microsoft, Mistral, OpenAI, and xAI all lead on various benchmarks. Chinese companies such as DeepSeek also pose legitimate competitive threats through their open-source models. Given the barriers posed by these scaling requirements, it is unlikely that we will see new breakout competitors at the foundational model layer anytime soon.
What remains less clear at this stage is how these foundational models will permeate the application layer. One of the earliest and most evident applications has been in coding, where AI tools are transforming developer productivity. The early competitors in this market are well-defined: Cursor, Windsurf, Claude for coding workflows, and Microsoft’s GitHub Copilot. However, even this vertical raises questions about long-term sustainability. Cursor, for instance, raised $900 million in June 2025 at a $9.9 billion valuation, just months after a $105 million Series B round in January 2025 at a lower valuation. The company’s growth has been explosive, with revenue reportedly doubling every two months. Yet, if incumbents like Anthropic (via Claude), Microsoft (with GitHub Copilot), and emerging players like Windsurf continue to leverage not only their capital moats but also their vast infrastructure and integrations, the long-term prospects for pure-play startups remain uncertain. If we look across other verticals, breakout winners are emerging. For example, Harvey AI has captured significant market share within the legal market.
Given the vast quantities of capital available to support the growth of foundational model providers evidenced by multi-billion-dollar funding rounds and partnerships with hyperscalers like AWS, Google Cloud, and Azure it is difficult to predict just how deeply these companies will extend into the application layer. For instance, a host of startups that raised at punchy valuations to build AI-powered financial advisors were likely dismayed by Anthropic’s July 2025 launch of Claude for Financial Services, a specialized tool designed for investment analysis, market research, due diligence, and competitive benchmarking. This move highlights how foundational players can quickly encroach on vertical applications, potentially commoditizing niche markets and disrupting early-stage ventures.
What holds true, however, is that there are entire industries lagging one or two steps behind in readiness to fully benefit from AI’s efficiencies. Sectors such as aviation (including air traffic control), haulage, construction, freight logistics, manufacturing, and even parts of healthcare remain heavily antiquated, with manual processes, siloed data on on-premise servers, and non-digital workflows. According to recent analyses, data-poor industries like these are scrambling to digitize to avoid being left behind by AI disruption, as AI thrives on rich, accessible datasets. For example, manufacturing a $16 trillion global market, there is a push towards digital twins, IoT integration, and predictive maintenance, but many firms still rely on legacy systems that make such adoption impossible.
Taking aviation as a case study, U.S. airlines generated approximately $250 billion in revenue in 2024 across major carriers like Delta, United, and American. Yet this is an industry handicapped by outdated workflows, including manual checklists, spreadsheet inputs, and walkie-talkie communications. These inefficiencies have been well-documented, prompting the Trump Administration to pledge $12.5 billion in 2025 for modernizing air traffic control systems as part of broader infrastructure reforms. This includes accelerating the integration of unmanned aerial systems and NextGen technologies to enhance safety and efficiency. Aviation is just one example, but the broader trend applies to trillion-dollar markets like construction (where AI could optimize supply chains and site planning) and freight (where predictive analytics could reduce downtime and fuel costs). In healthcare, for instance, fragmented electronic health records and manual administrative tasks represent a $4 trillion opportunity for digitization before AI can fully automate diagnostics and patient management.
While billions of dollars in venture capital continue to pour into the foundational model layer driving innovations in reasoning models, multimodal AI, and edge computing, the top of the stack feels highly unpredictable. To return to first principles, the core benefit of AI is its ability to automate complex tasks at scale. In the meantime, as clarity emerges on how far foundational giants will stretch into applications, significant opportunities exist in digitizing these lagging industries. It feels logical that opportunity therefore lies in software solutions that bring some of these antiquated industries to a point of data optimisation where they can benefit from the advances in AI platforms such as IoT sensors for freight, cloud migration tools for construction, or data unification software for healthcare. These could be described as the “picks and shovels” enablers for these industries which offer defensible moats and compounding returns once AI automation kicks in. Paying attention to these foundational digitization efforts in the short term creates an interesting long-term potential as the AI market matures.