Ed Zitron: Microsoft Has $80 Billion Of AI Chips It Can't Plug In
Market Talk with George Noble
197,563 views
The AI bubble panel: Ed Zitron, Gary Marcus and Julien Garran join George Noble, Jack Gamble (Nobody Special Finance) and Alessandro (Risk Takers) to ask whether the AI boom is the biggest misallocation of capital ever, and who is left holding it if it breaks.
Ed Zitron explains why he says you can't trust Anthropic's or OpenAI's numbers until they are audited, why $1.3 trillion of compute commitments don't add up, and his estimate that $200 billion to $350 billion of GPUs are sitting in warehouses and unpowered data centers, and why that is nothing like the dark fiber of the dot-com bubble.
Gary Marcus explains why "rogue AI" is really badly built software, why large language models can't reliably follow instructions, and why he says AI stocks now trade like meme stocks.
Julien Garran of MacroStrategy Partnership lays out the $1.7 trillion Big Tech has already spent on AI capex, another trillion a year planned, and his estimate that on a generous depreciation schedule it wipes out 98% of their non-AI income by 2034.
Jack Gamble breaks down the creative financing behind the data center build-out, including Amazon's chip sale-leaseback, and the Navier-Stokes proof story. George Noble asks what it means for gold, the dollar, Treasuries and the Fed when it unwinds.
We also get into OpenAI, Anthropic and the Anthropic IPO, Nvidia, Broadcom and Google's TPUs, Microsoft, Amazon and Meta, GPU depreciation and rental prices, private credit, pensions and 401(k)s, how this compares with the dot-com bubble, and what it all means for gold and the US dollar.
🎙️ The panel
Ed Zitron: Where's Your Ed At → https://www.wheresyoured.at | Better Offline → https://www.betteroffline.com
Gary Marcus: Marcus on AI → https://garymarcus.substack.com
Julien Garran: MacroStrategy Partnership → https://www.macrostrategy.co.uk
Jack Gamble: Nobody Special Finance → @nobodyspecialfinance
Alessandro: Risk Takers → @officialrisktakers
📅 Recorded: October 2, 2026
🎧 Listen on Spotify → https://open.spotify.com/show/2Y1GCTD...
🎧 Listen on Substack → https://georgenoble.substack.com/
🎧 Listen on Apple podcast → https://podcasts.apple.com/us/podcast...
🎟️ Register now for the George Noble Best stock ideas Online Summit Conference → https://noble-capevents.com/
🔗 Follow George Noble:
X: https://x.com/gnoble79
⏱️ Chapters
0:00 Coming up
1:00 Meet the panel
1:36 Is "rogue AI" real? Gary Marcus on AI agents and safety
6:58 Air-gapped computers and the AI doom stories
8:31 Why AI models can't reliably follow instructions
10:53 OpenAI, Codex and the Navier-Stokes proof
15:27 Why Ed Zitron doesn't trust Anthropic's numbers
18:14 Broadcom, Google TPUs and where the systemic risk sits
21:14 ARR: the revenue number that isn't in GAAP
25:56 Julien Garran: $1.7 trillion spent, and what has to pay it back
31:04 Why isn't Meta's Muse the killer app?
34:33 The 404 Media report on Muse
39:33 AI agents, crypto and the "agentic bank run"
44:14 Revenue isn't profit: $1.3 trillion of compute commitments
47:39 Gary Marcus: AI doesn't work like ordinary software
49:39 The economic fallout of AI capex
53:06 Gold
53:48 The Fed, Treasuries and the long end
57:13 Which inning are we in? And what it means for the dollar
58:52 Creative financing and Amazon's chip sale-leaseback
1:06:05 GPU depreciation: Gary Marcus on the 10-year claim
1:07:56 Fixed costs, China and fat margins
1:11:17 Why are GPU rental prices going up?
1:13:53 George Noble's shipping lesson
1:15:46 Ed Zitron's research: $200-350 billion of GPUs in warehouses
1:21:45 Big Tech's prisoner's dilemma
1:24:03 When does it blow up? Gary Marcus on meme stocks
1:24:57 Ed Zitron's close: the doom loop
1:28:18 Final thoughts
👍 Like, comment, and subscribe to support independent market insight.
The views expressed are the speakers' own. Nothing in this video is financial advice.
#AIBubble #EdZitron #GaryMarcus
20 Comments
Yep, that has the Smell of AI Slop Mathmatics all over it!!
Exactly.
It's been shown that any AI model has to hallucinate otherwise it's vulnerable to injection attacks.
Because any model can hallucinate, it can't be fully trusted, therefore a human has to be in the loop.
If a human has to be in the loop why would companies pay for a frontier model that might be a little better when they can run a local model that is just good enough to assist that human?
(In context, they've had their disagreements in the past, sometimes along personal lines)
These are the guys beating it to catchy music.
Ed Zitron: Microsoft Has $80 Billion Of AI Chips It Can't Plug In
Market Talk with George Noble
197,563 views
The AI bubble panel: Ed Zitron, Gary Marcus and Julien Garran join George Noble, Jack Gamble (Nobody Special Finance) and Alessandro (Risk Takers) to ask whether the AI boom is the biggest misallocation of capital ever, and who is left holding it if it breaks.
Ed Zitron explains why he says you can't trust Anthropic's or OpenAI's numbers until they are audited, why $1.3 trillion of compute commitments don't add up, and his estimate that $200 billion to $350 billion of GPUs are sitting in warehouses and unpowered data centers, and why that is nothing like the dark fiber of the dot-com bubble.
Gary Marcus explains why "rogue AI" is really badly built software, why large language models can't reliably follow instructions, and why he says AI stocks now trade like meme stocks.
Julien Garran of MacroStrategy Partnership lays out the $1.7 trillion Big Tech has already spent on AI capex, another trillion a year planned, and his estimate that on a generous depreciation schedule it wipes out 98% of their non-AI income by 2034.
Jack Gamble breaks down the creative financing behind the data center build-out, including Amazon's chip sale-leaseback, and the Navier-Stokes proof story. George Noble asks what it means for gold, the dollar, Treasuries and the Fed when it unwinds.
We also get into OpenAI, Anthropic and the Anthropic IPO, Nvidia, Broadcom and Google's TPUs, Microsoft, Amazon and Meta, GPU depreciation and rental prices, private credit, pensions and 401(k)s, how this compares with the dot-com bubble, and what it all means for gold and the US dollar.
🎙️ The panel
Ed Zitron: Where's Your Ed At → https://www.wheresyoured.at | Better Offline → https://www.betteroffline.com
Gary Marcus: Marcus on AI → https://garymarcus.substack.com
Julien Garran: MacroStrategy Partnership → https://www.macrostrategy.co.uk
Jack Gamble: Nobody Special Finance → @nobodyspecialfinance
Alessandro: Risk Takers → @officialrisktakers
📅 Recorded: October 2, 2026
🎧 Listen on Spotify → https://open.spotify.com/show/2Y1GCTD...
🎧 Listen on Substack → https://georgenoble.substack.com/
🎧 Listen on Apple podcast → https://podcasts.apple.com/us/podcast...
🎟️ Register now for the George Noble Best stock ideas Online Summit Conference → https://noble-capevents.com/
🔗 Follow George Noble:
X: https://x.com/gnoble79
⏱️ Chapters
0:00 Coming up
1:00 Meet the panel
1:36 Is "rogue AI" real? Gary Marcus on AI agents and safety
6:58 Air-gapped computers and the AI doom stories
8:31 Why AI models can't reliably follow instructions
10:53 OpenAI, Codex and the Navier-Stokes proof
15:27 Why Ed Zitron doesn't trust Anthropic's numbers
18:14 Broadcom, Google TPUs and where the systemic risk sits
21:14 ARR: the revenue number that isn't in GAAP
25:56 Julien Garran: $1.7 trillion spent, and what has to pay it back
31:04 Why isn't Meta's Muse the killer app?
34:33 The 404 Media report on Muse
39:33 AI agents, crypto and the "agentic bank run"
44:14 Revenue isn't profit: $1.3 trillion of compute commitments
47:39 Gary Marcus: AI doesn't work like ordinary software
49:39 The economic fallout of AI capex
53:06 Gold
53:48 The Fed, Treasuries and the long end
57:13 Which inning are we in? And what it means for the dollar
58:52 Creative financing and Amazon's chip sale-leaseback
1:06:05 GPU depreciation: Gary Marcus on the 10-year claim
1:07:56 Fixed costs, China and fat margins
1:11:17 Why are GPU rental prices going up?
1:13:53 George Noble's shipping lesson
1:15:46 Ed Zitron's research: $200-350 billion of GPUs in warehouses
1:21:45 Big Tech's prisoner's dilemma
1:24:03 When does it blow up? Gary Marcus on meme stocks
1:24:57 Ed Zitron's close: the doom loop
1:28:18 Final thoughts
👍 Like, comment, and subscribe to support independent market insight.
The views expressed are the speakers' own. Nothing in this video is financial advice.
#AIBubble #EdZitron #GaryMarcus