Adam Chambers

1. Key Themes

Physics Is Running Out, So Hardware Must Get More Inventive

Steven Latré (IMEC) frames the current moment as the end of the scaling era. Transistor shrinkage, the engine of 40 years of chip progress, is hitting atomic limits, which forces a more creative approach to chip design. The implication is that progress shifts from "just shrink it" to architecture, packaging, and interconnect innovation.

Latré: "we're at the level that a transistor has about the depth of one nanometer What is one nanometer? It's about 100 000 times smaller than a human hair That means that that we're at the end of physics of what we can still do" [00:02:08]

He continues: "what we do in hardware is going to be needs to be way more inventive than what we've been doing in the last 10 to 20 years to keep on scaling that that roadmap" [00:02:36]

The Bottleneck Has Moved from Training to Inference, and to Physical Constraints

Adam Chambers (investor) maps the chip landscape by where the bottlenecks sit. GPUs handle training well, but the constraint now lives in power, memory, cooling, and data transfer. This is a useful investing lens: back the companies that unblock the constraint, not the ones that ride the headline.

Chambers: "GPUs are very good at the training part and We're now looking at the bottleneck that sits in the inference layer and The bottlenecks that I look at the most and try and Look at companies that attack those bottlenecks sit in power and energy memory cooling and data transfer" [00:03:09]

Memory Is the Most Acute Shortage, and Pricing Power Is Extreme

Both speakers converge on memory as the number-one broken component. Chambers gives the market structure: three suppliers, roughly 95% share, with hyperscalers absorbing inventory.

Chambers: "there's only three companies in the world that produce memory Which is 95 of the market micron? sk hynix and samsung two of them are in korea and a lot of their shift has Been filled out by hyperscalers taking over the their whole inventory So you've seen prices rise by 700 percent this year just for the memory part of the chip" [00:04:52]

Latré: "memory memory is absolutely the number one Component that that is already broken to today" [00:07:49]

The Scaling Hypothesis Is Ending: Model Sizes Will Stop Growing

A non-consensus view from someone who sees the roadmap five to ten years out. Models already exceed human-brain parameter counts while the hardware running them is wildly less efficient, so the industry will pivot away from ever-bigger models.

Latré: "the latest open source models like Kimi K3 already have more parameters than what the human brain has but on the other hand that That K3 chip or the chip that we need To fuel that that K3 model is about a million times more or less energy efficient than what the human brain brain is" [00:07:49]

Latré: "We kind of went into a an era the last five years of building Bigger and bigger and bigger models. We went from 175 billion parameters to to kind of 10 trillion parameters right right now That's gonna stop and there's a lot of companies right now taking that shift as we speak Abandoning what we call the the scaling hypothesis" [00:08:19]

Photonics as the Cross-Cutting Technology That Attacks Multiple Bottlenecks

Photonics is described as the single technology that addresses power, cooling, and data transfer simultaneously. The near-term path runs from linear pluggable optics between clusters to co-packaged optics right next to the GPU.

Chambers: "It's attacking energy because it can transfer data quicker It's attacking cooling because you don't need as much cooling around these gpus if you're transferring through light And then it's attacking data transfer at the core... It's like the key technology that is likely to win" [00:11:04]

Chambers on the roadmap: "at the moment It's between the clusters and we're using linear pluggable optics But I think in the next five years we will see These optical components get closer and closer to the actual gpu and we will go towards a state of co-packaged optics" [00:09:36]

Latré calls it a "duck killer technology to to enable the chips of the future" [00:10:55] (likely "deal-breaker/killer" enabling tech, in his words).

Hardware Diversification Beyond the GPU

As software approaches diversify, hardware must follow. The GPU monoculture is expected to fragment, which is a thematic opening for specialized silicon.

Latré: "in the software world We're going to see way more different types of Algorithms and as a result, we will also need More diversification in the type of hardware so more than just just gpus as as well as models get bigger and bigger" [00:07:23]

Manufacturing, Not Chip Design Startups, Is the Real Constraint

Despite the surge in new chip companies, Latré says design capacity is not the limiter. The scarce resource is manufacturing the right components in the right regions.

Latré: "the bottleneck is manufacturing Manufacturing the right type of components In the in different regions... having the right type of components for that is I think the real bottleneck not necessarily the The number of companies that do this because every company can come with a Completely different unique design" [00:06:54]

Cooling Is Moving Physically Closer to the Silicon

Chambers expects liquid cooling to migrate ever closer to the GPU, with value accruing to manufacturers of that hardware.

Chambers: "bringing the liquid that takes heat away from the chip Closer and closer to the gpu itself Is how it's going to progress over the next five to ten years and I think a lot of the value will go to the manufacturing side of that" [00:03:35]

Europe-Plus-US Founder Profile as a Talent and Investing Heuristic

Chambers offers a specific pattern-matching rule for founders.

Chambers: "it's a very good idea to recruit or invest in people that have Had a academic and technical background in europe and had the wherewithal and business sense In the us" [00:12:29]

2. Contrarian Perspectives

Today's AI Will Look as Quaint as Dial-Up Internet

Latré's hottest take directly challenges the prevailing belief that LLMs are the destination. He argues they are a brute-force stopgap, and a software revolution tied more closely to hardware is coming.

Latré: "we're gonna laugh at how old-fashioned ai was today... I fundamentally don't think that Today's technology that this is what we're going to be talking about in 10 years from from now I think there's an actual also in software revolution is coming" [00:11:42]

He adds the analogy: "in 10 years from now, we're going to talk about large language models Which are very much a brute force approach in exactly the same way?" [00:12:08]

The Value Pyramid Hasn't Flipped, Software Is Just Immature

When Molly O'Shea suggests hardware is eating software's value, Latré pushes back: the current state of software is a brute-force phase, and value will return to software once it co-evolves tightly with hardware.

Latré: "I don't really think it's necessarily flipped I think it's more the immaturity of the software market right now... it's still very much a brute force approach And as a result the value that you can get out of it is pretty limited" [00:05:50]

Bigger Models Are a Dead End

The consensus assumption that scaling laws drive AI forward is rejected by the speaker closest to the long-horizon hardware roadmap.

Latré: "I don't really think that models will become necessarily bigger and and bigger" [00:07:49]

More Chip Startups Won't Solve the Constraint

Where many investors assume a wave of new chip companies unlocks capacity, Latré locates the problem in manufacturing and components rather than design or company count.

Latré: "it's rather the bottleneck is manufacturing... not necessarily the The number of companies that do this" [00:06:54]

3. Companies Identified

IMEC

Belgium-based semiconductor R&D organization, the world leader in early-stage chip development, with a unique clean room facility. Mentioned as the host guest's organization and as the hidden hub of the chip roadmap.

Latré: "We're the worldwide leader in chips But in actually the early stages of building a chip" [00:00:57]

Latré: "there's almost no chip in the world today That has not been touched by by iMac in some way We already exist for 40 years" [00:01:23]

Also: "all the biggest chip providers in the world actually first come to us first in Belgium To kind of design that's next generation chip" [00:01:23]

Micron

One of three memory producers controlling about 95% of the global memory market. Mentioned as a beneficiary of the memory shortage.

Chambers: "there's only three companies in the world that produce memory Which is 95 of the market micron? sk hynix and samsung" [00:04:52]

SK Hynix

Korean memory manufacturer, one of the three dominant suppliers.

Chambers: "micron? sk hynix and samsung two of them are in korea" [00:04:52]

Samsung

Korean memory manufacturer, one of the three dominant suppliers.

Chambers: "two of them are in korea and a lot of their shift has Been filled out by hyperscalers taking over the their whole inventory" [00:04:52]

Hyperscalers (generic)

Large cloud and AI infrastructure buyers absorbing memory inventory, driving prices up 700% this year.

Chambers: "hyperscalers taking over the their whole inventory So you've seen prices rise by 700 percent this year" [00:05:21]

Kimi (Moonshot AI's Kimi K3 open-source model)

Open-source model cited as having more parameters than the human brain, illustrating how far raw scale has gone versus hardware efficiency.

Latré: "the latest open source models like like Kimi K3 already have more parameters than what the human brain has" [00:07:49]

Strike / Villa Charlotte (Adam Chambers' firm)

Investment firm of Adam Chambers; he invests in bottleneck-attacking companies across power, memory, cooling, and data transfer.

Molly O'Shea: "Adam, you're from Strike / Villa Charlotte" [00:00:30]

Foundries (referenced)

Latré points to foundry buildout as super important for solving the manufacturing bottleneck.

Latré: "what the foundries is doing there Uh, it's super important" [00:06:54]

Sourcery (host media platform)

The podcast and newsletter, with weekly top deals and tech headlines.

Molly O'Shea: "check out our newsletter sorcery.vc where we deliver a once a week top deals and tech headlines email" [00:13:02]

4. People Identified

Steven Latré

IMEC executive whose daily job is forecasting what AI looks like 5 to 10 years out. Notable for the long-horizon view that LLMs are a brute-force phase.

Latré: "every day what I do actually at imec is thinking about What could be ai in the future? So how is ai going to evolve in like five to ten years from now?" [00:10:03]

Investor at Strike / Villa Charlotte who maps chip investing by bottleneck. Distinctive for a bottleneck-first lens and a specific Europe-plus-US founder heuristic.

Chambers: "The bottlenecks that I look at the most... sit in power and energy memory cooling and data transfer" [00:03:09]

Molly O'Shea

Host of Sourcery. Notable for referencing the "hardware era renaissance" framing and margin discussion.

O'Shea: "the margins for chips are fantastic and and for hardware versus software Which is getting eaten by ai" [00:04:16]

Tony Kim (BlackRock)

Interviewed by Molly at the RAISE AI summit in Paris, cited for the "hardware era renaissance" view.

O'Shea: "I recently interviewed tony kim of black rock at the raise ai summit. It's actually in Paris and We're talking about hardware. It's like the hardware era renaissance" [00:04:16]

Louie (Sourcery team)

Mentioned as coming up next to speak after the interview.

O'Shea: "I think louie's gonna come up next to say something" [00:13:02]

5. Operating Insights

Map Your Investment or Product Thesis to the Binding Constraint

Chambers's framework is operationally reusable: instead of asking "what is hot," ask "what physically limits the system and who removes that limit." He names four constraints (power, memory, cooling, data transfer) and only looks at companies attacking them.

Chambers: "The bottlenecks that I look at the most and try and Look at companies that attack those bottlenecks sit in power and energy memory cooling and data transfer" [00:03:09]

Hire and Back the "Deep Technical Europe + Commercial US" Hybrid

Chambers offers a concrete recruiting and investing filter: technical depth from a European academic background combined with U.S. commercial instincts.

Chambers: "a lot of the best founders that i've backed and i've seen Have that dichotomy between the two Um, so i'd say both for recruiting and investing" [00:12:29]

Plan on a 5-to-10-Year Hardware Lead Time

Latré's description of the hardware cycle is an operating constraint for anyone building in the space: unlike software, chips take years from concept to product, so roadmap decisions must be made a decade ahead.

Latré: "hardware is not like software It's not that you're prompted and suddenly 30 seconds later something comes out of it No, it's a process of five to ten years" [00:00:57]

Look for Technologies That Compound Across Multiple Bottlenecks

The photonics discussion shows a selection filter: prefer a technology that lowers energy, cooling, and data-transfer costs at once, because its value stacks.

Chambers: "spanning Most of the bottlenecks that we're seeing in the aion for build out right now It's like the key technology that is likely to win" [00:11:04]

6. Overlooked Insights

The Human-Brain Efficiency Gap Is a Roadmap for Where Opportunity Lies

Latré drops in passing that the hardware needed to run a frontier model is roughly a million times less energy efficient than the human brain. That single number frames a multi-order-of-magnitude efficiency opportunity, and the brain's 3D structure and data movement ability is the design cue he cites. Whoever closes even a fraction of that gap captures enormous value.

Latré: "that chip that we need To fuel that that K3 model is about a million times more or less energy efficient than what the human brain brain is" [00:07:49]

And on the design cue: "our human brain is also unbelievably good at We have an unbelievable 3d structure that is able to to do this" [00:10:26]

Foundry and Component Geography Is the Hidden Strategic Variable

Latré mentions almost offhand that manufacturing the right components "in the different regions" is the true bottleneck, and points to foundries expanding across regions. This hints that regional supply-chain positioning (and not chip design alone) is where scarcity and strategic value concentrate, which could favor equipment, materials, and packaging suppliers over designers.

Latré: "Manufacturing the right type of components In the in different regions and I think for example what the foundries is doing there Uh, it's super important" [00:06:54]