The $5T AI Opportunity🤖, The Only…

1. Key Themes

Theme: AI capital is concentrating in infrastructure, starving applications, and physical bottlenecks are the real constraint

Capital is flowing to infrastructure while strong application startups go unfunded

Investment is skewed toward the picks-and-shovels layer even as application companies post better fundamentals.

"Investment is concentrating in AI infrastructure while application startups with strong growth, retention, and margins struggle to secure funding."

The deal list reflects this: CScale ($145M, optical interconnect for "gigawatt-scale AI infrastructure"), Volantis ($88M, photonic inference architecture), and Efficient Computer ($97M+, energy-efficient processors) all raised sizable rounds on the compute and hardware layer.

The binding constraints on AI are electricity, capital allocation, and execution, not model quality

"AI progress faces constraints beyond model quality, including transformer grid-connection waits of up to five years, reliability falling to 36% across twenty steps at 95% per-step accuracy, and more than $800 billion in circular financing."

The newsletter ties these to "three underlying limits: electricity, capital allocation, and organizational execution."

Frontier-lab economics rest on enormous fixed commitments

"Anthropic's July 2026 run rate exceeded $65 billion versus nearly $4.6 billion in 2025 revenue, while about 80% of infrastructure commitments remain noncancelable or usage-independent."

"A $2 trillion valuation implies nearly 31x the July run rate, putting sustained free cash flow at the center of the valuation case."

Theme: AI as a margin-expansion tool for traditional businesses (roll-ups)

Boomer-era business turnover creates a roll-up window

"An estimated $5 trillion in US businesses could change hands by 2035, creating an opportunity to acquire service firms and use AI agents to raise margins from 5 to 10% to 30 to 40%."

The playbook covers "sourcing, diligence, work samples, and a 100-day integration plan, with execution risks including staff departures and scaling acquisitions faster than integration."

Theme: Fund formation is specializing into physical-world, defense, and deep tech

New funds cluster around "atoms" rather than pure software

"Tech is leading earnings growth and margin expansion, while trading at multiples roughly 20% below its 5-year average." (a16z, "Rotation to Atoms")

Recently closed funds mirror this: Zero Infinity Partners ($156M, "intersection of physical infrastructure and technology"), Protego Ventures ($125M, "Israeli defense technology"), and deals like Varda Space Industries ($251M, "space-based manufacturing") and Reverion ($175M, "carbon-negative power plants").

Theme: VC fund performance and capital concentration among large funds

Big funds absorb capital while headline IRRs decay

"For 2017 to 2019 vintages, top-decile IRRs declined while TVPIs increased, with the 2017 vintage moving from 28.7% to 25.7% IRR and 3.31x to 4.14x TVPI. Funds above $100 million captured 64% of 2025 capital."

"2017 median DPI remained at 0.37x."

Paper gains are rising, but realized returns (DPI) lag.

2. Contrarian Perspectives

Don't raise venture capital just because competitors, investors, or hype say you should

Against the prevailing fundraise-when-you-can culture, the newsletter argues for a strict test.

"Raise only when capital is the primary constraint on growth, since fundraising consumes founder time and dilutes equity and control. A competitor's funding, investor pressure, stalled growth, or market hype are not substitutes for a clear capital need."

Evidence: fundraising's costs (founder time, dilution, control) are certain, while the benefit is only real when capital is the actual bottleneck.

AI's constraint is not the model, it is physics, reliability math, and circular finance

Contrary to the consensus focus on model capability, the newsletter points to compounding error and infrastructure waits.

"reliability falling to 36% across twenty steps at 95% per-step accuracy"

The arithmetic (0.95^20 ≈ 36%) shows why long agentic workflows fail even with "good" models, and "transformer grid-connection waits of up to five years" cap deployment speed regardless of capital.

Investors and application startups should avoid chasing the infrastructure crowd

"infrastructure companies can raise and offer secondaries, application startups can prioritize survival, and investors can avoid duplicating the same exposure."

The contrarian read is that strong-fundamentals application companies are being overlooked, and portfolio duplication of infrastructure exposure is a hidden risk.

3. Companies Identified

Anthropic

Description: Frontier AI lab.
Why mentioned: Case study in revenue growth versus fixed commitments and valuation math.
Quote: "Anthropic's July 2026 run rate exceeded $65 billion versus nearly $4.6 billion in 2025 revenue, while about 80% of infrastructure commitments remain noncancelable or usage-independent."

Figma

Description: Design software company.
Why mentioned: Example of supervoting shares separating control from ownership.
Quote: "allowing Figma's CEO to hold roughly 9% of equity while controlling just over half the votes."

Jev

Description: Low-cost inference service for classification/routing decisions.
Why mentioned: Addresses overspending on frontier models for simple agent decisions.
Quote: "Jev, launched September 15, returns choices, scores, or probabilities at $0.042 per million input tokens with no output charge."

EliseAI

Description: AI platform for housing and healthcare operations.
Why mentioned: Largest hot deal.
Quote: "raised $350M in Series F funding at a $4B valuation, co-led by Andreessen Horowitz and Bessemer Venture Partners."

General Intuition

Description: Foundation models for physical AI.
Why mentioned: Highest-valuation round listed.
Quote: "raised $220M in new funding at a $6.2B valuation... to develop foundation models for real-world and physical AI applications."

Armadin

Description: Agentic cybersecurity platform.
Why mentioned: Large Series B backed by top-tier investors.
Quote: "raised $255.5M in Series B funding, co-led by Andreessen Horowitz and Accel, at a valuation above $2.5B."

Varda Space Industries

Description: Space-based manufacturing.
Why mentioned: Series D in the physical-economy rotation.
Quote: "raised $251M in Series D funding, led by Lux Capital and Natural Capital."

CScale

Description: Optical interconnect for AI datacenters.
Why mentioned: Infrastructure-layer capital concentration.
Quote: "to commercialize optical interconnect technology for gigawatt-scale AI infrastructure."

Volantis

Description: Photonic AI-inference architecture.
Why mentioned: Addressing memory and bandwidth limits.
Quote: "to commercialize its photonic AI-inference architecture designed to overcome memory and bandwidth constraints."

Efficient Computer

Description: Energy-efficient processor company.
Why mentioned: Moving from physical AI to datacenters.
Quote: "raised more than $97M in Series B funding, led by TQ Ventures, at a $650M valuation."

Reverion

Description: Carbon-negative power plants.
Why mentioned: Energy/physical-world raise.
Quote: "raised $175M in Series B funding, led by Kembara."

Jeeves

Description: Stablecoin-native financial infrastructure.
Why mentioned: Stablecoin volume growth.
Quote: "raised $110M in equity funding, led by CoinFund... as platform and stablecoin volumes accelerate."

Flow Engineering

Description: AI-powered hardware engineering and validation.
Why mentioned: AI applied to hardware design.
Quote: "raised $50M in Series B funding at a $750M valuation."

Miter

Description: AI-native construction workforce and operations platform.
Why mentioned: Vertical AI in construction.
Quote: "raised $40M in Series B funding, led by Battery Ventures."

Attio

Description: CRM built for the agentic era (sponsor).
Why mentioned: Sponsor; customers include Parallel, Turbopuffer, and Wordsmith.
Quote: "Teams like Parallel, Turbopuffer and Wordsmith already run on it."

Biotech/medtech raises (TriGemX Bio, AnaCardio, ai3Bio, Aptadir Therapeutics)

Description: Therapeutics companies.
Why mentioned: Continued healthcare funding.
Quote: "TriGemX Bio raised $94M... to advance elismetrep, its Phase 3-stage migraine treatment."

4. People Identified

Aaron Harris

Description: Author of the fundraising-discipline piece.
Why mentioned: Source of the "only reason to raise VC" argument.
Quote: "Raise only when capital is the primary constraint on growth."

Greg Isenberg

Description: Author of the AI roll-up playbook.
Why mentioned: Source of the $5T roll-up thesis.
Quote: "An estimated $5 trillion in US businesses could change hands by 2035."

Ramy Adeeb

Description: Author of "The Great AI Capital Misallocation."
Why mentioned: Frames infrastructure-versus-application funding imbalance.
Quote: "Investment is concentrating in AI infrastructure while application startups with strong growth, retention, and margins struggle to secure funding."

Ilya Strebulaev

Description: Author on supervoting and founder control.
Why mentioned: Explains separating ownership from control.
Quote: "Supervoting shares separate voting power from economic ownership."

Ruben Dominguez

Description: Newsletter author and curator.
Why mentioned: Publisher of the roundup and founder resources.
Quote: "Another week, another pulse check on Tech."

5. Operating Insights

Match your sales motion to market structure

"The right sales strategy depends on who signs the contract, how much risk that buyer carries, and whether evidence of success spreads across the market. Concentrated markets favor winning a marquee customer through the lighthouse play, while fragmented markets favor the landgrab play built around repeatable unit economics."

Route simple agent decisions to cheap classifiers, not frontier models

"Many agent decisions have only a few possible outcomes but still use expensive frontier models for simple classification and routing tasks."

Audit agent workflows for classification and routing calls that can be moved to low-cost inference ($0.042 per million input tokens in Jev's case).

Make GTM a learning system, not a fixed playbook

"Your GTM can't run on fixed sequences and static playbooks. It has to learn from every account, interaction and outcome, and adjust what happens next on its own."

6. Overlooked Insights

Supervoting has limits that founders often miss

"The structure does not guarantee board seats or bypass protective provisions, while voting proxies offer founders a narrower alternative to enhanced voting rights."

Control via votes is not the same as board or veto control, and proxies are a lighter-weight option.

Fund-size concentration and LP-base thinning

"Funds above $100 million captured 64% of 2025 capital, while median LP counts fell for $100 million to $250 million funds."

Mid-sized funds are seeing shrinking LP bases, suggesting a squeeze in the middle of the market.