We thought model cycles would outpace infra

We thought model cycles would outpace infra

They did. That broke the stack

$300B in data center spend this year

Model performance doubled twice in 12 months

Windows 10 dies in 60 days

Infra refresh cycles stretched from three years to six

Capex got delayed then drained

Endpoints are running out of time

GPT-5 launched with broken charts and hallucinations

OpenAI sold “vibe coding” to cover the gaps

You can’t build production off aesthetics

SoftBank posted $2.9B net income

Most AI infra shops can’t show $3M ARR

LPs already marked it down in Q1

401(k) rules are changing

Retail capital is coming for unregulated models

The foundation is already cracked

Biopharma VC dropped 40 percent in Q2

Only AI and clinical-stage deals survived

Everyone else held worthless paper

The same thing that scaled the AI boom

Is now making it unusable at scale

AI-native infra is a story

The hardware refresh cycle is a fact

Windows 10 sunsets in October

That’s when the forced upgrade hits

Most infra funds still think they have time

Some of the loudest AI infra funds haven’t shipped a single public case study in over a year

Most enterprise deployments are still stuck in “pilot”

Ask your CTO how many models they’ve deprecated this quarter

You already know which funds this burns

Plenty of firms will pretend this isn’t about them

It is

Mythology pays better than product