AI's New Economy: Value Shifts from Software to Silicon
By Ray Dombrowski · Reporting from Youngstown ·
BlackRock analysis suggests AI is driving a massive, physical revaluation of capital into foundational compute and hardware infrastructure.
The conversation surrounding AI, particularly on the recent episode of "Sourcery" with Molly O’Shea featuring BlackRock’s Tony Kim, paints a picture so focused on silicon and bandwidth that it risks forgetting where all this compute power has to land: in physical factories and worker hands. The most striking claim is the systematic revaluation of global capital—the shift from an era where value resided primarily in software and services (estimated at $10T+) to one where the "plurality of the dollars" now resides in foundational compute chips and hardware, potentially reaching $22T–$30T+.
On the podcast, Tony Kim argued that AI has fundamentally shifted the economic base layer from a software-centric world to a highly hardware-centric world. He detailed how this necessity drives a complete redesign of data centers, moving them through stages of technological evolution and requiring an unprecedented increase in compute power—up to 10,000x—and necessitating a "complete rebuild" of infrastructure. This is not just about faster chips; it’s about physics: reducing data transmission distances from kilometers down to millimeters.
The Great Re-materialization of Value
The key theme that jumped out was the return to physical reality. For years, I tracked labor market trends and industrial policy by looking at payroll—the actual dollars spent on human effort. Now, Kim argues that AI is causing a "physical world renaissance." He noted that chip companies maintain high profitability and that the industry has systematically shrunk from hundreds of players to a few behemoths with immense pricing power. The value stack, according to his analysis, is moving away from "asset-light high margin" cloud services toward "asset-heavy lower margin big dollars"—the foundational compute infrastructure itself.
While this concentration of capital in hardware seems inevitable given the technical constraints—such as memory intensity skyrocketing and the difficulty of manufacturing fabs (a 3–4 year lead time)—I am wary of treating this shift purely as a financial arbitrage opportunity. The sheer scale of investment into chips, accelerators, and next-gen architectures is undeniable. However, I keep returning to the human element. Kim dedicated significant time to robotics, noting that the development mirrors AI progress, but emphasized that the physical embodiment—the arms, limbs, hands—is still the most challenging part, requiring a manufacturing hardware business approach.
Beyond Silicon: The Industrial Bottlenecks
The argument for "co-design"—tightly integrating silicon design with model parameters—is technically sound and represents the leading path forward. But when I look at this from an industrial policy angle, the bottleneck shifts. We are talking about foundational models that require massive power density, pushing us toward 800-volt architectures and even considering utility-scale technologies like Small Modular Reactors (SMRs) or orbital data centers.
This isn't just a capital expenditure problem; it is an energy grid problem and a labor force problem. If the next generation of compute requires moving from megawatts to gigawatts, that demand must be met by physical infrastructure—transmission lines, power plants, and specialized construction talent. The market may be betting heavily on the "tomorrow" (the 5+ year horizon), but nothing builds itself.
Where Payroll Meets the Photon
The true test of this compute revolution won't be who designs the best chip or who achieves the highest memory intensity; it will be which region and industry can most efficiently solve the massive, real-world logistical problems associated with deploying that power. The promise of advanced robotics—especially for social embodiment in aging populations—is compelling because it directly addresses a demographic cliff that impacts payrolls globally.
The current focus on abstract layers like "context layer" or "ontology" is valuable for software architects, but the underlying reality remains rooted in physical throughput: moving data via fiber, converting electricity into usable power density, and building reliable factories to house these systems. The narrative of compute supremacy is compelling, yet it must be tethered to the slow, expensive, messy process of concrete pouring, copper wiring, and skilled labor deployment.
The next wave of value creation will not simply flow from the most powerful chip; it will accrue to the industrial players who can reliably bridge the gap between theoretical computational power and practical, scalable physical implementation.