The Twenty Minute VCOpenAI's Multi-Billion Deal with AMD & Polymarket, Vercel and Supabase Raise Mega Rounds
At a glance
WHAT IT’S REALLY ABOUT
OpenAI power plays, mega-rounds, and venture liquidity in AI era
- OpenAI’s AMD warrant deal is framed as a power-and-leverage move where OpenAI uses distribution and demand to extract upside, contrasting with NVIDIA’s ability to take equity because it holds chip monopoly power.
- The hosts argue OpenAI’s DevDay “apps in ChatGPT” and agent tooling are directionally important but demos lacked a killer use case, echoing past hype cycles (e.g., Slack integrations, custom GPTs) that didn’t fully change workflows.
- They debate whether today’s enormous seed/early rounds (e.g., $1B at $5B pre) “break venture math,” concluding venture is forgiving on price due to variance—but not infinitely forgiving when entry prices imply only modest upside even for massive outcomes.
- IPO and liquidity realities are tightening: Snyk is near the low end of 2025 IPO thresholds, PE bid appetite is weaker than expected, and LPs/endowments selling stakes signals growing pressure for secondaries and portfolio liquidity.
- Vercel/Supabase mega-rounds are described as “Captain Obvious” developer-trend infrastructure bets, while Polymarket’s $2B strategic investment is treated as a deregulation-driven regime change story and a sign of shifting capital formation (including SPAC terms improving).
IDEAS WORTH REMEMBERING
5 ideasOpenAI is leveraging user demand as strategic bargaining power.
Even while burning cash, OpenAI can “bestow market cap” on suppliers by committing to massive chip purchases, enabling it to demand warrants from weaker partners like AMD while conceding equity to stronger monopolists like NVIDIA.
The GPU layer is behaving like a rare component monopoly—until a second source is credible.
Buyers normally pressure component vendors to cost-plus margins, but NVIDIA’s scarcity and lock-in invert the stack economics; AMD is positioned as the necessary second source, similar to IBM-era Intel/AMD dynamics.
DevDay’s app marketplace/embeds may be useful, but UI and workflow reality will limit displacement of standalone apps.
Natural language interfaces struggle with multi-option, high-context tasks (e.g., travel booking, deep app workflows), so ChatGPT may absorb “quick answers” while heavy engagement still happens inside the native product.
Agent tooling threatens some automation platforms, but enterprise-grade agents will still require orchestration layers.
An 8-minute demo can commoditize simple agent creation, yet production use demands governance, integrations, monitoring, and complex workflows—leaving room for specialized vendors unless OpenAI fully commits to the surface area.
Venture is highly forgiving on price—until it isn’t.
Because outcomes are power-law, overpaying can be survivable; however, if entry prices imply only ~5x even for “wildly amazing” outcomes, portfolio-level math breaks when only a minority become extraordinary.
WORDS WORTH SAVING
5 quotesPaul Graham was right, Sam Altman understands power. He has more power than AMD, so he took 10% of their company for the privilege of buying stuff, selling stuff to him, and he has probably less power than NVIDIA, so he let them get equity for the privilege of selling him chips.
— Rory O’Driscoll
NVIDIA is making so much money, it's almost incomprehensible. So I think Jensen knows he's got to give up some of it.
— Jason Lemkin
The great thing about venture, it has maximum variance, which means that it is the most forgiving of getting the price wrong because you can get... because you have exponential growth on your side.
— Rory O’Driscoll
Liquidity doesn't evaporate 'cause people run out of money. Liquidity evaporates 'cause people get scared and wanna keep their money.
— Rory O’Driscoll
The more you do this, the more you just say to yourself, you just need to do big, exciting deals in trends that are absolutely obvious, and every time you try and make it harder than that, you just, you lose money.
— Rory O’Driscoll
High quality AI-generated summary created from speaker-labeled transcript.
