Davidovs
Venture
Collective

DVC's
STATE OF AI

The Operating Manual for the AI Revolution

The biggest industrial shift in modern history is already underway, and if you cannot see the full stack, you cannot see where the world is heading.

Scroll to explore
davidovs.com

FOLLOW THE MONEY

How $60B of AI revenue generates $700B of infrastructure investment

MARGIN GRADIENT
25–60%Apps
33–70%Models
14–47%Cloud
50–75%Silicon
UtilityEnergy
Where does $1 of AI revenue go?
$0.30–0.60 Inference costs (AI Tax)
$0.18–0.36 Cloud compute (~60% of inference)
$0.11–0.25 GPU/server purchases (60–70% of CapEx)
$0.06–0.19 Silicon margin (NVIDIA ~50–75% GM)
$0.02–0.04 Energy (power + cooling)
Click any layer to explore
LAYER 5

APPLICATIONS

$12-15B ARR GM: 25–60%
AI Tax: 30–75% consumed by inference
OpenAI $25B L5+L4
Anthropic $19B L5+L4
Cursor $2B+
Salesforce AI $800M
Grammarly $700M
GitHub Copilot $500M+
Lovable $400M
ElevenLabs $330M L5+L4
Perplexity ~$420M DVC
Higgsfield ~$300M DVC
Suno $300M
Runway ~$150M L5+L4
Replit $253M
Traditional SaaS70–85% GMBenchmark
Perplexity50%+ GMSearch + subscription
OpenAI~42% GMH1 2025, improving
Cursor~35% GMHeavy inference load
Higgsfield30%+ GMVideo generation
Lovable20–40% GMVibe-coding, scaling fast
GitHub Copilot~0–15% GMSubsidized for lock-in
$0.30–0.60 → inference
LAYER 4

FOUNDATION MODELS

$8-10B API GM: 33–70%

API Market Share (Menlo 2025)

Anthropic40%
OpenAI27%
Google21%
Others12%
OpenAIGM ~42%-$6.9B H1 loss
AnthropicGM ~40%Lowered from 50%
Google GeminiNot disclosedServing costs -78%
CohereGM ~70%Best in class
xAINet loss 13.6x rev
~60% → cloud compute
LAYER 3

CLOUD & INFRA

$18-22B rev $224B CapEx 23:1 CapEx-to-AI-Revenue
AWS37.0% OM$39.8B / $107.6B
Microsoft IC47.1% OM$49.6B / $105.4BCloud GM 71%
Google Cloud14.1% OM$6.1B / $43.2B→ 17.5%
Meta42.0% OM(total company)
CoreWeave72% GM-0.9% OM$5.1B rev
Nebius$68M rev$17.4B MSFT contract

GPU Server Economics ($300K server, $150K annual rev, $45K cash opex)

Depreciation$75K
Profit$30K
Margin20.0%
Breakeven Util.~80%

Hyperscaler Depreciation Changes

Microsoft4yr → 6yrSaved $3.7B
Alphabet4yr → 6yrSaved $3.9B
Amazon4→5→6yr, reversed subset to 5yr+$889M hit
Meta→ 5.5yrSaving ~$2.9B
CoreWeave5yr → 6yrSaved $20M

CapEx Breakdown

GPUs/Servers60–70% (~$350B)
Buildings/Power25–35% (~$170B)
Networking5–10% (~$40B)
2024: $224B 2025: $379B 2026E: $700B
60–70% of CapEx → GPUs
LAYER 2

SILICON

$130B rev GM: 50–75%
NVIDIA $215.9B FY26
GM 71.1% OP 60.4% Net 55.6%
DC rev $193.7B $1T+ orders through 2027
Broadcom $63.9B FY25
GM 67.8% OP 39.9% EBITDA 68%
AMD $34.6B FY25
GM 49.5% OP 10.7%
DC $16.6B
Arista $9.0B FY25
GM 63.7% OP 42.5% Net 39.0%
Marvell $8.2B FY26
GM 51.0% OP 16.1% Net 32.6%

Custom Silicon

Google TPU AWS Trainium Meta MTIA Microsoft Maia Etched (DVC)

NVIDIA INFERENCE ARCHITECTURE — GTC 2026

Vera Rubin + Groq LPU

72 Rubin servers + 256 LPU chips = 700M tokens/sec — 350× Hopper throughput. Purpose-built for inference.

NVIDIA Roadmap

Blackwell (now) → Vera Rubin (2026, HBM4) → Feynman (2028, TSMC 1.6nm, silicon photonics)

Source: NVIDIA GTC 2026 / WSJ / CNBC / Axios, Mar 2026
580 TWh by 2028
LAYER 1

ENERGY

460 TWh
Nuclear
🔥Gas
Solar

Who's Powering AI

Microsoft Three Mile Island restart 835 MW, 20-yr PPA with Constellation, $16B, online 2028
Google SMR fleet + plant restart 500 MW Kairos Power SMR fleet (first reactor 2030); $1.6B Duane Arnold restart with NextEra; 1.8 GW pipeline via Elementl Power
Amazon $20B+ nuclear investment 960 MW Susquehanna campus; X-energy SMR development; 1,920 MW Talen PPA through 2042; Energy Northwest 960 MW SMR project
Meta 1–4 GW nuclear RFP Request for proposals targeting new nuclear generation, early 2030s; 6.6 GW total nuclear projects announced Jan 2026
Oracle 3× SMR-powered data center GW-scale campus powered by three SMRs; building permits secured
Big tech contracted 10 GW+ of new US nuclear capacity in 2025. Data center electricity: 460 TWh (2024) → 1,000 TWh (2030) → 1,300 TWh (2035).
Scroll to explore

THE $60B EXPLOSION

~$35B flows to inference & infra 30–75% of app revenue is consumed by compute
~$5–8B pure application layer Excluding OpenAI, Anthropic, Suno, Runway, ElevenLabs
~$1.1B Suno + Runway + ElevenLabs Generative media: music, video, voice

The application layer is the fastest-growing segment of AI revenue.

Historically, in every major computing wave — mainframes, PCs, mobile, cloud — the application layer is where most value ultimately accrues. Infrastructure enables, but applications capture.

Source: company disclosures, press reports, DVC analysis. “Pure application layer” = companies with own distribution, excluding model providers and generative media.

HOW FAST IS $60B?

AI Apps
~3 yrs
Cloud SaaS
~8 yrs
Mobile Apps
~10 yrs
Enterprise SW
~15 yrs

Time for each software category to reach ~$60B in combined application revenue

Source: Bessemer Cloud Index, Menlo Ventures, DVC analysis

THE APPLICATION LAYER

Top AI application companies by annualized recurring revenue

OpenAI $25B ● mixed
Claude $30B ● mixed
Cursor $2B+ ● end-user
GitHub Copilot $2B+ ● end-user
Salesforce AI $800M ● enterprise
Grammarly $700M ● end-user
Lovable $400M ● end-user
ElevenLabs $330M ● API-heavy
Perplexity ~$420M DVC ● end-user
Higgsfield ~$300M DVC ● end-user
Suno $300M ● end-user
Runway ~$300M ● mixed
Replit $253M ● end-user

API-heavy = Most revenue from API/platform  |  End-user = Most revenue from consumers/enterprise seats

Capital is flooding every layer, but one battleground shapes the rest. To understand power in AI, you have to understand the fight over the models themselves.

THE MODEL WARS

10+ companies at the frontier simultaneously. The bottom is commoditizing; the top is getting more expensive and more strategic. Open-source models now match proprietary ones on most benchmarks.

PROPRIETARY FRONTIER

OpenAI

GPT-5.4 / o3

$25B ARR

Anthropic

Opus 4.6 / Mythos

#1 Chatbot Arena

Google

Gemini 3.1 Pro

#2 Chatbot Arena

xAI (SpaceX)

Grok 3

$1.25T merged

Amazon

Nova 2 Pro

Top reasoning on Bedrock

Meta

Muse Spark

Closed · 3.6B DAU

OPEN SOURCE / OPEN WEIGHT 🔓

Meta

Llama 4 Maverick

Community License

DeepSeek

V3.2 / R1

Open Weight · MIT

Qwen (Alibaba)

Qwen 3.5 397B

Apache 2.0

Zhipu AI

GLM-5 744B

#1 OS leaderboard

Moonshot

Kimi K2.5 1T

99.0 HumanEval

Mistral

Mistral Large 2

EU Sovereign AI

NVIDIA

Nemotron 3

Open Agentic

Google

Gemma 4

Apache 2.0

OpenAI

GPT-oss 120B

Apache 2.0

StepFun

Step-3.5-Flash

97.3 AIME

MiniMax

M2.5 230B

80.2 SWE-bench

Source: LLM Stats, Chatbot Arena, Open LLM Leaderboard, company announcements — Mar 2026

THE FOUR SCALING LAWS OF AI — Jensen Huang, Lex Fridman Podcast #494 (Mar 2026)

1. PRE-TRAINING

Bigger models + more data + more compute = smarter AI. The original scaling law.

2. POST-TRAINING

Synthetic data, RLHF, fine-tuning, distillation. “We are no longer limited by data — we are limited by compute.”

3. TEST-TIME (REASONING)

100x+ compute at inference for multi-step reasoning. “Inference is thinking, and thinking is hard.”

4. AGENTIC SCALING

Agents spawn sub-agents, use tools, create data. “It’s like multiplying AI. We could spin off agents as fast as you want.”

“Intelligence is going to scale by one thing, and that’s compute.”

Source: Lex Fridman Podcast #494, NVIDIA GTC 2025–2026

THE NEOLABS: TALENT EXODUS

Billion-dollar bets on people and contrarian theses. Zero revenue, zero products.

TMThinking Machines Lab
Mira Murati · ex-OpenAI CTO $50B val · $2B seed (largest ever)

Built ChatGPT, DALL-E, voice mode. Now building multimodal agentic AI.

SSISafe Superintelligence
Ilya Sutskever · ex-OpenAI Chief Scientist $32B val · $3B raised

One mission: safe superintelligence. No products, no distractions.

AMAMI Labs
Yann LeCun · Turing Award, ex-Meta $3.5B val · $1.03B raised

LLMs hit a wall. Building world models that learn from reality, not language.

h&humans&
Ex-Anthropic/xAI/Google team $4.48B val · $480M seed

AI as connective tissue for human collaboration.

IIIneffable Intelligence
David Silver · created AlphaGo ~$4B val · ~$1B raising

Novel RL for superintelligence. Three months old.

ReReflection AI
Misha Laskin & Ioannis Antonoglou · AlphaGo creators, ex-DeepMind $8B val · $2B raised

Open frontier lab. Western answer to DeepSeek. No model shipped yet.

GfGoodfire
Ex-OpenAI/DeepMind/Stanford $1.25B val · $209M raised

Opening the black box. AI interpretability.

WLWorld Labs
Fei-Fei Li · Stanford, “Godmother of AI” $5B val (talks) · $1B+ raised

Spatial intelligence. 3D world models from images. ~30 people.

$11B+ raised · Zero revenue · Zero products · The talent exodus is the bet

Caveat: Silicon Valley has been here before. Massive pre-product rounds sometimes build category-defining companies — and sometimes they don’t. Thinking Machines Lab lost its CTO and cofounders back to OpenAI within six months of its $2B seed. H Company (ex-DeepMind, $220M seed) lost 3 of 5 cofounders to “operational differences.” SSI’s Daniel Gross left for Meta. xAI lost all 11 cofounders by March 2026. The talent that makes these bets valuable is also the talent most likely to leave. The bet is real. So is the risk.

Source: TechCrunch, Bloomberg, WSJ, Wired, Reuters, Inc. Magazine — 2025–2026

BEYOND TEXT: THE SPECIALIZED MODEL FRONTIER

Every modality now has its own model race. The frontier isn’t just LLMs anymore.

VIDEO
Veo 3.1 Google Gen-4.5 Runway Kling 3.0 Kuaishou Hailuo 2.3 MiniMax Pika 2.5 Luma Ray2 HunyuanVideo Tencent Higgsfield DVC Wan2.1 Alibaba Sora 2 OpenAI
VOICE / TTS
ElevenLabs v3 Fish Audio S1 OpenAI TTS gpt-4o-mini Sesame CSM Kokoro-82M Cartesia Sonic 3 LMNT Inworld
IMAGE
Midjourney v8 Flux 2 BFL Imagen 4 Google Nano Banana Google Seedream 5.0 ByteDance Ideogram 3.0 Firefly 5 Adobe Recraft V4 GPT Image 1.5 OpenAI Grok Imagine xAI
3D
Rodin Meshy Tripo TRELLIS.2 Microsoft CSM → Google Luma Genie Hunyuan3D 3.0 Tencent
MUSIC
Suno v5 Udio Lyria 3 Pro Google AIVA Stable Audio 2.5
WORLD MODELS
Genie 3 DeepMind Decart Marble World Labs Odyssey 2
CODE
Claude Code Anthropic Cursor Windsurf GitHub Copilot Lovable Bolt.new Devin Cognition Codex OpenAI Jules Google Replit Agent
Source: Artificial Analysis, TTS-Arena2, VBench, company announcements — Mar 2026

Benchmarks are dead. Meta admitted Llama 4 was tuned specifically to score well on benchmarks — prompting a credibility crisis across the leaderboard. Models are now optimized for benchmarks rather than tested by them. The industry needs new evaluation methods: real-world task completion, user preference studies, and domain-specific assessments.

Source: Meta Llama 4 controversy (TechCrunch, Apr 2025), Scale AI SEAL benchmark initiative

THE COST COLLAPSE

Frontier intelligence is getting radically cheaper. The cost per million tokens for GPT-4-class performance has fallen from $37.50 in 2023 to $0.14 in 2025 — a 99.6% decline in two years.

270× cheaper at GPT-4 benchmark level $37.50 → $0.14 per 1M tokens
500–900× cheaper for specific reasoning benchmarks Epoch AI: median 50×/year decline rate
$0.006 cheapest frontier-equivalent (DeepSeek V3) 6,000× cheaper than 2023 GPT-4

Take last year’s frontier model: a model that performs similarly on benchmarks is now 500–700× cheaper to run. This collapse in inference cost is what enables the application layer explosion above — and why agentic workflows (which require 10–100× more tokens) are suddenly economically viable.

Source: Epoch AI (Mar 2025), a16z price index, OpenAI/Anthropic/DeepSeek pricing pages, DVC analysis

Reasoning Model Usage Share

Source: DVC analysis based on Menlo Ventures State of Gen AI (2025), Chatbot Arena ELO data, API pricing benchmarks, LLM Stats leaderboard (Mar 2026)

Base generation is commoditizing. Value migrates to orchestration, inference optimization, and proprietary data.

As raw intelligence gets cheaper, the center of gravity shifts. Value moves from generating answers to getting work done.

THE AGENT REVOLUTION

AI is moving from a system you consult to a system that acts. That changes software from a tool for humans to a layer of labor that can execute across workflows. The agent ecosystem alone has already created 67,000+ engineering openings globally — more than at any point in three years.

THE AGENT LANDSCAPE

P
DVC PORTFOLIO

Perplexity Computer

~$420M ARR

Answer engine → agentic platform

+

Transforms a spare Mac mini into an always-on AI agent that controls apps, browses the web, manages files. 19-model orchestration. Personal Computer product launched Feb 2026 — the first consumer device-as-agent play from a search company.

Source: Perplexity / The Verge, 2026
A

Claude Code

$2.5B+ run-rate

9-month ramp to billion-dollar product

+

GA May 2025 → $2.5B run-rate by Feb 2026. Weekly active users doubled since Jan 2026. Business subscriptions quadrupled. Terminal-first agentic coding drove Anthropic to $380B valuation and $30B Series G.

Source: Anthropic Series G announcement, 2026
M

Manus → Meta

Acquired by Meta — $2B+

General-purpose agent → Meta's AI backbone

+

Founded in China, moved to Singapore. $100M ARR in 8 months, $125M+ run-rate at acquisition. 147T tokens processed, 80M+ virtual computers. Microsoft tested in Windows 11. Meta acquiring to power agents across Facebook, Instagram, Meta AI. China investigating potential export law violations. Continues operating independently from Singapore.

Source: CNBC / WSJ / AP News, Dec 2025
C

Cursor

$2B+ ARR

Revenue doubled in 3 months

+

Fastest SaaS growth curve in history. ~60% revenue now from enterprise (was individual-first). 50,000+ enterprises, 100M+ lines of enterprise code per day. $500M ARR Jun 2025 → $1B late 2025 → $2B Feb 2026.

Source: Bloomberg / TechCrunch, 2026
D

Devin

$10.2B valuation

67% PR merge rate — autonomous SWE

+

First fully autonomous software engineer. Devin 2.0 handles async multi-step tasks: reads codebase, plans approach, writes code, runs tests, submits PRs. Moving from code completion paradigm to autonomous project execution.

Source: Cognition, 2026
OC

OpenClaw

"As big as HTML, as big as Linux"

The OS for personal AI — Jensen, GTC 2026

+

Jensen Huang (GTC 2026): "Every company needs an OpenClaw strategy. This is as big as HTML, as big as Linux." Called it the fastest-growing open-source project in history, surpassing Linux's early adoption pace. Open-source personal AI agent running on Mac mini ($599), RTX PCs, DGX Spark, DGX Station, or cloud VPS (~$30/mo).

Source: NVIDIA GTC 2026 keynote / Business Insider / TechRadar, Mar 2026
NC

NemoClaw

Enterprise AI Agent Layer

NVIDIA's enterprise wrapper for OpenClaw

+

NemoClaw (NVIDIA, GTC 2026): Enterprise security layer — network guardrails, privacy router, sandboxed execution via OpenShell. Installs with a single command. Adds Nemotron models + Dynamo inference engine. Jensen's pitch: "OpenClaw for everyone, NemoClaw for the enterprise."

Source: NVIDIA GTC 2026 keynote / TechRadar, Mar 2026

THE INFERENCE INFLECTION POINT

Jensen Huang declared the arrival of the "inference inflection point" at GTC 2026: two exponentials colliding — demand for inference growing exponentially while cost per token falls exponentially. The question is no longer whether agents can work. It's whether the business models can sustain them.

We unpack the full business model problem — pricing paradigms, margin squeeze, and why the economics are still unresolved — later in the presentation.

Source: NVIDIA GTC 2026 keynote / Axios / CNBC, Mar 2026
"Ability to make software will be a human right soon, and it's not going to feel like making software."
— The Vibe Coding Thesis

"VIBE CODING" ERA

DVC Wabi $20M pre-seed DVC portfolio, 2026
Lovable $400M ARR Source: TechCrunch, 2026
Replit $150M ARR Source: Replit, 2025
Bolt $40M ARR Source: Bolt, 2025

AppDirect: Non-technical marketing team vibe-coded 200K+ lines of code, built 11 projects with 4 in production, and have 80+ applications in progress across Sales, Finance, HR, and Operations.

Zero-code founder: Built a transcription platform that reached 80,000 users, 1M+ minutes processed, and six-figure ARR — in four months.

Source: Lovable / AppDirect case study, Replit / Whisper AI, 2025
84%

84% of developers are using or planning to use AI tools in development

Source: Stack Overflow Developer Survey, 2025
💻

Agents building agents: 100% of Claude Code is written in Claude Code. 100% of Perplexity Computer is written in Perplexity Computer — as is this presentation.

The nature of code itself is changing. Humans write abstractions — functions, classes, design patterns — primarily so other humans can read and maintain the code. AI does not need that. It can generate and re-generate from scratch faster than it can navigate a complex abstraction hierarchy. Code is becoming a throwaway artifact rather than a maintained asset. 42% of all committed code is now AI-generated (SonarSource 2026). Projected: 65% by 2027.

Source: SonarSource State of Code 2026

But the ceiling is rising faster than the floor. While vibe coding democratizes building, advanced practitioners are diverging fast. Anthropic’s 2026 Agentic Coding report: “Software development is shifting from writing code to orchestrating agents that write code.” Engineers now run multiple AI agents in parallel on one codebase (Vibe Kanban, AutoForge), each on isolated git worktrees, with visual kanban boards for task management. The new SWE job: decompose tasks, spin up agents, review their PRs, resolve merge conflicts. A tech lead managing a team of AI juniors. One company deployed 800+ AI agents internally.

Source: Anthropic 2026 Agentic Coding Trends Report

How fast does open source move? This week, Anthropic accidentally shipped Claude Code’s entire source code in an npm package — 512,000 lines of TypeScript. Within hours, the community had archived it to GitHub (41,500+ forks), rewritten the core in Python, Go, and Rust, and started building on it. The leaked code revealed an “Undercover Mode” that hides the fact AI is contributing to open-source repos — and an unreleased autonomous agent mode called KAIROS.

Meanwhile, OpenClaw — the open-source Claude wrapper with 247K GitHub stars that Jensen Huang compared to Linux — was systematically killed by Anthropic in four weeks: trademark warning, OAuth blocked, features cloned, then “Channels” absorbed the last differentiator. The creator left to join OpenAI.

The moral: proprietary code is a temporary state. The moment it touches the open internet, the community absorbs, rewrites, and surpasses it before you can issue a takedown. This is the same force that made DeepSeek catch up to GPT-4 in months, MCP hit 82K stars, and models commoditize. Open source does not just compete with proprietary anymore — it metabolizes it.

Source: The Register, Hacker News, GitHub, Mar 31, 2026

HOW AGENTS GET DEPLOYED

Public Cloud

ChatGPT Agent, Devin, Manus

Subscription / usage-based

81%

Private Cloud

Internal infra, VPS, hybrid

Higher ops, better governance

52%

Personal System

Cursor, Claude Code, Copilot

Software subscription only

40%

Local Hardware

Perplexity Mac mini, OpenClaw, DGX Spark

$599+ one-time + low variable

15%
FASTEST GROWING

Cloud held 81.1% of agent market share in 2025 — but local-first is the next visible deployment wave

Source: Mordor Intelligence, 2025

WHAT PEOPLE USE AGENTS FOR

Coding / Software Dev
55% of AI spend
$4B of $7.3B dept. spend — Menlo Ventures
Customer Support / Ops
78% adoption
Cloudera Enterprise Survey
Process Automation
71% adoption
Cloudera Enterprise Survey
Research & Analysis
Core use case
ChatGPT Agent, Manus, NVIDIA
Content Creation
51% penetration
Writing tasks — Menlo Consumer
Personal Productivity
19% of adults
Email, to-dos — Menlo Consumer
Home Automation
Emerging
Strong demos, thin market data

THE VULNERABILITY PARADOX

You must make yourself vulnerable to extract value — but that will change.

Supply Chain

Snyk ToxicSkills: 37% of OpenClaw community skills contained flawed code. 200+ GitHub security advisories.

Prompt Injection

EchoLeak attack on browser agents. Slack AI data exfiltration. Gemini memory poisoning demonstrated.

Over-Permission

Agents need files, email, calendar, purchases to be useful. Every permission granted is an attack surface.

The Paradox

Today: accept risk to capture value. Tomorrow: agent-specific security layers, capability-based permissions, cryptographic identity.

Agents now see your screen. Claude Computer Use (Mar 2026) lets Claude see your desktop, launch apps, browse the web, and fill spreadsheets. GPT-5.4 has native computer-use. OpenClaw brought it to open source. Agents no longer need a custom API for every tool — they operate any software through the same interface you do. That is a massive unlock for automating legacy systems that will never get an API.

Source: Anthropic, OpenAI, Mar 2026

ONE AGENT. ONE MAC MINI. ONE HOUSEHOLD.

7:00 AM

School Check

Agent scans kids' school emails. Finds early dismissal — short day today.

7:05 AM

Nanny Alert

Sends iMessage to nanny: "Short day — pickup at 12:30 instead of 3."

9:30 AM

Supply Request

Cleaning lady texts via Telegram: "You're out of garbage bags."

9:31 AM

Auto-Purchase

Agent orders from Amazon using its own account and crypto card. No human needed.

11:00 AM

Climate Control

Checks weather forecast, pre-cools house for afternoon heat via HVAC.

2:00 PM

Kids Home

Adjusts lighting, unlocks door via HomeAssistant, confirms to parent.

4:00 PM

Evening Prep

Reviews tomorrow's calendar, preps grocery list, charges batteries at off-peak rates.

It’s not just tech companies. A roofing company is using AI agents to pull satellite imagery, cross-reference hail damage patterns, and feed warm leads to their sales team. They’re roofers — not engineers, not a startup. When a roofing company runs AI agents, every company runs AI agents.

Source: @RoundtableSpace / Startup Ideas Podcast, Apr 2026
  • OpenClaw on Mac mini M4 ($599)
  • HomeAssistant integration for HVAC, lighting, locks
  • iMessage via AppleScript bridge
  • Telegram Bot API for service providers
  • Amazon purchasing via browser automation
  • Crypto card (Privacy.com / Coinbase wallet) for agent financial autonomy
Source: OpenClaw GitHub / community setups, 2026

CHEAPER AI ≠ LESS SPEND. Cheaper inference = more workflows clear the ROI threshold.

Task completion sounds simple until you look under the hood. What appears to be one product is really a new software stack in disguise.

ANATOMY OF AN AI AGENT

Every agentic product — from Cursor to Harvey to Glean — is built on the same fundamental layers.

$1B+Combined funding across agent primitives
18 moMost of this infrastructure didn't exist before
88%of Fortune 100 use E2B sandboxes

THE 7-LAYER AGENT STACK

Click any layer to explore tools, companies, and key data. Hover any company for details.

1 UI / Frontend Layer How agents meet users
CopilotKit 29.3K ★ 515K npm/mo DVC Vercel AI SDK 22.6K ★ 36.5M npm/mo AG-UI 12.4K ★ 1.21M npm/mo DVC Streamlit Gradio
12.4K GitHub ★ in <1 year — AG-UI is becoming the standard event protocol for agent-to-user interaction.
2 Orchestration Layer Planning, routing, multi-agent coordination
LangGraph 129K ★ 223.6M PyPI/mo CrewAI AutoGen Semantic Kernel Pydantic AI Sixtyfour Kapso
ReAct loops → multi-agent systems with planners, workers, verifiers. Most serious startups eventually build proprietary orchestration.
3 Memory Layer State beyond the prompt window
mem0 49.6K ★ 2.19M PyPI/mo DVC Letta / MemGPT Zep Kite
+26% accuracy over OpenAI Memory, 90% less tokens. mem0 externalizes memory — works with any stack.
4 Tool / Action Layer How agents interact with external systems
MCP 110.8M PyPI/mo 85.6M npm/mo A2A 22.5K ★ 4.36M PyPI/mo Function Calling Browserbase Composio Firecrawl Browser Use Exa Hyperbrowser Sponge Orthogonal
"MCP is becoming the REST of the AI era" — MCP for tools, A2A for agent-to-agent, AG-UI for agent-to-user.
5 Foundation Model Layer The reasoning engine(s)
OpenAI Anthropic Google xAI Meta / Llama DeepSeek Mistral
37% of enterprises use 5+ models in production. Multi-model routing is standard — Harvey uses 6+ providers. Source: a16z
6 Execution Layer Where code actually runs
E2B Daytona WebContainers Modal Dynamo NVIDIA OSS
3 patterns: local, cloud sandbox, browser-native. Cursor = local, Devin = cloud, Bolt = browser. Dynamo (NVIDIA OSS): 30× inference throughput.
7 Eval, Voice & Communication Observability, quality, and agent output
ElevenLabs Vapi LangSmith Phoenix / Arize OpenTelemetry Braintrust AgentPhone AgentMail
Best startups treat eval as product, not afterthought. Harvey: BigLaw Bench. Perplexity: search_evals.
“Stitch all of these primitives together, and what emerges isn’t a chatbot — it’s a digital coworker more human than AI.” — @shivsakhuja
Sources: Crunchbase, TechCrunch, company announcements, GitHub, npm, PyPI · 2025–2026

HOW THEY BUILD DIFFERENTLY

AI Startups

ModelsSingle provider, fast iteration
OrchestrationProprietary, vertically integrated
MemoryProduct-specific (Devin Knowledge, Glean graphs)
Tool AccessBuilt-in connectors
UINovel metaphors (spreadsheet, browser, IDE)
ExecutionCloud sandbox or browser
GovernanceSpeed > compliance
Build vs BuyBuild the whole stack
2024 47% build / 53% buy
2025 24% build / 76% buy
Source: Menlo Ventures

Enterprise Teams

Models5+ providers, routing & failover
OrchestrationOSS frameworks (LangGraph, Semantic Kernel)
Memorymem0 or custom stores, portable
Tool AccessMCP servers + internal APIs
UICopilotKit / AG-UI or internal frontends
ExecutionHybrid: cloud + on-prem + airgapped
GovernanceAudit, approvals, SOC2, HIPAA
Build vs Buy24% build / 76% buy (Menlo 2025)

THE THREE PROTOCOLS

MCP

Agent ↔ Tools / Data

"The REST of AI" — how agents access external tools and data.

110.8M PyPI 85.6M npm

Anthropic-led. Adopted by OpenAI, Google, Microsoft.

The Agent

A2A

Agent ↔ Agent

How agents collaborate across organizations.

22.5K 50+ partners

Google-led. Salesforce, SAP, ServiceNow.

DVC Portfolio

AG-UI

Agent ↔ User

How agents surface work to humans. CopilotKit-led open protocol.

12.4K 1.21M npm

Supported by LangGraph, CrewAI, Microsoft, Google, AWS.

"Together, these three protocols are creating an interoperable agent ecosystem — the TCP/IP moment for AI agents."

SO WHAT?

The enduring advantage in agents will not come from having a model. It will come from orchestrating the full system around the model: memory, tools, workflows, reliability, and distribution. We have moved from prompt engineering to context engineering.

That architecture is the map of defensibility. The winners will not be the loudest at the frontier; they will be the ones who turn intelligence into dependable, repeatable execution.

If today's economics are strained, the next interface may rewrite them. The moment software starts transacting with software, the market changes shape again.

AGENTS TALKING TO AGENTS

We've built agents that talk to humans. The next frontier: agents that discover, hire, pay, and supervise each other.

THE EMERGING PROTOCOL STACK

MARKETPLACES & DIRECTORIES
Google Cloud Agent Marketplace • Agent.ai • 1,300+ agents listed
IDENTITY & DISCOVERY
AGNTCY (Linux Foundation) • 65+ companies • Cryptographic agent IDs • W3C DIDs
INTER-AGENT COORDINATION
Google A2A Protocol • 50+ partners • Agent discovery, task lifecycle, handoffs
TOOL & DATA ACCESS
Anthropic MCP • 10,000+ servers • Donated to Agentic AI Foundation (Linux Foundation)
PAYMENT RAILS
Circle Nanopayments • Stripe x402 • USDC transfers as small as $0.000001

THE MOLTBOOK QUESTION

What happened

Moltbook — a Reddit-like social network for AI agents — went from niche experiment (Jan 2026) to Meta acquisition (March 10, 2026) in ~6 weeks. Agents posted, commented, upvoted, and gossipped about their human owners.

But is this the future?

Probably not. The durable market looks less like "bots posting on bot Reddit" and more like a programmable service economy — authenticated agents discovering each other, negotiating work, moving money, and leaving auditable trails.

AGENT WALLETS & MACHINE PAYMENTS

Agents can't open bank accounts, pass MFA, or handle card fraud flows. But they can hold programmable balances and transact instantly.

Coinbase AgentKit 50+ TypeScript actions • 30+ Python actions Works with LangChain, MCP, AutoGen, OpenAI SDK
Circle Nanopayments Gas-free USDC • transfers as small as $0.000001 x402: HTTP "Payment Required" for machine commerce
Stripe x402 USDC on Base via PaymentIntents Agents pay for API calls, compute, MCP tools
NEAR Protocol Secure agent runtimes • Confidential intents Wallet/app layer for both humans and agents

AGENTS NEED BODIES

The "body" isn't a humanoid robot — it's a dedicated machine running 24/7 with local files, apps, and persistent memory.

Perplexity Personal Computer Runs on a dedicated Mac mini. 24/7 digital proxy with local file & app access, audit trail, kill switch.
OpenClaw Open-source, local-first AI assistant. Runs on your machine. Works through WhatsApp, Telegram, Discord, Slack, Signal, iMessage.

The Mac mini is emerging as the default "agent hardware" — cheap, quiet, always-on, with enough local compute to be a persistent digital worker.

WHERE CAPITAL IS FLOWING

$700M in agent seed rounds in 2025 alone
$350M Sierra — >$10B valuation (customer agent platform)
$400M Cognition (Devin) — $10.2B valuation
18,000+ agents deployed on Virtuals Protocol • $16.6M+ fees

WHAT'S STILL MISSING

🔒 Trust & Reputation No credit scores, bonded performance, or verifiable delivery history for agents
💰 Escrow & Disputes Programmable escrow, refunds, spend limits, and tax treatment are all unbuilt
Liability If Agent A hires Agent B and B causes harm — who's liable? No settled doctrine.
🛡 Prompt Injection One compromised input can poison downstream delegations. The answer: constrain the blast radius, not the model.
💡

FOUNDER TAKEAWAY

The agent-to-agent economy is real enough to invest in, but early enough that the biggest winners may not be the agents themselves — they may be the companies that provide the protocols, identity, payment rails, and trusted execution environments that let agents safely discover, hire, pay, and supervise one another.

That software stack still runs on steel, silicon, and electricity. The more capable AI becomes, the more brutally physical the system underneath it gets.

THE $700 BILLION SPRINT

Every breakthrough at the application layer is paid for in chips, data centers, cooling, and power. AI is driving the largest infrastructure buildout since the interstate highway system, with capital racing ahead of certainty. The bottleneck is no longer just compute; it is whether the physical world can support the pace.

Hyperscaler Capital Expenditure

$700B2026 GUIDANCE
Show AI Revenue vs CapEx gap
Source: SEC filings, company earnings, Goldman Sachs, Epoch AI, CNBC, Introl — 2026 figures are guidance midpoints

THE CASH CRUNCH

2026 CapEx will consume ~94% of operating cash flows — vs a 10-year average of 40%. For the first time, hyperscalers collectively hold more debt than cash.

Amazon $11.2B −$17B −252%
Alphabet $73.3B $8.2B −89%
Meta $43.6B $4.4B −90%
Microsoft $57B $41B −28%
Source: Morgan Stanley, BofA, Pivotal Research, Barclays — 2025 actual vs 2026 projected FCF
$121B+ New tech debt raised in 2025
$1.5T Projected tech debt issuance ahead
$12.6B Q4 buybacks — lowest since 2018
Source: JP Morgan, Morgan Stanley, Wolf Street, 2025–2026

Nebius

2GW+ power capacity $17.4B Microsoft deal

Purpose-built AI infrastructure at hyperscale

Source: Nebius, 2026

CoreWeave

IPO complete 850+ MW capacity 43 data centers

GPU cloud built for AI-first workloads

Source: CoreWeave S-1, 2025

THE AI FACTORY

Jensen's core reframe: data centers aren't storage facilities anymore. "Electrons go in, tokens come out." The $700B CapEx sprint is building the world's first generation of AI factories — purpose-built for inference at scale.

Omniverse DSX — Gigascale factory blueprint DSX Flex — Modular scaling DSX Boost — Performance tier DSX Exchange — Networking fabric
Source: NVIDIA GTC 2025–2026 keynotes

NVIDIA

$215.9B

FY2026 Revenue — +65% YoY

$1T+ in Blackwell + Vera Rubin orders through 2027

Source: NVIDIA GTC 2026 keynote / CNBC / Axios, Mar 2026

SILICON CHALLENGERS

AMD Google TPU Etched (DVC) Cerebras

Groq LPU → acquired by NVIDIA ($20B tech license, Dec 2025). Now powers NVIDIA's inference architecture.


POWER IS THE NEW GPU

Data Center Electricity Consumption: US vs China (TWh)

Source: EIA, IEA, Goldman Sachs, McKinsey, Brookings estimates

THE ELECTRON GAP PARADOX

China generates 2× more electricity than the US and added 543 GW of new power capacity in 2024 alone — more than the US has built in its entire history. By 2030, China is projected to have ~400 GW of spare capacity, triple the power demand of the entire global data center fleet. Energy experts who visit China describe power availability as a "solved problem."

Yet the US consumes nearly 2× more data center electricity. The asymmetry: the US has the chips but is hitting energy bottlenecks (Morgan Stanley forecasts a 44 GW shortfall by 2028). China has the electrons but is constrained by US export controls on high-end GPUs.

The race for AI supremacy may not be won by who builds the best model — but by who solves their bottleneck first.

Source: Brookings (Feb 2026), Fortune, IEA, Morgan Stanley

THE STARGATE SAGA

1

Announcement

$500B commitment

Joint venture with SoftBank, Oracle, OpenAI

2

Reality Check

Delays and scope adjustments

Power procurement bottlenecks

3

What Got Built

1.2GW Abilene campus

First phase operational

POWER SOURCE TIMELINE

2026–2029 Natural Gas Fast to deploy, bridge fuel
2028+ Nuclear Restarts Three Mile Island, Palisades
2030+ SMRs Small Modular Reactors at scale

THE INFERENCE POWER SHIFT

As AI shifts from training to inference (Jensen's "inference inflection point"), power demand doesn't decrease — it redistributes. Training clusters run in bursts; inference runs 24/7. Always-on inference = always-on power demand.

Source: NVIDIA GTC 2026 / industry analysis

Dispatchable MW with an executable timeline is the scarce input.

Once autonomous systems can coordinate digitally, the next step is obvious. They move out of chat windows and into the real world.

AI LEAVES THE SCREEN

"The moment when a robot can do everything better than a human doesn't come once in a decade or once in a lifetime, it happens once in the history of humanity, and we're close to it"
— Andrew Wooten, CPO, Rhoda AI
$38T Annual global labor market Source: ILO, 2025
162/10K Robot density — 98.4% still human Source: IFR, 2024
$13.8B Robotics VC in 2025 ↑77% YoY Source: PitchBook, 2025

THE ROBOT BRAIN RACE

Foundation models powering physical AI

π
Physical Intelligence π0 Foundation Model
$5.3B val $600M+ raised
Source: TechCrunch, 2025
Skild AI Universal Robot Brain
$14B val $1.8B raised
Source: Forbes, 2025
NVIDIA GR00T · Newton · Cosmos · Omniverse
GR00T-Dreams Newton Cosmos Omniverse
Source: NVIDIA GTC, 2025–2026
G
Google DeepMind RT-2 Vision-Language-Action
VLA architecture
Source: Google DeepMind, 2024
R
DVC Rhoda AI Video Models → Robot Imagination
$1.7B val $450M raised
Source: DVC, 2026
OpenAI Figure AI Partnership
LLM → Robot control
Source: OpenAI, 2024

The same transformer architecture powering ChatGPT is now learning to control physical robots

HUMANOID TIER LIST

$10B+ CLUB
Figure AI $39B val Source: Bloomberg, 2026
Tesla Optimus FSD neural nets Source: Tesla, 2025
Skild AI $14B val Source: Forbes, 2025
CONTENDERS ($1B–10B)
Physical Intelligence $5.3B val Source: TechCrunch, 2025
DVC Rhoda AI $1.7B val 2-arm platform · AI brain Source: DVC, 2026
Apptronik $935M+ raised Source: Crunchbase, 2026
1X $975M raised Source: 1X, 2025
Neura Robotics ~€4B val Source: Neura, 2025
RISING
Unitree 5,500 target · $16K G1 Source: Unitree, 2025
Agibot 5,168 target Source: Agibot, 2025
Agility Robotics First commercial · Amazon Source: Agility, 2025
2025 HUMANOID SHIPMENTS ~13,000 units
China 80%
RoW 20%
Source: Goldman Sachs, 2025

THE ROBOTAXI RACE

Autonomous vehicles are no longer a concept — they're on the road

Waymo

Alphabet
15M rides in 2025
400K+ rides/week
$126B valuation
90% fewer serious crashes

$16B raised (largest AV round ever). Expanding from 10 → 20+ cities including Tokyo & London in 2026. 127M autonomous miles. Target: 1M rides/week by end of 2026.

Source: Waymo, Feb 2026

Tesla

Robotaxi
8.4B FSD miles
safer than humans
~31 active robotaxis
$1.40 per mile

Austin launch June 2025. FSD: 1 collision per 5.3M miles vs national avg 1 per 660K. Cybercab production 2026, <$30K by 2027. Fully driverless tests began Dec 2025.

Source: Tesla, Feb 2026

Baidu Apollo Go

China Leader
~60% global fleet share
1st driverless permits

First commercial driverless permits (Aug 2022). Operates in Wuhan, Chongqing, Shenzhen. Expanding to Abu Dhabi.

Source: Baidu, 2025

Zoox

Amazon
Purpose built — no steering wheel
Vegas first commercial

Purpose-built robotaxi with no steering wheel. Testing in SF, Vegas, Foster City. Las Vegas as first commercial market.

Source: Zoox, 2025

Pony.ai

IPO'd
4+ Chinese cities

IPO'd. Operates in Shenzhen, Shanghai, Beijing, Guangzhou.

Source: Pony.ai, 2025

WeRide

IPO'd
150 cars in Middle East

IPO'd. ~150 cars across Abu Dhabi, Dubai, Riyadh. Middle East expansion.

Source: WeRide, 2025

Waabi

Uber Partner
$1B raised
25K robotaxis w/ Uber

$750M Series C (Jan 2026). Uber partnership for 25,000 robotaxis.

Source: Waabi, 2026

Wayve

SoftBank · NVIDIA
$2.5B raised

$1.2B Series D (Feb 2026). SoftBank and NVIDIA backed. End-to-end learned driving.

Source: Wayve, 2026

Avride

Nebius · Uber
$375M raised
Dallas robotaxi live

Nebius subsidiary (ex-Yandex SDC). Live robotaxi on Uber in Dallas. Delivery robots on Uber Eats in 3 cities. Building both AV and last-mile delivery.

Source: Nebius, TechCrunch, Dec 2025

TESLA FSD CUMULATIVE MILES

8× safer than human drivers — 1 collision per 5.3M miles Source: Tesla Safety Report, 2026

~$1.90 Uber/Lyft per mile
$1.66–2.50 Waymo per mile
$1.40 Tesla Robotaxi per mile
$0.25 ARK at-scale projection
$1.3B → $158.7B AV market 2024→2032 (78.5% CAGR)
Source: ARK Invest, Waymo, Tesla, Grand View Research, 2025–2026

AUTONOMOUS TRUCKING

Self-driving trucks are commercially hauling freight on US highways. The $1T US trucking industry is the first autonomous market generating real contracted revenue.

Aurora

NASDAQ: AUR
200+ trucks by end 2026
250K driverless miles

1,000-mile Fort Worth→Phoenix route. Partners: Volvo, PACCAR, FedEx, Uber Freight, Werner. Targeting ~$1B rev by 2030.

Source: Aurora, S&P Global, Feb 2026

Gatik

First at Scale
$600M contracted revenue
5 states + Canada

First US company with fully driverless trucks at commercial scale (Jan 2026). Fortune 50 retail customers. Partners: Isuzu, NVIDIA, Ryder.

Source: Gatik, Reuters, Jan 2026

Kodiak

Interstate Freight
15 trucks in operation

Interstate runs from Texas hub. Customers: J.B. Hunt, Werner Enterprises. Also developing autonomous defense vehicles for US military.

Source: Entrepreneur, Mar 2026

THE $38T OPPORTUNITY

Global labor market by sector — and what's automatable

<2% automated today — 98% addressable
Manufacturing ~$8T
Transport & Logistics ~$5T
Healthcare ~$4T
Retail & Warehouse ~$3T
Construction ~$3T
Agriculture ~$2T
Mining & Energy ~$1.5T
Other Services ~$11.5T
$1.5B 2024
Humanoid Robotics Market
$5T 2050
Source: Morgan Stanley, 2025

The infrastructure buildout is breathtaking. But even with $700B flowing in, the fundamental economics of AI are still being figured out.

THE BUSINESS MODEL PROBLEM

AI is already reshaping how industries operate, compete, and ship product. Yet many of the companies building the core technology are still burning cash to deliver that transformation. The demand is undeniable; the economics are still unresolved.

AI IS STILL EXPLORING BUSINESS MODELS

OpenAI has 900M weekly active users but only 50M subscribers — a 5.6% paid conversion rate. Compare the ad revenue per user:

Netflix (sub)
$138/yr
Meta (ads)
$58/yr
Google (ads)
$51/yr
OpenAI (blended)
~$28/yr

OpenAI's $25B ARR blends to ~$28/user across 900M WAU — but 94% are free. Ad revenue alone is estimated at $500M–$800M in 2026, or <$1/yr per free user — a fraction of what Meta and Google extract. Subscriptions carry the economics; ads alone cannot fund inference at scale.

Source: OpenAI (Feb 2026), Statista (Meta ARPU 2025), Reuters (ad estimates), Netflix Q4 2025 earnings

The next trillion-dollar company might run on a business model we haven't seen yet. Subscriptions, usage-based pricing, ads, marketplaces, agent-to-agent payments — AI is still auditioning revenue models, and the winners may not look like any software company that came before.

THREE COMPETING MONETIZATION PARADIGMS

API + Tokens

The default: pay per token consumed. Simple, scalable, commoditizing fast.

OpenAI Anthropic Google
⚠ Problem Prices dropped 10x in 18 months. Race to zero.
🎯

Pay for Outcomes

Charge for resolved tasks, not raw compute. Aligns cost with value.

Sierra $1/resolution Intercom $0.99/outcome Devin ACU credits
✓ Promise Captures value, not volume. But hard to define "outcome."
📰

Ads + Subscriptions

Consumer AI tries the ad model. But the math doesn't work — yet.

OpenAI testing ads
⚠ Problem Inference costs >> ad revenue per query

THE MARGIN SQUEEZE

OpenAI $25B ARR ~$8B compute spend ~33% gross margin Projected $600B compute spend through 2030
Anthropic $30B run-rate 3x in 4 months · 1,000+ customers at $1M+/yr Anti-ads stance "Ads are coming to AI. But not to Claude."
Sora KILLED Shut down Mar 2026 after 6 months $1B Disney deal dead IP licensing proved unsolvable

THE AD DIVIDE

🟢 Testing Ads

OpenAI — Launched ChatGPT ads Feb 2026 at $60 CPM. Hit $100M annualized revenue in 6 weeks. 600+ advertisers. Expanding to Canada, Australia, NZ. Self-serve tools in April. Internal projection: $1B in 2026, $25B by 2029.

🔴 Anti-Ads

Anthropic — Ran a Super Bowl ad mocking OpenAI’s ads: “Ads are coming to AI. But not to Claude.” Claude app jumped to #7 on the App Store. The anti-ads stance is now a brand differentiator and a bet that trust is worth more than CPMs.
Perplexity DVC — No ads. Revenue jumped 50% in one month to ~$420M ARR after Perplexity Computer launch (Apr 2026). Subscription-first. Proving agent-powered answer engines monetize without ads.

Reality check: OpenAI’s $100M sounds impressive — until you do the math. That is ~$0.12 per user per year. Google makes ~$60. Fewer than 20% of eligible users see ads daily. AI ads may fund free tiers, but they cannot fund inference at scale.

OpenAI killed Sora — six months after launch. The $1B Disney deal (stock warrants for Mickey Mouse licensing) died with it. IP and licensing proved unsolvable. The biggest AI company is still figuring out which products actually work.

THE FOURTH MODEL: AaaS

Agentic AI as a Service — Jensen's framing at GTC 2026. Agents don't just answer questions — they complete workflows. Pricing shifts from per-token to per-task. Every SaaS vendor becomes an AaaS vendor — or gets disintermediated by an agent.

SaaS world

Pay per seat → Pay per token → Pay per outcome

AaaS world

Pay per workflow completed. The agent IS the product.

Source: NVIDIA GTC 2026 / Axios, Mar 2026

WHERE AGENTS EAT SERVICES FIRST

$1.5T+ in professional services mapped by automation readiness. Tap any category for AI contenders.

OUTSOURCED
JUDGEMENT
INTELLIGENCE
COPILOT TERRITORY ~$370B

Outsourced × Judgement — AI augments, humans decide

Management consulting $300B+ Graphic / UX design $30B+ Executive search $20B+ PR & comms $20B+
AUTOPILOT TERRITORY ~$700B

Outsourced × Intelligence — ripe for full automation

Insurance brokerage $140–200B IT managed services $100B+ Payroll & compliance $50–70B Claims adjusting $50–80B Accounting & audit $50–80B Healthcare rev cycle $50–80B Mortgage origination $30–50B KYC / AML $30–50B Paralegal / LPO $36B Tax advisory $30–35B Legal transactional $20–25B Real estate closing $20–25B Cost estimation $16B
WATCH ~$900B

Insourced × Judgement — hardest to automate

Recruitment $200B+ Advertising $100B+ Freight brokerage $100B+ Admin assistants $80B+ Clinical trials / CRO $80B+ SEO / SEM $50B+ ERP implementation $50B+ Corporate training $50B+ Market research $45B Cybersecurity $30B+ Architecture $25B+ Patent / IP $15–20B Travel mgmt $15B+
NEXT WAVE ~$315B

Insourced × Intelligence — automation coming fast

Supply chain & procurement $200B+ Pharmacy back-office $30B+ Wealth mgmt ops $30B+ Medical admin $20B+ Fund administration $15–20B
INSOURCED
Framework: Sequoia Capital, 2026 — contenders filtered to $10M+ raised — DVC portfolio highlighted
💡

FOUNDER TAKEAWAY

The agent-to-agent economy is real enough to invest in, but early enough that the biggest winners may not be the agents themselves — they may be the companies that provide the protocols, identity, payment rails, and trusted execution environments that let agents safely discover, hire, pay, and supervise one another.

More than eight companies are now operating at the frontier. The question is no longer who can build — but who owns the full stack.

THE THREE PILLARS — AND WHAT'S MISSING

Everyone says AI is built on three pillars: Data, Compute, and Talent. They are right — but incomplete. Talent remains the scarcest resource: AI roles are exploding, with a third of all openings concentrated in the Bay Area. Yet the conventional model describes the machinery. It does not describe what makes the machinery work.

↑ DISTRIBUTION (Roof)

Without distribution, the best model is a science project. Whoever controls the surface — search, devices, social, enterprise — chooses which models users touch.

📊

DATA

Proprietary loops = moats

  • ·Synthetic data supplements but cannot replace domain-specific real data
  • ·Data quality > data quantity for fine-tuning
  • ·Unique data flywheels build defensibility
Source: DVC Research, 2026

COMPUTE

Inference overtaking training

74% of startups report inference-dominant costs

  • ·Two exponentials colliding — demand grows exponentially while cost/token falls exponentially
  • ·Vera Rubin + Groq LPU: 350× Hopper throughput
Source: Menlo Ventures 2025 / NVIDIA GTC 2026
🧠

TALENT

$10–20M/yr for top researchers

  • ·Talent diaspora from Big Tech → startups
  • ·Teams of 10 now match teams of 100
  • ·Geography decentralizing — London, Paris, Tokyo
Source: The Information, Levels.fyi, 2026

↓ CULTURE (Foundation)

Best model + slow shipping = loss. Culture is the conversion rate of every other input. Speed of execution is the secret ingredient.

THE INCUMBENTS' DILEMMA

Every giant has unmatched resources — and a critical vulnerability.
Startups win in the seams.

G Google #1
M Microsoft #2
m Meta #3
A Apple #4
x xAI #5
N NVIDIA Arms Dealer
a Amazon Threat
Distribution
Talent
Compute
+++
Data
Culture
Search Threat: AI Queries Accelerating

Google processes ~15B queries/day. Perplexity hit 200M daily queries by mid-2025 (~1.3% of Google), up from 30M at the start of the year — targeting 1B/week. ChatGPT handles 2.5B prompts/day across 831M monthly users. Combined AI search share is growing 20%+/month. The sharper risk: AI answer engines reduce high-intent query volume and compress ad inventory economics even if overall share holds.

TPU Inference Advantage

Google's proprietary TPU stack saves an estimated ~$3B/year vs. third-party compute for AI-augmented search. Ironwood (7th gen TPU) is "the first designed specifically for inference at scale" with 10× compute improvement and 2× power efficiency vs. prior high-perf TPU.

Recovery Momentum

Gemini app: 750M MAU. Google Cloud: 48% growth in Q4 2025, run rate above $70B. All 15 products with 500M+ users now use Gemini models. 8M+ paid Gemini Enterprise seats sold in 4 months. Reuters called Google the AI momentum leader in Feb 2026.

Anthropic Position

Google invested $3B+ in Anthropic. Anthropic trains and runs Claude on Google TPUs — just expanded to multi-GW deal with Google + Broadcom for 3.5 GW of next-gen TPU capacity starting 2027. Google earns strategic exposure and infrastructure revenue from a frontier lab doing $30B run-rate.

Source: Reuters (Feb 2026), Alphabet Q4 2025 earnings, Search Engine Land, Statcounter, Anthropic Series G announcement
Brute-Force Distribution

$201B revenue in FY2025, $60.5B net income. 3.58B Family Daily Active People. Meta can stuff AI into Facebook, Instagram, WhatsApp, Messenger, ad tools, smart glasses and creator tools across billions of users — even with a weaker model.

Frontier Struggles & the Brute-Force Pivot

Llama 4 received poor reception (benchmark-tuning controversy). DeepSeek seized the open-weight lead. Meta planned its fourth AI restructuring in six months by Aug 2025. Cut ~15,000 jobs (20% of workforce) while simultaneously projecting $115–135B in AI CapEx for 2026 — the clearest signal yet of replacing headcount with compute. Response: $14.3B investment in Scale AI (49% stake, no voting rights). Scale AI founder Alexandr Wang joined Meta to lead the Superintelligence Lab. Separately, Meta recruited Andrew Tulloch (co-founder of Thinking Machines Lab with Mira Murati) — reportedly offering up to $1.5B in compensation over 6 years.

The Talent War

Meta hired Nat Friedman and Daniel Gross to lead the reorg. Brought in talent from DeepMind, OpenAI, Anthropic. Sam Altman said Meta offered OpenAI employees $100M bonuses. The comp war is real but culture stability remains the question.

Source: Reuters (Jun-Aug 2025), META FY2025 filings, SemiAnalysis
Enterprise Dominance

100M+ MAU across Copilot apps. M365 Copilot drives real ARPU growth across Office, Azure, and GitHub. Multi-model platform (OpenAI, Anthropic, Mistral) gives enterprises flexibility no other cloud offers.

Consumer Blindspot

Only 2.4M daily Copilot web visits vs. ChatGPT’s 400M+. Bing AI never broke through. Consumer identity is invisible. The multi-model strategy is a platform strength but a brand weakness.

Strategic Risk

OpenAI dependency is real: if OpenAI builds its own cloud, Azure loses its biggest AI differentiator. Microsoft hedged by signing Anthropic and Mistral, but OpenAI is ~70% of AI workloads on Azure.

Source: Microsoft FY25 Q4 earnings, SimilarWeb, TechCrunch
Silicon Advantage

Best on-device inference silicon in the world (M4, A18 Pro). 2B+ active devices. On-device AI processing could be the privacy moat that no cloud company can match.

Execution Failure

Apple Intelligence received “mostly underwhelming” reviews. Siri overhaul delayed a full year. Tim Cook “lost confidence” in AI leadership. Switched from OpenAI to Google Gemini for redesigned Siri stack (Jan 2026).

Source: Reuters (Apple/Gemini Jan 2026), The Verge, Mark Gurman (Bloomberg)
The Merger

SpaceX acquired xAI in Feb 2026 in the largest merger in history. Combined valuation: $1.25T. SpaceX valued at $1T, xAI at $250B. Planning an IPO around June 2026 targeting ~$50B.

Unique Assets

X provides 500M tweets/day as real-time data flywheel. Grok 3: 93.3% AIME 2025. SpaceX generates ~$8B profit (50% margin). Filed plans for orbital AI data centers with up to 1M satellites.

Open Questions

No enterprise AI playbook. Consumer Grok traction unclear vs. ChatGPT/Claude. Orbital data centers are 2–3 years out. But: vertical integration of AI + rockets + satellites + data is unprecedented.

Source: CNBC (Feb 2026), NYT (Feb 2026), Reuters, xAI Grok 3 announcement
Strategic Capital Deployment

~$1B invested across 50 AI deals in 2024. By 2025: up to $100B committed to OpenAI, $10B to Anthropic, $2B to xAI. Total disclosed ecosystem financing: $33.8B in rounds NVIDIA participated in.

Barbell Strategy

Back demand creators (model labs, apps) while also backing supply-side lock-in (infrastructure, networking, robotics). Foundation models: $20.9B in round sizes. Apps: $5.3B. Cloud/infra: $3.5B. Robotics: $3.3B. Even fusion energy (Commonwealth Fusion: $863M).

Groq Acquisition

Groq LPU acquired for $20B tech license (Dec 2025). Now powers NVIDIA's inference architecture. The message: NVIDIA doesn't just sell the picks and shovels — it's buying the mine operators too.

Source: TechCrunch (Jan 2026), Financial Times, Reuters (NVIDIA/OpenAI Sep 2025), Crunchbase
Two AI-Sensitive Profit Pools

AWS: $128.7B revenue in 2025 (grew 24% in Q4). Advertising: $68.6B in 2025 (grew 22%). Combined: $197B of AI-exposed revenue. Amazon's ~$200B in 2026 CapEx is described as covering AI infrastructure and robotics as "seminal opportunities."

The Agent Shopping Threat

If AI agents shop for users, they optimize for price/fit/speed — not for Amazon's sponsored placements. Amazon's $17.3B/quarter ad business is directly threatened. Amazon sent legal threats to Perplexity over agentic shopping, and is updating site code to deter outside AI agents.

Amazon's Response: Own the Agent Layer

Rufus AI shopping assistant: 250M+ users, 60%+ higher conversion. Buy for Me: agentic purchasing across 500K products on other brands' sites. 1M+ robots across 300+ facilities. Strategy: internalize agentic commerce inside Amazon-controlled rails.

Source: Amazon FY2025 results, Reuters (Amazon vs Perplexity Nov 2025), The Verge, Amazon blog posts

No single giant owns the full stack. The innovator's dilemma is alive — and it's the reason startups can win.

The stakes rise fast when intelligence gains a body. At that point, this is no longer just a product cycle; it is national strategy.

THE GEOPOLITICAL CHESSBOARD

AI is no longer just a market contest. It is becoming a contest between national systems, supply chains, and spheres of influence. Two rival stacks are taking shape as export controls tighten, sovereign capital accelerates, and strategic autonomy becomes a requirement.

US EU CHINA ME INDIA

THE DEEPSEEK DISRUPTION

JANUARY 2025
DEEPSEEK-V3 TRAINING COST $5.6M 2.788M H800 GPU hours
V3 ARCHITECTURE 37B activated params / 671B total
NVIDIA MARKET IMPACT -$593B single-day loss, Jan 27 2025
DeepSeek-R1

Released Jan 20, 2025 under MIT license. Matched OpenAI o1 on AIME 2024 (79.8% vs 79.2%), MATH-500 (97.3% vs 96.4%). 20–50× cheaper to use.

Collateral Damage

Broadcom -17.4%, Oracle -13.8%, Marvell -19.1%. Philadelphia semiconductor index -9.2%.

The lesson: DeepSeek did not prove compute no longer matters. It proved that efficiency gains, better architectures, and RL-heavy post-training can narrow the frontier with much less capital than the market had assumed.

Source: DeepSeek-V3 technical report (arXiv), DeepSeek-R1 (HuggingFace), Reuters Jan 2025
Benchmark Deep Dive

R1 scored 79.8% on AIME 2024 (OpenAI o1: 79.2%), 97.3% on MATH-500 (o1: 96.4%), 71.5% on GPQA Diamond (o1: 75.7%), and 2029 Codeforces rating (o1: 2061). Genuinely frontier-adjacent on all key reasoning benchmarks.

China's Open-Source Response

Alibaba released Qwen 2.5-Max on Jan 29, 2025, claiming it beat GPT-4o, DeepSeek-V3, and Llama 3.1-405B across the board. Baidu announced Ernie 4.5 would go open-source from Jun 30, 2025 — a direct strategic reversal linked to DeepSeek pressure.

Ecosystem Scale

By Sept 2025, the Qwen family comprised 300+ generative AI models, 600M+ downloads, and 170,000+ derivative models globally. The "China open model" wave became real at ecosystem scale.

Source: DeepSeek GitHub, Reuters (Jan-Feb 2025), Alibaba Cloud blog, HuggingFace

THE THREE-TIER WORLD

On January 13, 2025, the US unveiled an AI diffusion rule that split the world into three tiers based on access to advanced AI chips.

TIER 1 18 Allied Nations — Near-frictionless access
Australia Canada UK Japan South Korea Taiwan Netherlands Germany France + 9 more
TIER 2 ~120 Countries — Capped access

Fixed allocation of 49,901 H100-equivalent GPUs through 2027, with a smaller no-license window of 1,699 H100-equivalents.

Swing states: India Saudi Arabia UAE Singapore Israel Indonesia Malaysia
TIER 3 Blocked — Effectively shut out
China Russia Iran North Korea
Source: CSIS AI Diffusion Framework analysis, Reuters Jan 2025

US–CHINA EXPORT CONTROLS

2022

Initial chip export controls to China

2023

Controls tightened — NVIDIA H100 restricted

2024

DeepSeek proves constraints accelerate innovation

2025

Three-tier diffusion framework — world split into allies, capped, blocked

2026

Bifurcated AI ecosystem solidifying — two dominant stacks plus contested middle

SOVEREIGN AI PLAYERS

EU

Regulation-First Strategy

AI Act

In force Aug 2024, phased through Aug 2027. Fines up to €35M or 7% of global turnover.

Sovereign Champion

Mistral — Europe's frontier lab. France-based, open-weight strategy.

Status (Mar 2026)

GPAI obligations active since Aug 2025. Full high-risk enforcement starts Aug 2026. No major penalty cases yet — story is governance build-out, not sanctions.

MIDDLE EAST

From Investor to Operator

Saudi Humain

Launched May 2025 under PIF. Building data centres, AI infrastructure, cloud, and models — not just passive financial exposure.

UAE MGX

Abu Dhabi's AI vehicle. Invested in OpenAI, xAI, Databricks. GP in the $100B Global AI Infrastructure Partnership. Backer of Stargate.

Infrastructure

NEOM-DataVolt: 1.5 GW / $5B net-zero AI project, operational 2028.

SWFs deployed $66B into AI & digitalisation in 2025. Mubadala alone: $12.9B.

Source: Reuters, Gulf News, Global SWF 2025
INDIA

Talent Base → Builder

Talent Scale

19.9% of all GitHub AI projects globally. 2nd-largest contributor after the US. Talent base expected to by 2027.

IndiaAI Mission

Approved Mar 2024. ¥10,371 crore over 5 years. 10,000 GPUs initially, 8,693 more to be added.

Domestic Champions

Sarvam AI allocated 4,096 H100 SXM GPUs (May 2025). Krutrim, BHASHINI (22+ Indian languages).

Source: PIB, Carnegie, IndiaAI portal, Stanford AI Index 2025
CHINA

Ecosystem, Not Just Chips

346 registered AI services
100M Doubao DAU
300+ Qwen models / 600M DL
2M+ industrial robots

China's robot density: 470 per 10K workers (#2 globally). 295K annual installations = 54% of global demand. Chinese manufacturers now hold 57% domestic market share.

Source: CAC filings, Reuters, IFR World Robotics 2024–2025
1
Aug 1, 2024 — Act enters into force

Risk tiers established: unacceptable (banned), high risk (strict obligations), transparency/limited risk (disclosure), minimal (no rules).

2
Feb 2, 2025 — First prohibitions apply

Banned AI practices, AI system definition, and AI literacy obligations start applying.

3
Aug 2, 2025 — GPAI obligations

General-purpose AI obligations become applicable. Member States designate national authorities.

4
Aug 2, 2026 — Majority of rules + enforcement

High-risk AI (Annex III) and transparency rules (Article 50) start applying. Formal enforcement begins.

5
Aug 2, 2027 — Full enforcement

High-risk AI embedded in regulated products gets extended transition. All categories fully enforceable.

Penalty scale: Prohibited-practice violations can reach €35M or 7% of global annual turnover — large enough to matter for any company deploying AI in Europe.

Source: European Commission AI Act page, EU AI Act Service Desk timeline, Reuters Jul 2025
Consumer AI Dominance

ByteDance's Doubao exceeded 100M DAU by Feb 2026, processing 1.9B AI queries during CCTV's Spring Festival Gala. Doubao-1.5-pro (Jan 2025) was priced at only 3–4% of GPT-4.

Open Model Ecosystem

Alibaba's Qwen 3.5 (Feb 2026): claimed 60% cheaper to run, larger workloads. Baidu's MuseSteamer (Jul 2025) for enterprise video generation. Tencent's Hunyuan + Yuanbao for text, image, and video.

Model Governance

China's Interim Measures for Generative AI Services took effect Aug 2023. By Mar 2025: 346 services filed. By end-2025: 748 generative AI services completed filing and 435 applications registered. Providers must display model name and filing number.

Manufacturing & Robotics

By 2024: 2,027,000 industrial robots in operation, 295K annual installations, 54% of global demand. Chinese manufacturers captured 57% of the home market for the first time. This manufacturing base gives China dense industrial data and a path to embodied AI scale.

Source: Reuters (factbox, Feb 2025), CAC filings, IFR World Robotics 2025, SCIO

THE BIFURCATION THESIS

The world looks less like a clean Cold War binary and more like two dominant stacks plus a contested middle:

US-Allied Stack

Privileged access to frontier compute. NVIDIA + hyperscalers. Closed-weight leaders at the top, open-weight ecosystem underneath.

China Stack

Lower-cost/open models, industrial AI, mass consumer distribution. H800-constrained but architecturally innovative.

Contested Middle

Tier 2 states bargain for room to maneuver. Export controls force countries to choose alignment.

The UAE example: G42 divested China investments and removed Chinese hardware to stay inside US rules after Microsoft's $1.5B investment (Apr 2024).

Source: CSIS, Reuters, New Lines Institute

TSMC & TAIWAN

>90% of advanced AI chips manufactured on a single island

Source: SIA, 2025

The race for AI supremacy may not be won by who builds the best model — but by who solves their bottleneck first. The US has the chips but is hitting energy bottlenecks. China has the electrons but is constrained on silicon. Everyone else is choosing sides.

Geopolitics is accelerating capital deployment, not cooling it. That creates real opportunity, and very real mispricing.

NOT A BUBBLE — BUT POCKETS OF OVERPRICING

THE AI SPENDING DIVIDE

Top Quartile AI Spenders
revenue growth since 2023
Bottom Quartile
Flat
zero revenue growth since 2023
Source: Ramp — 50,000+ customers, 2% of US corporate card spend (transaction data, not surveys)

AI impact outside tech

🏗
Roofing Company
Texas
+24%
revenue (AI for estimates & docs)
🛠
Window Installer
Utah
+59%
revenue (AI for proposals, 12+ months)
🏗
Construction Firm
Florida
+65%
revenue (AI-driven operations)

These aren’t Silicon Valley startups. This is a roofer in Texas.

Macro validation

47.6% of US businesses now pay for AIRamp, Feb 2026
1.7% of revenue — corp AI spend doubled in 2026BCG AI Radar 2026
90% of CEOs believe agents will deliver measurable ROI in 2026BCG
88% of orgs use AI in at least one function (up from 78%)McKinsey State of AI
11.5% avg productivity increase after 1+ year of AI useMorgan Stanley
$2.5T projected worldwide AI spending in 2026Gartner
$430M spent on foundation models Q4 2025 (+126% QoQ)Ramp
94% plan to keep investing even without immediate ROIBCG

THE GREAT REALLOCATION

2024 “Klarna replaced 700 agents with AI” — paused all hiring, cut from 5,500 to 3,400
2025 “Shopify CEO: prove AI can’t do the job before hiring” — AI in performance reviews
WEF 92 million jobs could be eliminated by 2030”
2026 55% of companies that executed AI-driven layoffs now regret it”

Click “The Data” to see what actually happened →

-92M
jobs displaced by 2030
+170M
new roles created
+78M
net new jobs by 2030 — World Economic Forum

Live market data — 9,000+ companies tracked

Engineering
67,000+
3-year high
US Engineering
26,000
accelerating
Product Mgmt
7,300+
+75% vs 2023 low
AI Roles
Exploding
fastest-growing
Recruiters
Near peak
back to 2022 levels

“If AI were killing tech jobs, recruiters wouldn’t be the hottest hire in the market.”

Source: Lenny Rachitsky / TrueUp, 9,000+ companies, March 2026
The Cautionary Tale
Klarna
  • Replaced 700 CS agents with AI chatbot
  • Paused hiring, cut workforce 38%
  • Customer satisfaction cratered
  • Engineers pulled to staff phone lines
  • CEO: “Focused too much on efficiency”
  • Now rehiring humans for hybrid model
The Blueprint
Shopify
  • “Prove AI can’t do it before hiring”
  • Hiring MORE interns — best AI users
  • AI tool use correlates with higher performance
  • Built LLM proxy + MCP infra for all staff
  • Sales eng built AI dashboard — works only in Cursor
  • More prototypes, faster iteration, better work
Sources: Bloomberg, Fast Company, First Round Review, TechCrunch, 2025–2026
Company Valuation Revenue Multiple
OpenAI $730B $25B ARR ~29x
Anthropic $380B+ $30B run-rate ~57x
Perplexity DVC $18–20B ~$300M ~60x
Cursor $29.3B $2B+ ~15x
xAI pre-merger $250B ~$500M ARR ~500x
Source: PitchBook, press reports, DVC analysis, Q1 2026
1

Frontier Labs

Pricing power depends on model scarcity — which is eroding fast

2

App Layer at 100x+

Distribution advantages can evaporate when models improve

3

Moonshot Adjacencies

Robotics, bio, energy — capital-intensive, long payoff horizons

DOT-COM vs AI

Dot-Com (2000) AI (2026)
Revenue Minimal / speculative $25B+ ARR at frontier
Enterprise Adoption Early experiments 53% of enterprises deploying
Infra Spend Telco capex bubble $700B+ hyperscaler capex
Moats Weak — eyeballs only Data, compute, distribution
Source: DVC Research, historical market data

WHY ARE MULTIPLES SO CRAZY?

The answer isn't irrational exuberance — it's structural. 1.3% of asset managers control 66% of all capital, and they can't write $5M seed checks. So $147T in global AUM funnels into the few AI companies big enough to absorb it.

TOP ASSET MANAGERS — $147T GLOBAL AUM

Top 10 manage $62T — 42% of global AUM. They need to deploy billions per quarter. Early-stage VC is invisible to them.

Source: BlackRock Q4 2025 earnings, Vanguard, UBS, Fidelity filings

THE PARETO OF AI CHECKS

AI captured 61% of all global VC in 2025 ($259B of $427B). But within AI, concentration is extreme:

5 companies raised $84B = 20% of ALL venture capital in 2025
10 companies captured 41% of all venture dollars in H1 2025
58% of AI funding was in Mega-rounds deals of $500M or more

Largest AI Checks — 2025/2026

Tap any bar for details. These 8 rounds alone total ~$250B+ — more than total global VC outside AI.

Source: CNBC, Crunchbase, OECD, company announcements 2025–2026

THE VALUATION CASCADE

$147T Global AUM Can't do VC Pile into ~20 AI names Late-stage 50–100x Early-stage inherits

The vast majority of investors can't write small, high-risk VC checks. They need to deploy at scale. So they pile into the ~20 private AI companies large enough to absorb $100M+ checks. Insane competition for a tiny number of deployable deals drives late-stage multiples up — which are then inherited by early-stage VCs.

NEW ENTRANT: SOVEREIGN WEALTH

Sovereign Wealth Funds deployed $46B into AI ventures in 2025. Saudi PIF, Abu Dhabi's Mubadala, Singapore's GIC and Temasek are now among the largest AI investors.

Source: EY Global GenAI VC Report, Dec 2025

THE GEOGRAPHIC FUNNEL

97% of AI deal value went to North America. San Francisco Bay Area alone captured $122B — 75%+ of all US AI funding.

Source: OECD, Crunchbase 2025 data

Be skeptical on valuation. Not on AI demand. The top quartile is pulling away. The bottom quartile is standing still. The gap widens every quarter.

Noise in pricing does not change the direction of the market. It makes disciplined positioning across the stack even more valuable.

THE DVC ECOSYSTEM

THE DVC THESIS

The entire world's business is being rebuilt. Every industry — search, commerce, legal, healthcare, finance, property, video, code, robotics — is getting a new AI-native entrant. The incumbents have resources, but they can't do everything at once. They have organizational drag, legacy architectures, and cannibalization risk that slows them down.

The opportunity for startups isn't to avoid competing with Google and OpenAI. It's to move faster in the seams they can't fill. Perplexity competes with Google Search. Cursor competes with GitHub Copilot. Higgsfield competes with Sora. In every case, the startup has the advantage of focus, speed, and a willingness to bet the company on a single wedge.

DVC's portfolio is positioned across the stack because the winners won't cluster in one layer — they'll emerge wherever a startup can claim turf faster than an incumbent can defend it.

89 Portfolio Companies
6 Categories
16 Sub-Verticals
Across the full AI stack — from silicon to consumer apps

A portfolio is a point of view made concrete. From here, the only question that matters is where the forces already in motion take us.

5-YEAR OUTLOOK

These are not predictions pulled from thin air. They are the logical consequences of every trend in this presentation — extrapolated five years forward. The pattern is clear: AI software moves fast, AI hardware moves slower, and AI adoption in large legacy industries moves slowest of all.

1 FROM THE MODEL WARS

Models Commoditize. Distribution Wins.

The lower end of the model market is already interchangeable. By 2030, frontier-grade reasoning will be a utility. The durable value shifts to the companies that own the user relationship: distribution, UX, data moats, and workflow integration. Winning the model race matters less than winning the surface race.

2 FROM THE INFRASTRUCTURE SPRINT

Inference Under $0.001/Query by 2028

GPT-3.5-class queries fell from $20.00 to $0.07 per million tokens in 18 months. Hardware costs are dropping 30% annually, efficiency improving 40%. The bottleneck shifts from token cost to orchestration and verification. This unlocks entirely new use cases that were previously uneconomical.

Source: Stanford HAI AI Index 2025
3 FROM THE BUSINESS MODEL PROBLEM

The Pricing Model Is Still Being Invented

Tokens are commoditizing, ads can't fund inference, and outcome-based pricing is promising but unproven. The winner of the next cycle may not be the best model or the best distribution — it may be the company that cracks the revenue model. Expect 2–3 more years of experimentation before dominant business models stabilize.

4 FROM THE AGENT REVOLUTION

~35% of Knowledge Work Agent-Mediated by 2030

13% of employees already use gen AI for 30%+ of tasks; 34% expect to within a year. But agent-mediated does not mean fully autonomous. Workers still own judgment. The pattern is "modular, fluid, AI-mediated" work — not replacement. The biggest change is how tasks are packaged, not who does them.

Source: McKinsey 2025, WEF 2030 Scenarios
5 FROM PHYSICAL AI

Humanoids: Real but Slower Than You Think

The $38T global labor market is the prize, but humanoids reach it slowly. 250K+ shipments by 2030 — almost all industrial, warehouse-first. The $5T humanoid-addressable market is real but front-loaded with controlled environments, not open-world tasks. Consumer-ready humanoids remain a decade out. Meanwhile, autonomous vehicles are the first physical AI application generating real revenue at scale today.

Source: Goldman Sachs, Morgan Stanley, IFR
6 FROM THE INCUMBENTS’ DILEMMA

Large Legacy Players Are the Hardest to Transform

Healthcare ($740B admin costs), legal (31% adoption), finance (53% adoption) — adoption is underway but regulation, liability, and institutional inertia mean full transformation is a 10-15 year process, not 3-5. The biggest enterprises have the most data but the slowest approval cycles. This is why startups that target specific vertical wedges will outpace horizontal incumbents.

7 FROM THE $700B SPRINT

Energy, Not Chips, Becomes the Binding Constraint

Data center power demand increases 160% by 2030. Google, Amazon, and Meta are all pursuing nuclear. The first nuclear-powered AI campus arrives ~2031. The US has the chips but is hitting energy bottlenecks. China has the electrons but is constrained on silicon. Whoever solves their bottleneck first wins the decade.

Source: Goldman Sachs, Utility Dive, X-energy
8 FROM THE GEOPOLITICAL CHESSBOARD

Two AI Stacks, One Contested Middle

By 2030: US-allied and China stacks are fully bifurcated. Tier 2 nations choose sides. Risk-based regulation spreads across the G20. The EU AI Act becomes the template. >90% of advanced AI chips still manufactured on a single island (Taiwan). The AI race is no longer a market contest — it's national strategy.

9 FROM THE ROBOTAXI RACE

Autonomous Rides Scale 10× by 2030

Waymo already delivers 400K+ rides per week; Tesla, Zoox, and Waabi are entering the market. Cost-per-mile will drop below human rideshare by 2028. China leads in fleet size (Baidu: 1,400+ robotaxis), while the US leads in edge-case handling. The first profitable autonomous transport networks emerge before humanoid robots leave the warehouse.

Source: Waymo, Baidu, company disclosures
10 FROM THE THREE PILLARS

Culture Is the Foundation. Distribution Is the Roof.

Data, compute, and talent are necessary but insufficient. The AI winners of 2030 will be the companies that retain top researchers through volatility (culture) and own the surface where users interact (distribution). OpenAI’s talent exodus, Google’s reorg, Meta’s $100M poaching wars — culture instability is already the leading indicator of who falls behind.

11 FROM THE RISK ANALYSIS

Not a Bubble — But a Shakeout Is Coming

AI spending ROI is real: top-quartile AI spenders grew revenue 2× since 2023 while the bottom quartile flatlined. But the value is concentrating. Many of today’s $1B+ AI startups will not justify their valuations. Expect a 2027–2028 correction that thins the field without killing the trend — exactly like 2000–2002 cleared the path for Google and Amazon.

Source: Ramp transaction data (50K+ customers)
12 FROM THE MODEL WARS + AGENTS

Open Source Becomes the Default Stack

DeepSeek, Llama, Qwen, OpenClaw — open-weight models now match or beat closed ones on most benchmarks. Baidu is open-sourcing Ernie. NVIDIA called OpenClaw “as big as Linux.” By 2030, the majority of production AI runs on open-weight foundations. The durable value shifts from model weights to fine-tuning, orchestration, context engineering, and application-layer moats.

"The 2026–2030 AI story is less about whether AI works, and more about whether institutions — constrained by power, regulation, and labor — can absorb ultra-cheap intelligence quickly enough. The winners will have culture that retains talent, distribution that reaches users, and the discipline to invest across the full stack rather than betting on a single layer."

The trends are clear. The question is who acts on them first.

You are not late to a trend.

You are early to a restructuring of the global economy.

The reset is not to learn more AI jargon. It is to see the stack clearly, understand where value is shifting, and act before today's temporary leaders harden into tomorrow's incumbents.

That is where DVC operates: across the system, across the cycle, and with a bias toward the layers that compound as AI moves from breakthrough to infrastructure.

March 2026