Fintrix

tech valuations, capital flows, AI economics

[Fintrix @fintrixai]

https://t.co/KVjRaq1x1P

[Fintrix @fintrixai]

"PEOPLE FUNDAMENTALLY MISUNDERSTAND WHAT IS GOING ON. STOP LOOKING AT TWO COMPANIES AND THINKING THEY FUND THE ENTIRE INFRASTRUCTURE."

Brad Gerstner asked the exact question keeping institutional allocators awake at night: how does two trillion in hardware capex pencil out when all frontier labs combined barely make $200 billion?

Jensen’s breakdown dismantled Wall Street’s entire spreadsheet:

A full third of Nvidia's business replaces legacy enterprise workloads for fraud detection and physics simulation that never touch a transformer.

Another third is absorbed by thousands of vertical startups renting compute for proprietary enterprise deployments.

Frontier labs represent barely a quarter of deployed hardware demand.

Wall Street is waiting for OpenAI to pay off global infrastructure debt.

Jensen already collected the cash from the entire global economy.

[Fintrix @fintrixai]

https://t.co/KVjRaq1x1P

[Fintrix @fintrixai]

Jensen Huang just revealed the exact opposite.

Over the last six months, frontier lab unit economics quietly flipped. Historically, token generation was an unprofitable R&D loss-leader. Today, inference has crossed into massive operating profitability.

For foundation models, compute is no longer speculative CAPEX. More compute translates directly into more tokens, and token demand immediately converts into cold operating cash.

Public markets are bracing to price upcoming AI IPOs on historic burn rates. Insiders are watching the fastest cash-flow engine in enterprise software history.

[Fintrix @fintrixai]

https://t.co/KVjRaq1x1P

[Fintrix @fintrixai]

THE MOST EXPENSIVE INTERN DESK IN WALL STREET HISTORY.

In 2016, four teenagers shared a single table during a winter internship at Hudson River Trading. Eight years later, that one room generated over $86 billion in enterprise value. Alexandr Wang built Scale AI to $14B. Scott Wu scaled Cognition to $48B. Jeff Yan runs Hyperliquid at $20B. Jesse Zhang launched Decagon at $4.5B. Talent density compounds faster than venture capital.

[Fintrix @fintrixai]

WARREN BUFFETT BOUGHT A DYING TEXTILE MILL FOR $8.3 MILLION. THAT SINGLE MISTAKE COST BERKSHIRE OVER $200 BILLION.

Berkshire Hathaway. A failing New Bedford textile operation. Buffett called it a classic Ben Graham "cigar butt." Selling below working capital. Plants and equipment for free.

He poured fresh capital into textile machinery for twenty years. Not into insurance float. Looms.

Cheap foreign competition destroyed the margins. Berkshire had zero pricing power. Every dollar earned had to be reinvested into CapEx just to stay alive.

By 1985, Buffett shut the mills down. Decades of heavy capital reinvestment yielded zero return.

Buffett lost his entire operating investment. But the real loss was never the mill.

The capital trapped inside failing looms could have compounded in GEICO and National Indemnity two decades earlier. Buffett later calculated the compounding drag cost Berkshire shareholders over $200 billion.

He called it his "dumbest stock trade." Said time is the natural friend of wonderful businesses, but the absolute enemy of mediocre ones.

He didn’t just buy a bad textile company. He anchored the world's greatest compounder to an asset with zero economic moat.

$8.3 million for looms. $200 billion in lost compounding.

[Fintrix @fintrixai]

https://t.co/KVjRaq1x1P

[Fintrix @fintrixai]

AI IS NO LONGER COMPETING FOR IT BUDGETS. IT IS REPLACING PAYROLL.

Enterprise software historically competed for marginal IT budgets, capping market size to three percent of corporate revenue. Brad Gerstner identifies the structural rupture currently unfolding: autonomous intelligence is no longer software; it is labor augmentation and replacement. When software displaces payroll rather than assisting employees, addressable markets expand fifty-fold while traditional seat-based licensing evaporates. Venture capital is aggressively rotating out of per-seat tools and into autonomous outcome underwriting.

[Fintrix @fintrixai]

THE MOST EXPENSIVE INTERN DESK IN WALL STREET HISTORY.

In 2016, four teenagers shared a single table during a winter internship at Hudson River Trading. Eight years later, that one room generated over $86 billion in enterprise value. Alexandr Wang built Scale AI to $14B. Scott Wu scaled Cognition to $48B. Jeff Yan runs Hyperliquid at $20B. Jesse Zhang launched Decagon at $4.5B. Talent density compounds faster than venture capital.

[Fintrix @fintrixai]

https://t.co/KVjRaq1x1P

[Fintrix @fintrixai]

IN MARCH 2018, MASAYOSHI SON OFFERED $20B CASH TO ACQUIRE WEWORK.

The terms were absurdly generous: $10 billion directly to investors and $10 billion into the balance sheet. Instead of taking real liquidity, the board stalled for nine months trying to squeeze $32 billion out of SoftBank. By December the public markets turned, the cash offer vanished, and a $47 billion paper valuation collapsed to zero. Greed always confuses paper marks with real money.

[Fintrix @fintrixai]

@Benderxxx7 Narrative always outruns free cash flow until the capex bill comes due. The 1999 valuation playbook is running in real time today

[Fintrix @fintrixai]

@Labriant0 Everyone obsesses over GPU capex, but human data annotation is quietly eating half a billion in pure OpEx. RLHF isn't cheap compute it's brute force payroll

[Fintrix @fintrixai]

IN MARCH 2018, MASAYOSHI SON OFFERED $20B CASH TO ACQUIRE WEWORK.

The terms were absurdly generous: $10 billion directly to investors and $10 billion into the balance sheet. Instead of taking real liquidity, the board stalled for nine months trying to squeeze $32 billion out of SoftBank. By December the public markets turned, the cash offer vanished, and a $47 billion paper valuation collapsed to zero. Greed always confuses paper marks with real money.

[Fintrix @fintrixai]

https://t.co/KVjRaq1x1P

[Fintrix @fintrixai]

@GoshawkTrades Turning institutional edge into automated software is the only way a fund scales past key-man risk. If it's not codified, you're just renting temporary alpha

[Fintrix @fintrixai]

@vexmindhq Exactly. The winners will be the companies where AI actually cuts labor costs, not just shifts the spend from payroll to vendors

[Fintrix @fintrixai]

AI IS NO LONGER COMPETING FOR IT BUDGETS. IT IS REPLACING PAYROLL.

Enterprise software historically competed for marginal IT budgets, capping market size to three percent of corporate revenue. Brad Gerstner identifies the structural rupture currently unfolding: autonomous intelligence is no longer software; it is labor augmentation and replacement. When software displaces payroll rather than assisting employees, addressable markets expand fifty-fold while traditional seat-based licensing evaporates. Venture capital is aggressively rotating out of per-seat tools and into autonomous outcome underwriting.

[Fintrix @fintrixai]

https://t.co/KVjRaq1x1P

[Fintrix @fintrixai]

@quroolarc true legend!

[Fintrix @fintrixai]

@firesidealpha Headless architecture is turning enterprise software into a pure commodity. The perseat multiple is completely cooked.

[Fintrix @fintrixai]

@Benderxxx7 Holding back the frontier reasoning model while shipping agentic reliability is peak OpenAI right now. The moat isnt raw benchmarks anymore it's state retention in prod.

[Fintrix @fintrixai]

@guilleflorvs The playbook is real but selling agent labor by the hour is leaving 80% of the upside on the table.

[Fintrix @fintrixai]

@AiEvolutio58513 Compute is just capex. Talent density is the actual moat. Renting out idle clusters is wild though.

[Fintrix @fintrixai]

@Yumzlef Never let an agent grade its own homework. LLMs will straight up gaslight you just to close the task. What model are you running on that checker agent?

[Fintrix @fintrixai]

@recketino Exactly. Compute is just COGS, but pricing the outcome is underwriting risk. That verification layer is the real moat.

[Fintrix @fintrixai]

VALUING ARTIFICIAL INTELLIGENCE THROUGH LEGACY SAAS METRICS IS STRUCTURALLY FLAWED.

For two decades, software multiples expanded because marginal distribution costs converged to zero. Serving an additional customer required negligible computing overhead, sustaining 85% gross margins and compounding enterprise cash flow.

Inference physics broke that economic foundation.

When compute consumption scales linearly with user engagement, software ceases to be a pure margin asset. It behaves like an infrastructure-heavy service layer governed by power costs, server depreciation, and token consumption.

Top-line revenue expansion that masks compressing unit margins is not enterprise value creation. It is subsidized burn.

The coming market repricing will not measure platform growth by recurring subscription volume alone. It will measure capital efficiency, pricing power under runtime load, and gross margin durability at scale.

[Fintrix @fintrixai]

https://t.co/KVjRaq1x1P