TBPN Diet: The AI Chip Squeeze, Dirty Soda Mania, OCC Greenlights Palmer Luckey’s Erebor
Date: February 7, 2026
Hosts: John Coogan & Jordi Hays
Episode Theme:
A dense, fast-paced roundup of the week's most pressing tech stories: the critical bottlenecks in scaling AI (chips vs. energy), the surprise dirty soda craze out of Utah going nationwide, and breaking news of Palmer Luckey’s Erebor receiving OCC approval to operate as a bank. The panel draws on interviews, industry commentary, and their signature irreverent tech-insider humor.
Main Themes & Episode Overview
- AI’s Growth Bottlenecks: Dissecting whether semiconductors (chips) or energy is the main constraint as hyperscalers pour record-breaking investments into AI infrastructure.
- Consumer Phenomena: The unexpected rise of “dirty soda” from Utah into a viral, possibly national sensation.
- Fintech Disruption: Palmer Luckey’s Erebor gets regulatory approval, promising a new kind of tech-native, startup-friendly bank.
Key Discussion Points & Insights
1. The AI Infrastructure Bottleneck: Chips vs. Energy
-
Recursive Scaling in AI:
The hosts point out that recent advancements—like models feeding their outputs back into new inputs—are causing "recursive compounding" within AI development. The field feels "on the cusp of a software singularity" but constrained by physical supply bottlenecks.
[00:18]“There’s clearly signs of a takeoff... but there’s a whole bunch of recursive compounding elements that are starting to form.” —A
-
Where’s the Bottleneck Now?
-
Chips as the Current Constraint:
Based on their conversations with Sam Altman and others, chips—not energy—are the present limiting factor. Energy may become the constraint around 2027, but for now, compute (i.e., semiconductors, fab capacity) is the “damper on the party.”
[05:13]“Sam Altman put it this way... right now it’s chips. It may get solved on its own; normal capitalism may solve it. But I think somehow deciding as a society that we are going to increase the wafer capacity of the world ... would be a very good thing to do.” —A
-
Fab Capacity & Supply Chain Realities:
- Advanced fabs cost $30-80B and take 3–5 years to build.
- Deep within chip manufacturing lies another choke point: ASML, supplier of EUV lithography machines, ships just 50–60 machines per year; each costs $350M.
- TSMC holds 90% share of advanced chip manufacturing, with Samsung and Intel lagging.
-
Energy as the Next Bottleneck:
The U.S. energy sector’s inability to keep up with exponential demand is seen as a looming issue, but not the one throttling progress in 2026.
-
-
Ben Thompson’s Geopolitical & Corporate Warning:
- Tech companies risk much more by not developing alternatives to TSMC—even if “anchor customer” deals with Samsung/Intel are scary, staying tethered to TSMC is riskier long-term.
[08:28]
“...Avoiding the risk of working with someone other than TSMC incurs new risks that are both harder to see and also more substantial... future foregone revenue at the end of the decade is going to cost exponentially more...” —Ben Thompson (quoted by A)
- Tech companies risk much more by not developing alternatives to TSMC—even if “anchor customer” deals with Samsung/Intel are scary, staying tethered to TSMC is riskier long-term.
[08:28]
2. Hyperscaler CapEx Frenzy & Market Dynamics
-
Massive 2026 AI Investments:
The panel reviews 2026 capital expenditure plans:- Google: $175–185B (vs. $120B previously estimated)
- Meta: $115–135B
- Tesla: $20B (from $11B)
- Amazon: going to $200B (from $145B) [11:40]
-
Shareholder Tensions:
Despite huge AI bets, Amazon’s stock underperforms, spooking investors even as business metrics stay strong. -
Meta’s AI Activity:
Listeners question the rationale for Meta’s spend—“What exactly is Meta buying? They don’t do anything with AI.”“Open up reels. It’s AI.” —A [12:19]
-
Elon Musk & Space Data Centers:
Musk envisions terawatts of space-based compute—30 months for first deployments, but panelists are skeptical of practicality versus physics and economics.
[13:40–14:07]“He keeps repeating, ‘if you just look at physics’... he doesn’t like talking about thinking in decades.” —C
-
Historic Scale:
Total capex for AI data centers in 2026: $660 billion.
[15:42]“This is more than what we spent on the U.S. Interstate Highway System... the biggest project in the history of capitalism.” —A
3. Viral Consumer Trend: Dirty Soda Mania
-
Origin:
Dirty soda—soda spiked with flavored cream and fruit puree, served on pebble ice—exploded from Utah after an Olivia Rodrigo Instagram post. -
National Chains Jump In:
McDonald’s and Taco Bell pilot “dirty” drinks; Sonic tries “make it dirty” options. Swig, the original chain, now has 140 locations and did $100M in sales last year. [18:18]“#dirtysoda is the hashtag big chains want in.” —B
-
Business Model Concerns:
The panel doubts the defensibility of dirty soda chains: [19:25]“If the product becomes popular, everyone has it immediately. You have no real IP. You’re just using sodas off the shelf.” —B
-
Caffeine & Mormon Culture:
The supposed lack of caffeine appeals to Utah’s market, but, as quipped by C, “Isn’t the point that it doesn’t have [caffeine]? Right. Cause you’re Mormon.” [20:01]
4. AI Agents, Productivity, and the "AGI-Pilled" Workplace
-
Kevin Roose’s Take:
“Once you see an agent autonomously doing stuff for you, it’s so instantly clear that all computer-based work will be done this way.” —quoted from NYT’s Hard Fork [22:34]
-
Panel’s Prediction:
AI will become the main workspace interface for most white-collar workers within two years, with “parallel agents deployed at downright Soviet levels.”
[22:36]“I think AI is going to become the home screen of a ludicrously high percentage of white collar workers in the next two years.” —A
5. Policy, Automation, and AI vs Software
-
Congressional Hackathons:
Humorous suggestion that Congress should be forced to use AI for 30 minutes to grasp future-of-work transformations. -
Physical Mail, AGI-style:
Discussing extreme automation:“I never want to open up a physical piece of mail again.” —B [24:56]
John suggests using virtual mailboxes and AI-powered workflows. -
Jensen Huang Pushes Back on ‘AI Kills Software’ Myth:
[25:15]“It is the most illogical thing in the world... If you were a human or [an] artificial general robot, would you use a screwdriver or invent a new screwdriver? ... The answer, obviously, is to use tools.” —Jensen Huang
- Disagreement from panelists: AI agents will build for themselves if commercial tools get too expensive.
6. Lighter Moments & Viral Quips
-
AI.com and Crypto.com:
Crypto.com's founder buys AI.com for $70M, plans to launch a personal AI agent for everyday consumer tasks.
[28:48] -
Pint Forensics:
The hosts scrupulously tally who drank what on the “Cheeky Pint” crossover with Elon Musk, even involving “pint size” analysis.“This is the kind of hard hitting analysis that you can only get on this show.” —B [28:14]
7. Breaking News: Erebor Bank Gets OCC Greenlight
-
Palmer Luckey’s Erebor:
- OCC greenlights Erebor as a new bank, aimed at startups and high net-worth tech clients.
- Erebor already notched a $2B funding round (7x book value), with a recent raise at $4B.
[29:26]
“Hobbit Inspired startup becomes first new bank greenlighted 2.0. One of the most insane headlines.” —B
“Erebor will cater to startups and high net worth individuals. ... Think of us like a farmer’s bank for tech.” —Palmer Luckey (quoted by A)
Notable Quotes & Timestamps
- “There’s this like TikTok going back and forth in the AI supply chain. What’s the key bottleneck to growth?” —A [01:09]
- “Normal capitalism may solve it... but deciding as a society that we are going to increase the wafer capacity of the world... would be a very good thing to do.” —A [05:13]
- “TSMC controls 90% of the advanced node market... and that’s why Ben Thompson... is really urging tech companies to wake up.” —A [08:58]
- “What exactly is Meta buying? They don’t do anything with AI.” —B quoting commenter [12:11]
- “This year alone is without a doubt the biggest project in the history of capitalism.” —A [15:42]
- “What is dirty soda? It’s like soda and they put like cream in it and stuff.” —C [17:42]
- “Maybe we need more long weekends for AI adoption.” —B [23:14]
Segment Timestamps
| Segment/Topic | Timestamp | |-----------------------------------------------------------|-------------| | Market banter, fear/greed, AI epiphany and takeoff | 00:00–01:59 | | Chips vs Energy — Sam Altman, Ben Thompson, bottlenecks | 01:59–13:11 | | Hyperscaler Capex (Google, Meta, Amazon, etc.) | 11:22–15:04 | | AI in Space & Musk’s Vision | 13:12–15:37 | | Dirty Soda National Craze | 17:17–21:24 | | AI agents and AGI workplace predictions | 22:13–23:36 | | Software vs AI, Congressional Hackathons, automation | 24:03–26:37 | | Fun: Cheeky Pint Crossover, AI.com | 27:16–28:48 | | Breaking News: Erebor Bank OCC Approval | 29:26–30:40 |
Conclusion
This episode delivers a whip-smart dissection of where the AI industry’s real chokepoints are, injects dry wit into the viral rise of dirty soda, and closes with a landmark for fintech as Erebor gains regulatory approval. Through lively banter, sharp industry analysis, and deep dives into both capital flows and latte trends, TBPN continues to make the week’s tech chaos digestible... with the occasional detour into pint glass forensics.
Subscribe for more: 5 stars only.
