BREAKING: BlackRock's Tony… | Sourcery Newsletter Summary

1. Key Themes

The World is Shifting from Software-Centric to Compute-Centric

The prevailing model of the last two decades — where cloud providers resold cheap hardware and software captured all the margin — has broken down. AI has fundamentally repriced the base layer of compute.

"At the end of the day, cloud computing, everyone says it's software, but it was really reselling CPUs with hard drives."

"A $10,000 server is a $1,000,000 server."

The market cap consequence is already visible: roughly $10T in non-Mag 7 software/services/internet, $20T in Mag 7, and $30T in non-Mag 7 compute, chips, and hardware — proportions that were closer to reversed before AI.

"The models themselves, like it or not, have consumed the market cap out of software and services."

Memory is the Next Critical Bottleneck ("RAMpocalypse")

The market is fixated on compute, but Kim argues memory intensity will become the defining constraint as AI architectures mature and begin mirroring the human brain.

"Today we're all talking about compute, compute, compute. I think the primacy of memory will become even more important."

Supporting data: DRAM contract prices rose ~90% in Q1 2026 QoQ; SK Hynix describes its DRAM, HBM, and NAND capacity as sold out for the year; Samsung, SK Hynix, and Micron control over 95% of global DRAM production. A structural duration mismatch compounds the issue:

"It takes 3-4 years to build a chip fab or a memory fab, but yet there's a shortage today."

A $10 Trillion Physical Rebuild is Underway — Infrastructure, Not Just Software

The AI buildout is a hardware and physical science renaissance. The 5 largest US hyperscalers have committed $660B–$725B in capex for 2026 alone. JPMorgan puts global AI-related capex through 2030 at $5.5T; McKinsey's data center estimate reaches $6.7T.

"The trillion dollars of CapEx this year and the $10T over the next 5 years that are coming, is to move data centimeters and millimeters. That's AI."

This redesign spans every layer: copper-to-light interconnects, 800V power architectures, solid-state transformers, SMRs, and rack-level materials science.

China is Winning the Robotics Race — A Wave of IPOs is Coming

Kim sees China dominating the robotics body manufacturing layer, with 130–140 robotics companies and a pipeline of 30–40 potential IPOs. Barclays estimates Chinese firms accounted for 85% of humanoids shipped in 2025. The West leads on AI brains, but the two are already combining:

"You can mix and match Chinese physical robot with a Western brain, and I know that's happening."

"Token Flow" is the New Framework for Enterprise Value

Kim proposes a new mental model for understanding which software companies survive AI disruption. Enterprise value accrues to whoever sits inside the flow of intelligence tokens — companies that can repackage tokens with proprietary context and data.

"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."

The stack: intelligence tokens flow in → data layer (Databricks, Snowflake, MongoDB, BigQuery) → context layer (Palantir-style ontology) → agents act. SaaS moats are under pressure; speed of offense replaces defensibility.

"Moats are always the breach, aren't they? It's more about offense in my opinion, can you move faster?"

2. Contrarian Perspectives

Semiconductors Were Never a Commodity — Their Margins Prove It

The consensus view for 20 years was that chips were a commodity input. Kim calls this the "20-year lie."

"People always said chips are a commodity, but yet they have the highest profitability of any sector. They're higher margins than software, pharmaceuticals, industrials, telecom, anything."

The structural evidence: hundreds of chip companies consolidated into a handful of dominant players with massive pricing power; VC largely exited the category. The result is the exact opposite of commodity economics.

"The number of companies have collapsed, and then those that have survived are behemoths with huge pricing power. The complete opposite of commodity."

The Biggest Robotics Market May Be Companionship, Not Labor

Against the dominant "factory automation" framing for humanoid robots, Kim makes a case for social robots addressing loneliness and aging populations — a use case almost entirely absent from mainstream robotics investment theses.

Birth rates across Asia sit well below the 2.1 replacement level; nursing home operators rank among the best-performing companies in the world. The form factor he envisions is "closer to C-3PO than Optimus — half size, approachable, multilingual."

"Even though you might not have perfect motor function, if you could embody some intelligence, empathy, I think that could unlock a really, really interesting market."

PE Roll-Ups May Be the Highest-Leverage AI Disruption Play

Rather than building new AI-native software, Kim identifies a contrarian pattern in private equity: buy existing customers, then refactor the delivery layer underneath with AI, collapsing cost structures dramatically.

"You abstract away all of those layers of the stack and just say, I will take on all of your insurance. Pay me X. Today you used to pay 100, I'll charge you 20."

He is watching for 500-person companies generating the revenue of 10,000 as the proof-of-concept pattern.

3. Companies Identified

BlackRock World's largest asset manager, $15.3T AUM Kim's employer and platform; context for the scale of capital allocating into these themes. iShares ETF platform exceeds $6T; alternatives book at $715B growing fastest.

Microsoft, Alphabet, Amazon, Meta, Oracle The five largest US hyperscalers Collectively committed $660B–$725B in capex for 2026, close to double 2025 levels — the primary drivers of the $10T infrastructure buildout.

SK Hynix, Samsung, Micron The global memory oligopoly Control over 95% of global DRAM production; all have crossed $1T in market value. SK Hynix describes its capacity as sold out for the year. Central to the RAMpocalypse thesis.

"Only three companies control memory."

d-Matrix Next-gen AI accelerator company Featured alongside Kim on a panel at RAISE Summit on next-gen accelerators; positioned as a company to watch in the AI chip co-design space.

PsiQuantum Quantum computing company Appeared with Kim on a quantum computing panel at RAISE; one of Kim's frontier bets targeting the 2030 convergence window.

Lumentum Optical interconnect / data center optics Featured with Kim on a panel about optics in data center design; beneficiary of the copper-to-light transition inside data centers.

Broadcom Semiconductors / XPU / AI chip co-design Featured with Kim on an XPU + AI chip co-design panel at RAISE; a key player in the custom silicon and co-design trend.

Palantir Enterprise AI / data ontology Named as the reference for the "context layer" (ontology) in Kim's token flow framework — the company whose architecture concept Kim uses to define enterprise AI value.

Databricks, Snowflake, MongoDB, BigQuery (Google) Data infrastructure / data layer Named collectively as the data layer in Kim's token flow enterprise stack — the tier holding proprietary and third-party data that agents interact with.

Unitree Chinese humanoid robotics Won Shanghai STAR Market IPO approval in June 2026 at ~42B yuan valuation, 73 days from application to approval. Leading example of China's robotics IPO wave.

AgiBot Chinese humanoid robotics Targeting a HK$40B–HK$50B Hong Kong listing; part of the 30–40 Chinese robotics IPO pipeline Kim identified.

EngineAI Chinese humanoid robotics Founded 2023; filed confidentially for IPO after opening a Shenzhen factory building a humanoid every 15 minutes. Cited as evidence of China's manufacturing velocity.

Anthropic AI foundation lab Filed confidential S-1 on June 1, targeting late 2026 Nasdaq debut; last private round at $965B valuation. Part of Kim's thesis on big lab IPOs.

OpenAI AI foundation lab Filed confidential S-1 days after Anthropic; reportedly weighing delay to 2027. Kim cited upcoming lab IPOs as a key near-term proof point.

4. People Identified

Tony Kim Managing Director & Head of Global Technology Team, BlackRock Fundamental Equities Central interview subject. Oversees public and private tech investments spanning semiconductors, memory, AI infrastructure, robotics, and quantum. Began AI compute investments in 2019–2021, ahead of the generative AI wave.

"I'm a public investor, I'm a private investor, I do both."

Molly O'Shea Host, Sourcery Newsletter / Podcast Author and interviewer. Conducted the conversation at the RAISE AI Summit in Paris.

"It was such a pleasure to sit down with Tony, he articulates systems very well with a good sense of humor."

5. Operating Insights

Your Company Must Be Inside the Token Flow — Or It Faces Existential Risk

Kim's token flow framework is a direct diagnostic tool for operators. Map your product against the three-layer stack: data layer, context layer, agent layer. If you cannot identify where you sit in the flow of intelligence tokens — either reselling them, enriching them with proprietary context, or packaging them into a workflow outcome — you are 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."

The "Still Cool in 5 Years" Test Should Gate Every Investment and Product Decision

Kim applies a single durability filter to all positions: will this still be relevant beyond the 3-year vortex? For operators, this translates to a discipline of not over-indexing on today's hot capabilities while neglecting to build compounding advantages (data, context, relationships) that persist.

"You always wanna be betting on not what's cool today. Will you still be cool in five years?"

Talent Reallocation is an Immediate Bottleneck — Retrain Software Engineers Into Hardware Roles

The physical sciences renaissance is stalling not on capital but on people. Kim relayed a dinner conversation with one of the largest memory companies where custom co-design work is blocked for lack of qualified designers. The actionable implication for operators and workforce planners is explicit:

"We gotta repurpose some of these software programmers into memory co-design architects."

6. Overlooked Insights

Orbital Data Centers Could Render Today's Terrestrial Infrastructure Decisions Obsolete

Mentioned only briefly, this frontier bet carries significant implications. If compute migrates to orbit, the entire basis of current data center siting, power sourcing, and cooling design could be disrupted before the current buildout fully depreciates.

"Moving the burden of terrestrial compute into space could have huge implications of how current data centers are even being built."

Kim's Investment Framework Has a Hard Discard Rule — Not Just a Preference

It is easy to read Kim's 90/10 split (vortex vs. frontier) as a soft allocation guideline. But the underlying rule is sharper: anything that fits neither the 3-year AI vortex nor the frontier thesis with nonlinear asymmetric potential gets cut entirely.

"I want nonlinear asymmetric potential."

For portfolio managers and venture investors, this implies a more ruthless pruning discipline than most public discussions of diversification suggest — the middle ground of "interesting but not clearly in or out" gets no capital.