Are large AI infrastructure deal announcements transparent, verifiable commitments, or deliberately vague marketing constructs?

Steeve Morin · Feb 24, 2025

Stargate's $500B announcement is not credible and, even if true, is just vertical scaling — more of the same rather than a differentiator

Announce-first-figure-it-out-later style; big spend on compute is brute-force scaling, and with ~90% GPU margins the headline number could be far smaller

63:56 20VC: Why Google Will Win the AI Arms Race & OpenAI Will Not | NVIDIA vs AMD: Who Wins and Why | The Future of Inference vs Training | The Economics of Compute & Why To Win You Must Have Product, Data & Compute with Steeve Morin @ ZML

Andrew Feldman · Oct 6, 2025

The headline multi-billion-dollar AI deal announcements are systematically overstated — 'up to X over five years' means the real number could be a fraction, and nobody ever audits whether the promised jobs or factories materialize

'Up to' is an unfalsifiable marketing construct and no one holds announcers to account afterwards

6:40 20VC: Cerebras CEO on Why Raise $1BN and Delay the IPO | NVIDIA Showing Signs They Are Worried About Growth | Concentration of Value in Mag7: Will the AI Train Come to a Halt | Can the US Supply the Energy for AI with Andrew Feldman

Andrew Feldman · Oct 6, 2025

The NVIDIA–OpenAI $100 billion announcement was deliberately structured to be un-analyzable — unbounded amount, unspecified timeframe, no valuation given

If you want an investment to be clear you state amount, valuation and closing; the opposite construction is a choice, and it's a reasonable one for both parties

Scope: says the structure is reasonable for both parties

13:16 20VC: Cerebras CEO on Why Raise $1BN and Delay the IPO | NVIDIA Showing Signs They Are Worried About Growth | Concentration of Value in Mag7: Will the AI Train Come to a Halt | Can the US Supply the Energy for AI with Andrew Feldman

Andrew Feldman · Oct 6, 2025 · hedged

The vagueness of the NVIDIA–OpenAI deal terms likely reflects genuinely many moving parts that couldn't be described clearly without disclosing more than the parties wanted.

Complex multi-part arrangements can't be summarized without revealing details the companies chose to withhold.

Scope: speculation about the parties' motives

14:44 20VC: Cerebras CEO on Why Raise $1BN and Delay the IPO | NVIDIA Showing Signs They Are Worried About Growth | Concentration of Value in Mag7: Will the AI Train Come to a Halt | Can the US Supply the Energy for AI with Andrew Feldman

David Cahn · Oct 27, 2025 · hedged

Announcing deals in gigawatts rather than dollars obscures the scale: at roughly $40–60B per gigawatt, 100GW is an $8 trillion question and 250GW a $20 trillion question, and that capital is not funded

Most people don't know how many dollars a gigawatt is, so gigawatt framing hides how much the ante has been raised

26:59 Gigawatt framing obscures the true dollar scale of commitments

Jonathan Ross · Feb 17, 2025

Much of the current data center buildout is fake supply from real estate people who don't understand the business and whose projects won't actually get built

Entrants promise capacity in months without having ordered generators (90-month lead times) and without knowing data centers need water — data centers are not real estate

33:36 Much of announced data center capacity is fake supply from inexperienced entrants

Andrew Feldman · Mar 24, 2025 · hedged

The claim that today's data center buildout is dominated by unqualified 'tourists' is overstated — there are some tourists, but much of the gigawatt-scale construction is being done by highly sophisticated builders

The early movers were Bitcoin mining companies who built near low-cost power to run power-hungry compute, and they are now leaders on some of the largest projects

21:33 Tourist narrative is overstated most builders are sophisticated