Dharmash Mistry @Dharmash

Morgan Stanley just mapped out the entire AI infrastructure supply chain, and it reveals who actually gets paid at every layer of the trillion dollar buildout (Save this). This heatmap breaks the AI infrastructure value chain into two dimensions those who owns and operates the data centers at the top and what physical and technical components get built underneath to make those data centers function. At the top sit the owners/operators, the hyperscalers like Meta, Alphabet, Amazon and Microsoft, alongside data center REITs, private equity giants like Blackstone and Brookfield, enterprises and neoclouds including CoreWeave and Nebius. These are the companies writing the massive capex checks that fund everything below them. Below that sits the actual build out, split into seven layers, semi production, processors, server components, servers, network, internal power/cooling and power supply. Semiconductor production is dominated by names your audience already knows well, Nvidia and AMD for GPUs, TSMC adjacent foundries, ASML and Applied Materials for capital equipment, and Micron and SK Hynix under memory/storage. But the less obvious money is in the physical infrastructure layers most retail investors never look at. Server components include passive parts from Yageo and Murata, thermal solutions from Sanyo Denki, and PCB substrates from companies like Unimicron. Network infrastructure includes InfiniBand and Ethernet gear from Nvidia and Arista, plus optical/DCI routing from Cisco and Ciena. Internal power and cooling is arguably the most underappreciated category here. It includes liquid cooling specialists like Vertiv and CoolIT, power electronics from Siemens and Eaton, and uninterruptible power supply makers like ABB and Legrand, all companies solving the literal heat and electricity problem created by cramming more GPUs into less space. So who benefits from all of this? Everyone in every box benefits in some way but the real insight is that value doesn't concentrate at just the GPU layer anymore. The hyperscalers at the top are distributing capex across seven distinct physical layers which means the picks and shovels opportunity set has expanded well beyond Nvidia into cooling, grid infrastructure, and power generation. Milk Road Pro is tracking each one of these layers, come join us for just a dollar using the link below!

The board exercise, once a year: With £100m, frontier models, no legacy and no sacred cows — which of the four routes would we use to attack ourselves first? Full framework (the Six Ps) in this week's Director Brief.

Same mechanism is now moving from information to transactions. What does agentic commerce do to the consumer interface for retail banks, insurance, travel, consumer goods?

Three companies your board probably hasn't discussed — and the four ways AI-natives attack. Every challenger picks one route in: reinvent the product, the workflow, the economics, or the distribution.

Squads of the last 4 teams (and top 4 ranked by FIFA) - Spain, Argentina, France, England - the Premier league leads with 41/ 104 (39%) - @premierleague 41 - LaLiga 29 - Ligue 1 13 - Series A 8 - Bundesliga 6

Congratulations to Spain, fully deserved #FIFA2026 world cup winners. Looking at this another way - medals table by club (gold, silver, bronze) - shows the concentration within the squad (familiarity & talent ecosystem - playing like a club side?) @FIFAWorldCup

thedirectorbrief.com/p/issue-003-so…

The Six Ps of AI strategy, the one-page board AI dashboard, and the £100m clean-sheet test. link below

Only 5% of companies show real financial gain from AI. That 5% posts 4x the shareholder return of everyone else.

Kimi K3 may be an important inflection point for AI. Potentially negative for Anthropic and OpenAI while being net positive for essentially every other company in the world. I mean that very literally. Although the real “Sputnik moment” would be an open-source frontier model that was also token efficient unlike Kimi K3 which is 50-70% more expensive to run than GPT 5.6 per Artificial Analysis. Rationale:   A world where there are only 2-3 dominant frontier labs with 90% inference margins is net negative for every other layer while being awesome for those 2-3 labs. Those labs would become monopsonies for power, data centers, semiconductors and hyperscalers and would obviously vertically integrate over time into all those layers while also completely subsuming the application/software layers.    Anything that lowers margins and increases competition at the model layer is good for every other AI layer: power, semiconductors, hyperscalers, neoclouds and yes even software.   This is why Jensen is so supportive of open-source. An open-source model requires the *exact* same amount of compute to run as a closed frontier model of similar size and architecture. Kimi K3 is roughly the same price as GPT 5.6 Terra on a per token basis, which actually suggests that it is less computationally efficient as I am sure that GPT 5.6 is priced to a higher margin than K3. And given that K3 is a token wastrel, i.e. token inefficient, it is significantly more expensive per task than GPT 5.6 and Grok 4.5, which are much more token efficient. Cost per token and token efficiency (i.e. intelligence density per token) are the drivers of intelligence per unit of cost. The winning AI companies will be those that offer the most intelligence per $ over time.   Lower margin % at the model layer = more margin $ at every part of the infrastructure layer and is a godsend for software. This can happen either through open-source models like K3 at the frontier *or* having a vertically integrated model company like Meta, SpaceX or Google at the frontier. Both outcomes result in a lower margin % at the model layer as vertically integrated model companies don’t really care where the margin $ come from. This is why it was so painful for OpenAI and Anthropic when Google was right there with them from a model competitiveness perspective and why Grok 4.5 and Muse 1.1 were just as important as Kimi K3. 
The reason Kimi K3 is only *potentially* negative for Anthropic and OpenAI is 1) the @ericvishria point that the Claude and ChatGPT products and harnesses may be more important than their models today and 2) the hypothesis that they have much more advanced model checkpoints internally that are already being used for RSI. In the latter scenario, reaching RSI even a few months ahead of other labs might be enough to cement a permanent lead. Time will tell on both points. And likely fairly quickly. Caveat would be that since Kimi K3 is not token efficient and thereby actually more expensive than ChatGPT 5.6, we may need to see a more token efficient open-source model at the frontier or see Grok 5/Composer 4/Muse 2 at multiple points on the Pareto frontier for this potential risk to Anthropic and OpenAI to play out. And I am sure they will both vertically integrate as quickly as possible while continuing the product/harness strength they have shown over the last 8 months.

keyboard_arrow_left Previous keyboard_arrow_right Next

4/Full six-week programme — free, published in full: thedirectorbrief.com/p/the-director… Try Week 1 before your next board meeting. #AIGovernance @The_IoD @NACD

3/The result: 6 weeks, 30 min/week, zero coding, zero company data anywhere near it. By week 6 — a scheduled AI brief, a saved skill, a vibe-coded tool, sharper board questions.

1/Who in this boardroom has used an AI tool to build something consequential in the last 30 days?