BlackRock's Tony Kim on AI's Next Winners?

1. Key Themes

The Great Value Migration: From Software to Compute

Kim's central macro observation is that AI has fundamentally inverted the tech value stack. What used to be a software-centric market has become compute-centric, and this shift is visible directly in market capitalization data. "There's roughly 10 plus trillion market cap in software and services and internet... There's 22, 23 trillion in Mag 7. And then there's another 30 plus trillion in chips and hardware... 10, 20, 30. Before AI, in BCE era, it was probably reversed." 00:00:00 He frames this as a durable four-year trend, not a blip: "you've seen in the last four years a transformation in value that has systematically been happening for the last four years." 00:00:27

The Complete Data Center Rebuild

Kim argues the entire physical infrastructure of computing is being reconstructed from scratch because AI workloads have entirely different physics than cloud computing did. "The internet as we know it was built... around the birth and the dawn of cloud computing... megawatts, not gigawatts... CPU with a hard drive... And now those compute servers are millions and tens of millions of dollars." 00:16:38 The redesign cascades through power, cooling, and data movement: "the data center is changing from... we would transmit data kilometers. And now building to building. And then it's within the building... rack to rack... within the rack... within the chip." 00:06:11

Model-Chip Co-Design as the New Competitive Frontier

Rather than generic compute, the leading labs are now tightly integrating silicon design with model architecture. "The best models obviously are trained on and built on the best compute and most optimized inference... this notion of co-design of tightly integrating the design of your silicon to match the parameters and the specs of the model... is like the new path that many of the leading foundation labs are pursuing." 00:08:27 He cites Broadcom's new "jalapeno chip" as a direct example of this trend. 00:08:53

"Rampocalypse": Memory Is the Next Bottleneck, Mirroring the Human Brain

Kim identifies memory (DRAM/HBM) scarcity as an underappreciated and escalating crisis, driven by AI architectures increasingly resembling brain structure. "Increasingly, more and more of chip and model development is starting to mirror the human brain... the human brain is very memory intensive. And today is more compute intensive. But... the memory intensity has just skyrocketed." 00:11:21 He flags a critical supply mismatch: "it takes three or four years to build a chip fab or a memory fab. But yet everyone is a shortage today... there's this like mismatch of demand supply, duration. And this is causing a lot of angst in the market." 00:13:10

Token Flow as the New Enterprise Operating Model

Kim proposes a new mental model for enterprise value creation centered on "token flow" — everything reduces to who creates, sells, or repackages AI tokens. "Follow the flow of tokens. Either you create tokens, compute. You then serve the tokens, foundation labs. And then you put a harness, package, context around the token... If you're not in that flow, it's a problem." 00:54:42 Enterprises will be defined by proprietary data plus a "context layer": "can you embody all of the knowledge of your company in a... context layer? A layer of the secrets and the ways of your company." 00:54:00

China's Manufacturing Scale Advantage in Robotics

Kim highlights an overlooked geographic imbalance: China's robotics ecosystem dwarfs the US in company count and public-market financing velocity. "There's, I think, 130, 140 robotics companies in China... I see 30, 40 potential IPOs this year in China. Alone this year. And there's, what, 0, 1, 2 in the United States maybe this year." 00:44:57 He attributes this to funding structure differences: "the U.S., the lack of depth in the private markets in China. So they're using public markets as a funding mechanism, unlike in the U.S." 00:45:28

2030 as the Convergence Point for Deep Tech

Multiple frontier technologies — quantum, space data centers, fusion/SMRs, next-gen power architecture — are independently converging on the same target date. "All roads converge to 2030. It's like quantum computing, utility scale, logically error-corrected, million-qubit quantum computer, 2030. SMRs, fusion small nuclear reactors with regulatory approval... 2030." 00:32:46

Semiconductors: The Misunderstood "Commodity" That Never Was

Kim pushes back hard on the long-standing narrative that chips are commoditized, pointing to consolidation dynamics and margin data as proof of durable pricing power. "People always said chips are a commodity, but yet they have the highest profitability of any company in the world... higher margins than software, pharmaceuticals, industrials, telecom, anything." 00:39:53

PE/VC Roll-Ups Reimagining Legacy Industries with AI-Native Cost Structures

A new category of "buy the incumbent, rebuild the stack" companies is emerging, using AI to compress headcount and cost radically. "A new company with 500 people that can do and generate the revenue of 10,000 people... they're just approaching it in a radically rethought process." 00:58:59 Kim cites a live example: "Long Lake, just bought Amex Global... their travel business... you have a small player buying a large player." 00:58:34 (Molly O'Shea)

2. Contrarian Perspectives

Moats Are a Losing Framework — Speed Is What Matters

Kim explicitly rejects the venture-standard obsession with defensibility in an era of exponentially improving models. "Moats are always the breach, aren't they? So it's more about offense, in my opinion. Can you move faster?" 00:27:08 This is a direct challenge to the standard investor framework of durable competitive advantage.

The "Cool Today" Trap: Present Relevance Signals Future Irrelevance

Kim argues that chasing what's fashionable today in AI is often a red flag for exit value five years out, inverting normal momentum-investing logic. "You always want to be betting on not what's cool today. Will you still be cool in five years? Because... you become yesterday's news in five years, because you, even though you are cool today." 00:35:48 This directly informs his lower allocation to "the app layer," which he says is "struggling to find their place" in the token-flow stack. 00:56:06

Social/Companionship Robots, Not Industrial Robots, May Be the Bigger Market

Rather than betting on humanoid robots for manufacturing, Kim's personal conviction is that AI-embodied companionship for aging and lonely populations is the larger overlooked opportunity. "I have a soft spot... one of the things I'm most interested in in the robotic side is around not so much the manufacturing robot... It's around loneliness, around social embodiment... to bring consumer and or commercial like social robots more so than industrial use." 00:47:14 He backs this with demographic data: "the birth rates are well below 1.0. And you need 2.1, 2.2 to stay even... some of the best performing companies in the world are nursing home companies." 00:48:15

Full Human-Level Motor Function Robots May Be Unnecessary

Against the industry's obsession with achieving perfect dexterity (hands, articulation), Kim suggests a "good enough" R2-D2/C-3PO-style approach could unlock a bigger consumer market faster than pursuing Optimus/Figure-style perfection. "Do you need to have a fully figured perfect motor function with... all the hand articulation? Or could you get something that can appeal to that? And I think that's possible." 00:50:42

China May Win the Robot Body Even If It Loses the Robot Brain

Kim predicts a bifurcated global supply chain where Western AI labs provide the "brain" but Chinese manufacturers provide superior, cheaper hardware bodies — a geopolitically uncomfortable but economically logical outcome. "What you might have is ultimately you can mix and match Chinese physical robot with a Western brain. And I know that's happening... those Chinese robots are amazing, right? So why don't we stick a Western brain in there." 00:46:43

3. Companies Identified

Broadcom — Semiconductor/networking giant. Cited as the leading example of model-chip co-design in practice via its new AI accelerator. "Obviously, they did that with a new jalapeno chip that recently came out." 00:08:53

D-Matrix — Next-generation compute architecture (accelerator) company. Featured as a RAISE panel topic on "next-gen computer architectures." 00:01:31

PsyQuantum — Quantum computing company. Panel partner for Kim's quantum computing discussion. 00:01:47

Lumentum — Optics company. Panel partner for discussion on "bringing optics to kind of next-gen data center design." 00:01:47

SambaNova — AI chip/systems company. Kim describes personal involvement: "One of my companies I was involved with, Samba Nova, Lipu was supposed to be one of the speakers." 00:02:16

SK Hynix — Memory chip maker, one of only three major DRAM/memory players globally, referenced amid its IPO. "There's only three main players... And SK Hynix is about to go public." 00:15:43 (Molly O'Shea)

Boston Dynamics (owned by Hyundai) — Robotics company behind the hydraulic Atlas humanoid robot. "That's Korean, actually. That's Hyundai, that owns Boston Dynamics." 00:46:18 (Tony Kim); Atlas described by Molly O'Shea as capable of picking up "like a fridge." 00:39:10

Figure AI — Humanoid robotics company focused on commercial/daily use cases like package sorting and auto manufacturing, visited by Molly O'Shea. "Figures is more of like daily use package sorting, commercial stuff, making cars." 00:39:10

Assembly AI — Voice/speech AI infrastructure company, cited as a standout growth story. "Speech model companies are crushing it... There's a company called Assembly AI that is like growing incredibly fast." 00:51:32 (Molly O'Shea)

Databricks and Snowflake — Data platform companies described as capturing outsized value from the downstream data layer buildout. "The downstream effects of it now hitting the data layer... the Databricks, the Snowflakes, they're hitting some of that extra premium in the market." 00:51:58 (Molly O'Shea)

MongoDB — Database platform for AI-native application development, described as benefiting from the proliferation of agent-built apps. 00:52:27 (Molly O'Shea)

Palantir — Cited for pioneering the "ontology" concept as the enterprise context layer. "I think Palantir calls this ontology." 00:54:27 (Molly O'Shea)

General Catalyst Creation Fund and Long Lake — Private equity/venture vehicles executing "buy and rebuild" roll-ups; Long Lake's acquisition of Amex Global's travel business cited as a landmark example of a small player absorbing and re-architecting a legacy incumbent. 00:58:57 (Molly O'Shea)

SpaceX — Referenced in context of orbital data center ambitions and its pending IPO. "That is a huge weight... especially with SpaceX coming and the whole IPO on that." 00:38:41 (Molly O'Shea)

4. People Identified

Tony Kim — Head of BlackRock's Global Technology team, public and private tech investor. Central guest; described his own multi-decade investment framework built around compute primacy, token flow, and long-duration deep tech bets extending to 2030 and beyond.

Jamin (Coatue) — CIO of public markets at Coatue, referenced by Molly O'Shea as corroborating Kim's memory/agent-proliferation thesis from a separate conversation. "We had a conversation with Jamin from Coatue. He's the CIO of public markets over there... they were talking about with the proliferation of agents, memory is only increasing more." 00:15:14 (Molly O'Shea)

Henri — Founder/organizer of the RAISE Summit in Paris, praised for building the event from a small gathering into "Europe's biggest or most targeted AI conference" in just a few years. "It was the second year of development that Henri had kind of pioneered and built this event." 00:02:38 (Tony Kim)

Eric Schmidt — Former Google CEO, mentioned as having previously appeared at RAISE, signaling the conference's rapid rise in stature. 00:03:26 (Molly O'Shea)

Scott Wu (Cognition), Andrew Feldman (Cerebras), Rodrigo Gang (SambaNova), Michael Hurlston (Momentum/onsemi), CJ (MongoDB) — Named as fellow guests in the broader RAISE interview series, indicating a network of leaders across AI coding agents, AI chips, and enterprise software. 01:07:37 (Molly O'Shea)

5. Operating Insights

Allocate Capital Across Explicit Time Horizons, Not a Single Thesis

Kim describes a structured portfolio construction discipline: roughly 90% of capital deployed against a 3-year "now" window with a required belief in 5+ year durability, and a smaller deliberate allocation to decade-long "frontier of the frontier" bets. "Within this three-year window... that is probably where 90-plus percent of my investment... is going in... But then there's also, you've got to be betting on the frontier of the frontier." 00:29:33 This is a transferable operating discipline for any capital allocator: separate the "now" bet from the "does it survive" bet from the "moonshot" bet, and size each deliberately.

Underwrite Exit Multiples Against Growth Deceleration, Not Current Growth

Kim flags a specific and often-missed risk: many current AI-hot companies won't reach free cash flow within a 5-year window, and multiple compression compounds with growth deceleration to create a "bind." "A lot of companies will not even have free cash flow by 2031... if you're just following the trend of today... it will not be what you think it is in five years. And the multiple that people will pay will go down. And your growth rates are decelerating. And now you're in a bind." 00:35:24 Operators and investors should stress-test valuation not just on trajectory but on multiple durability at exit.

Identify Where You Sit in "Token Flow" Before Building Your Business Model

Kim's practical heuristic for evaluating any AI-adjacent company: does it create tokens, serve tokens, or repackage tokens with proprietary context? Companies unable to answer this are structurally exposed. "You must be in this token flow to either resell or repackage the tokens with your context and your very specific application... If you're not in that flow, it's a problem." 00:55:09 This is a useful operating/positioning framework for founders building anywhere in the AI stack.

Bet on Talent Gaps as a Source of Structural Advantage

Kim notes a severe, underappreciated talent shortage in physical/hardware disciplines (chip design, custom memory architecture) that creates outsized opportunity for those who can source or retrain talent. "We cannot get people to design custom memory... There are no people. We've got to repurpose some of these software programmers into memory co-design architects." 00:42:44 Operators in hardware-adjacent AI infrastructure should treat talent acquisition/retraining as a core strategic lever, not an HR afterthought.

6. Overlooked Insights

The Full-Solution Abstraction Play: "I'll Just Do the Whole Thing"

Buried in the token-flow discussion is a subtle but potentially enormous business model insight: rather than selling a software layer or a service, some companies are choosing to abstract away the entire stack and simply take over an entire business function outright at a steep discount. "I will do all of your claims processing. I will do all of your insurance processing. So you don't know what are you. Are you an app company? Are you a service company? Are you a compute company?... No, I'm just selling you the whole solution. Today you used to pay 100. I'll charge you 20." 00:56:58 This wasn't dwelled on, but it describes a fundamentally different competitive category from typical SaaS or vertical AI tools — full business-process substitution priced at 80% discounts to incumbent costs, which could be one of the most disruptive AI business models but received almost no elaboration in the conversation.

Space-Based Data Centers Could Obsolete Current Terrestrial CapEx Planning

Kim mentions orbital data centers almost in passing alongside quantum and SMRs, but the implication is understated: if this timeline holds, it could strand or devalue the very terrestrial data center buildout (the trillion-dollar CapEx cycle) that dominates today's AI investment narrative. "If you keep pushing on that progression, it could open... a rethink... and moving the burden of terrestrial compute into space. And that can have huge implications, huge implications of how current data centers are even being built." 01:03:05 Given that Kim frames the current data center buildout as the primary investment vortex of the next 3 years, a credible 2030 orbital alternative represents a significant, underdiscussed long-duration risk to today's infrastructure valuations — a point made in a single sentence but with outsized implications if realized.