Inside the AI Value Chain: Reflections from the Road
This June, our investment team embarked on their annual AI road trip across the US West Coast, meeting with executives from 18 companies that command a combined market cap of approximately $16 trillion.1 Representing a diverse cross-section of the technology sector—from industry giants to early-stage startups—the public and private firms we spoke with span the entire AI value chain, including leading hyperscalers, semiconductor pioneers, software developers, and cybersecurity providers. These firsthand conversations offer differentiated insights that shape our investment theses in this rapidly evolving landscape. Here we share our key reflections from the trip.
Enterprise Adoption, Advertising, and Software Enterprises: Enterprise AI adoption is accelerating, but it is still in early stages. In one of our meetings, we came to understand that many of the companies in the top 5% of AI adoption are now consuming three times more tokens than those at the median—a usage gap that has doubled over the past year and continues to widen. Despite this concentration of consumption, we share the view of the leadership teams we met that the medium-term outlook points to a multi-winner landscape rather than a winner-take-all market. This dynamic is already visible in the developer tool space, where Anthropic is finding rapid success with Claude Code while OpenAI continues to command strong traction.
Advertising: AI is actively expanding the market for digital advertising by driving higher conversion rates and providing automated creative tools. In Search, we are still in the early stages of a higher-monetization phase. Longer, conversational queries in AI Overviews are expected to monetize at significantly higher rates than traditional search over time. For example, one major hyperscaler noted that it is only 30% penetrated with its AI advertising products, which are already delivering a 15% to 30% return on ad spend lift. Another highlighted that ad growth in the social media space is being driven primarily by higher average revenue per user, as AI-enabled ad creatives boost click-through and conversion rates.
Software: On the software front, we heard that vendors are actively restructuring their technology stacks to support autonomous AI agents. Value is shifting from being a Systems of Record (SoR) that serve as a shared reference point businesses to Systems of Action (SoA) that focus on the automation of data, moving from passive dashboards to active agents that solve problems. Value will accrue to vendors that can correctly, safely, and repeatedly turn data into actions. Software businesses are beginning to adapt to this shift. However, the long-term economic models, pricing structures, and monetization strategies for agent-based software remain highly uncertain.
Watch Brook Dane's live update from the road.
AI Buildout, Bottlenecks, and Semiconductors
Power: In our conversations, power was repeatedly and unanimously cited as the most critical and longest lead-time constraint in AI scaling. The extreme difficulty of securing multi-gigawatt sites is ushering in behind-the-meter solutions and bringing about a shift toward smaller, distributed data centers, which are highly suitable for inference workloads. Beyond power, supply-chain bottlenecks are expanding to include compute, networking, memory, silicon wafers, and specialized components like lasers. Limited advanced-packaging and foundry capacity are pushing enterprises to secure second-source partners: TSMC has been somewhat of a gatekeeper for leading-edge chips but Samsung is now entering the space, followed by Intel, and some companies are exploring proprietary fabrication facilities like Tesla’s Terafab.
Semiconductors: A key takeaway from our meetings is that the semiconductor value chain is not a zero-sum game. Multiple parallel technologies are thriving simultaneously to meet different architectural needs. For example, we came to believe that copper and optics could coexist and grow to meet rising bandwidth demands over varying physical distances within data centers. Similarly, merchant graphics processing units (GPUs) will continue to win on flexibility and fast time-to-market, while custom application-specific integrated circuits (ASICs) will capture high-volume, targeted workloads requiring maximum cost and power efficiency. Hyperscalers are developing their own custom silicon: Google has its TPU, Amazon has Trainium, and Microsoft is releasing Maia in the second half of the year. We estimate Broadcom’s custom ASIC business has become a $25 billion+ enterprise2 as demand and adoption take off. We think rapidly accelerating inference workloads are expanding the market enough for multiple vendors to succeed.
Watch Sung Cho’s live update from the road.
Agentic Commerce, Cybersecurity, and Robotics
Agentic Commerce: The conclusion we draw from our conversations on this topic is that the underlying technical infrastructure for agentic commerce is largely ready. Yet, agent adoption is likely to accelerate in some areas and move more slowly in others. Executives emphasized that the main hurdle to greater adoption now is consumer trust. Agents are still perceived as somewhat unrefined, prone to making incomplete decisions, poor booking plans, or misunderstanding context. We expect the adoption curve to begin with low-value, low-consideration purchases where the margin for error is relatively higher, before expanding to higher-value transactions.
Cybersecurity: While advanced AI models are discovering new attack paths and exploiting vulnerabilities at an unprecedented pace, this heightened threat environment has not yet translated into a major surge in enterprise cybersecurity spending. We heard that although security teams are highly concerned about these rapid, AI-driven exploits, budget allocations have remained relatively flat. However, when spending does catch up, leadership teams expect the market to heavily favor "best-of-suite" vendors over niche, "best-of-breed" players. This shift is driven by the structural advantage of suite providers, who can aggregate massive datasets across their entire product portfolios to train and deliver more advanced, AI-driven security intelligence.
Robotics: While humanoid robotics captured significant mindshare, the tech leaders we met acknowledged that the technology is still in the very early stages. The consensus is that widespread commercial viability for humanoids is still years away. In the near term, autonomous mobile robots will likely continue to be the primary growth driver in the robotics space, with purpose-built robots deployed across industrial applications.
To hear more of our observations and ways to invest in the evolving tech landscape, our Fundamental Equity team is ready to explore these opportunities with you.
1 Goldman Sachs Asset Management. As of June 19, 2026.
2 Broadcom's 2Q FY26 earnings call on June 3, 2026.
