Bill Maris just put a number on a week of vibes: 4.76x DPI for small funds, 2.42x for the big ones. The megafund story finally has a number on the other side. And it's the same week Brad Gerstner was telling everyone to hold their position. Funny how the math says the opposite of his book. This is Tech Podcast Podcast. Today: Maris on fund size, Google's pricing sword, and the first malicious-AI story of the week from Security Now. Plus Maris calling AI "still Zork" — which, after a week of Suleyman selling superintelligence, feels like the cold shower we needed. Let's start with the arithmetic. Start with the DPI split, because DPI means actual distributions to paid-in. Maris has sub-$750M funds nearly doubling the returns of the megafunds. That's the whole game. The a16z pitch is "capital flows through compute, so you need scale." Maris is at a Liquidity Conference saying scale is the thing dragging your returns down. And it lands directly on the token-cost story — Uber burning its AI budget by April, Microsoft pulling Claude Code licenses. If you can't price compute, size doesn't save you. Which is where the "Google holds the sword" line gets scary. Maris says Google can cut token prices eighty percent on its own infrastructure. One incumbent reprices the entire market overnight. That reframes Nadella's whole token-as-compounding-asset idea. You can't compound something a competitor can devalue by eighty percent on a whim. Right, so half of this week's "AI is a durable moat" commentary turns into "you're hostage to Google's pricing team" commentary. Maris said it with a mechanism, not a metaphor. And on the timeline — "AI is still Zork." From a credentialed investor, that's the sharpest counter to Suleyman's imminence pitch we've gotten. Text adventure versus superintelligence. One of those guys is selling a frontier lab and one's standing at a liquidity event with no product to pump. I know whose timeline I'm pricing in. Volpi backs the practitioner read on the Uncapped episode — AI genuinely reshaping software development costs. So it's real capability and a runaway bill at the same time, depending on who controls the price. Zork with a metered API. That's the honest version of 2026. Then Security Now goes the other direction — Anthropic's red team work, mapped to MITRE ATT&CK. First actual malicious-AI taxonomy this week instead of threat theater. And it's the accountability question from a new angle — if attackers move faster than defenders, who eats it when an autonomous process goes sideways? Nobody's holding that bag yet. MITRE at least gives it a vocabulary, which beats Suleyman's "won't take your job" reassurance from earlier in the week. Name the behavior, then argue about liability. Queue Maris for the number. The rest of the rundown is people explaining why the number can't be true. From Podcast Alpha:
Sub-$750M venture funds average 4.76x DPI vs 2.42x for funds above $1B. 95% of top-10% performers are smaller vehicles. This is arithmetic, not preference. Google could cut token prices 80% using its search cash machine and make OpenAI and Anthropic’s business models immediately unviable.
4.76x DPI for sub-$750 million funds versus 2.42x for the billion-plus crowd. The line reads like a tax return. Maris is sitting at a liquidity conference doing arithmetic that makes most of this week's mega-fund cheerleading look like cope. And it lands right on the compute-flywheel argument we tracked Monday. If 95% of top-decile performers are smaller vehicles, the whole "you need a megafund to bankroll tokens and GPUs" frame has a structural counter sitting next to it — with a number, not a vibe. The Google line is the one that should scare people, though. Maris says Google could cut token prices 80% off its search cash and make OpenAI and Anthropic's models unviable overnight. Quote: "If I were Google, that's what I'd do." That extends the accountability question we kept hitting all week — who owns the P&L when the platform you're built on can reprice the floor out from under you? Nadella's compounding-token-capital framing looks very different if one incumbent holds that sword. And then the gut-check: he calls current AI the Zork era. Command-line, no persistent memory, session resets. One frontier CEO this week was selling superintelligence; Maris is calling it a text adventure from 1980. There's also the access piece: Maris is tying late AI IPOs at elevated valuations to passive 401(k) holders, who get the overpriced shares forced on them through index inclusion. No choice in the matter. So the small funds capture the 4.76x in private, and retirement savers buy the markup automatically when it finally goes public. Sacks and Maris both call it indefensible, and the math agrees with them. Crypto Briefing writes:
Building a successful venture firm requires focusing on significant macro trends like AI. - Past successes can hinder innovation in new ventures by reinforcing outdated methods. - The AI era is fundamentally changing software development costs and assumptions. - Investors clinging to outdated software development models risk poor investment decisions.
Volpi's whole frame here is that software's cost assumptions just got blown up by AI — and investors still pricing companies on the old build cost are gonna make bad calls. That's the one takeaway in this rundown that actually has teeth. It's the second practitioner this week on that exact point. Nadella had the token-capital version on June 8 — Volpi's coming at it from the venture side, saying if you don't have the technical fluency to model the new cost curve, you can't even price the deal. Which is a polite way of saying half of VC is underwriting software economics that don't exist anymore. Coming right after the Maris piece — small funds beating megafunds on the math — the spreadsheet keeps embarrassing the keynote today. What I want from a Volpi episode is the actual number, though. "AI changes the cost" is true and vague. Does Jack Altman push him to say by how much, and for what kind of company? That's the difference between a useful episode and a comfortable one. And the brand-reputation stuff is the soft part — "you need a good name to attract founders," sure. Volpi at Index has earned that. But it's the chapter you skip to get back to where he's quantifying the build-cost collapse. Steve Gibson and Leo Laporte, writing in Security Now:
Discover how Anthropic's secretive red team and the MITRE ATT&CK framework are mapping the chilling rise of malicious AI use, revealing cyber threats that now move faster than defenders can respond.
Security Now's got Anthropic's red team report — a full year of cataloguing how people actually abuse Claude, and they mapped it against MITRE ATT&CK. That's the part that earns the coverage: real attacker tactics, an actual taxonomy, not just threat theater. And buried in the same episode — researchers built the first AI-enabled internet worm. Code that spreads itself, autonomously, using AI. That was a thought experiment six months ago. Which sits in a funny tension with where we landed an hour ago. Maris is out there calling AI "still Zork" — text-adventure territory. Meanwhile the security desk is logging a worm that propagates on its own. Right — it can't reliably book your dentist appointment, but it can write malware that moves faster than the people paid to stop it. Both things are true, and that's the whole genre right now. Gibson's headline on the Anthropic findings is basically one word: it's bad. And the MITRE mapping means they can point at specific techniques defenders already track, not just vibes. From Aaron Stanley at Brazil Crypto Report:
The stablecoin fragmentation problem is more serious than it looks. USDT and USDC promise universality, but every market trades, converts, and accounts for them differently. Checker sits in the middle and makes that complexity invisible for the financial institutions that need to move money across borders.
Okay, after all that Maris arithmetic, this one's a palate cleanser — Checker, Jack Chong, a stablecoin liquidity network wiring up FX banks and neobanks across emerging markets. The framing I actually like here is "the global financial system runs on duct tape." It's a concrete problem — fragmented liquidity between counterparties — not a token-price moonshot. And the pitch is a network of regulated counterparties. Which, fine, but the whole crypto promise was disintermediation, and now the missing piece is... more intermediaries with licenses. The detail that makes it worth listening is the local part — local relationships, local capital, local market structure, with one global product underneath. That's a real structural claim about why stablecoins haven't moved money the way people promised. Chong's path is the tell on whether that's real — Hong Kong, Oxford, learned Arabic in Jordan for a diplomacy career, then ends up building stablecoin rails in New York. If anyone's going to argue emerging-market liquidity is a relationships problem, it's the guy who came up through diplomacy. Here's Brandon Zemp at BlockHash:
For episode 742 of the BlockHash Podcast, host Brandon Zemp is joined by Chi Zhang, co-founder and CEO of Kite, which is building the base layer for the agentic internet. Her extensive background encompasses AI, big data, and product management.
Episode 742, BlockHash, Chi Zhang from Kite — building the "base layer for the agentic internet." Which is a sentence I've now heard about forty times this cycle with forty different chains attached to it. The resume's real though. Databricks data engineering, founding AI leader at dotData, a Berkeley Ph.D. in ML statistics. She's not a pitch-deck founder. Sure, and that's exactly why "base layer for the agentic internet" frustrates me — someone who shipped at Databricks should be able to tell me what Kite actually settles. Payments between agents? Identity? Who's the counterparty when the agent overspends? That's the part I'd press on. After the Maris piece earlier — Google able to cut token prices 80% on its own rails — an agent-payment layer is interesting only if it survives an incumbent repricing the whole market underneath it. Right. And the accountability question doesn't go away just because you put it on a chain. Same broken-liability problem from the Anthropic red-team angle, different wrapper — who eats it when the autonomous process goes sideways? I'd queue it for the architecture, not the tagline. If Zhang names the actual mechanism — settlement, spend limits, agent identity — there's a real episode in there. If it's vision talk, it's the third "agentic" pitch this month. Got thoughts on today's episode, a story we should track, or a correction we should make? Send us a note at techpodcastpodcast at lantern podcasts dot com. We really do read them.
We've put links to every story from today's briefing in the show notes, so if something caught your ear, you can jump straight to the source and read more.
That's Tech Podcast Podcast for today. This is a Lantern Podcast.