Interview, Fireside Chat, Conference Presentation
BlackRock's Tony Kim on AI's Next Winners? Chips, Memory, Robotics & Quantum
- The global technology sector is transitioning from a software-centric to a compute-centric model, a shift expected to systematically continue as AI drives value transformation and consumes market capitalization previously held by software.
- Base layer compute value has increased 10,000x, with server costs rising from $10,000 to millions of dollars, a trend predicted to fundamentally reshape market capitalization and lower margins compared to the high-margin SaaS era of the 2000s to 2020s.
- Data center designs will be redesigned to move data transmission distances from kilometers to millimeters, shifting from copper to light-based regimes to address power density, bandwidth, and heat constraints dictated by the laws of physics.
- A symbiotic co-design relationship between AI models and compute architecture is emerging, where model specs inform chip design, leading to architectures that increasingly mirror the human brain with a shift from compute-intensive to memory-intensive operations.
- A "Rampocalypse" of memory shortages is expected to persist due to a three-to-four-year duration mismatch between building chip fabs and current demand, causing market angst before the primacy of memory becomes more critical than compute.
- Approximately 90% of investment capital is anticipated to be allocated to a three-year window focusing on immediate AI demand and supply dynamics, while the remaining focus targets frontier technologies for a five-to-ten-year horizon, specifically milestones for 2030 such as utility-scale quantum computing, SMRs, fusion, 800-volt power architectures, and solid-state transformers.
- Many companies are predicted to lack free cash flow by 2031, requiring a belief system that their fundamentals could flip positive beyond that point as the AI industry evolves toward scalability and approval.
- China is expected to lead in the physical manufacturing and embodiment of robots due to lower costs and mass production scale, with 130 to 140 robotics companies identified and 30 to 40 potential IPOs expected in China this year compared to zero to three in the U.S.
- A future robotics market is predicted where Western AI brains are mixed with Chinese physical robot bodies, driven by the convergence of AI intelligence and physical embodiment, with social robotics for elderly care and combating loneliness expected to unlock a significant consumer and commercial market.
- Traditional enterprises are expected to redesign by creating an "ontology" or context layer around which agents interact using tokens, while "inference clouds" or "edge clouds" will emerge as the last mile for tokens to provide developer kits to small and medium businesses.
- Private equity firms and new efficient companies are expected to disrupt traditional industries by buying incumbents and reimagining the entire service stack to sell complete solutions, as traditional companies struggle to adjust to AI-driven efficiency unless acquired.
- The current fears regarding "X-pocalypse" or "Rampocalypse" are expected to subside within the next 12 months as compute and memory walls are plowed through by relentless demand.
- Major AI foundation labs are expected to go public in the next 12 months, accompanied by anticipated progress on orbital data centers which could trigger a radical rethink of terrestrial construction.
- The market is expected to re-adjudicate value and margin stacking as revenue shifts from high-margin SaaS to token-based revenue streams, with software programmers needing to be repurposed into memory co-design architects to address custom memory shortages.
- The "lost art" of analog computer design and custom chip architecture is predicted to resurge as the industry demands specific solutions, and the "token flow" framework will define the future of enterprise where value is captured by resellers or those providing the whole solution stack.
- Traditional company struggles and "scare" cycles regarding AI capex and financing are expected to continue every six months but will ultimately be overcome, while the physical sciences layer including material science and new memory design is expected to become a major unlock.