Lead storyThe Information
A 109-page SEC filing shows Meta created a new non-voting share class so it can raise AI capital at massive scale while preserving Mark Zuckerberg's more than 60% voting control, via a one-time dividend of one new Class C share per Class A/B share. Management told the board in May it may need "hundreds of billions of dollars" of capital starting as early as fall 2026; the filing says raises by Alphabet, SpaceX, Anthropic, and OpenAI are making big capital "increasingly" hard to secure before rivals soak up funding. The overhaul was also shaped by Zuckerberg's worry that California's Proposition 40 wealth tax, on next month's ballot, could force him to leave the state.
Why it matters
Meta is the first mega-cap to say plainly that AI needs hundreds of billions and that even the richest balance sheets cannot fund it from cash flow alone. Selling non-voting shares is a way to raise at scale without losing founder control, and the filing frames the race as a scramble for scarce capital, not scarce chips. For founders and investors, the binding constraint has shifted: whoever can raise the most money at the best terms wins the capacity.
Story 02The Information
DeepSeek is raising roughly $15 billion with CATL and Tencent in the mix, per AI Agenda's Deals & Debuts. Nebius lined up a $3.5 billion credit facility with Citi. Flapping Airplanes raised $1 billion at a $6 billion valuation (Index, Lux). Blacksmith added $66 million at $1 billion (Sequoia) just after Sequoia led its Series C, and Coherence Neuro raised a $45 million Series A (Founders Fund, OpenAI Startup Fund).
Why it matters
Money is moving faster than ever and spreading across models, infrastructure, and applications at the same time. A $15 billion round would make DeepSeek one of the best-funded labs in the world, while Nebius borrowing billions and a $1 billion Series A show the buildout is being financed by debt and equity at record scale. For founders, the bar for "enough capital" just moved again: your competitors are raising war chests, not rounds.
Story 03OpenAI
OpenAI says its Calif team used GPT-5.6 Sol to find critical vulnerabilities in PHP, the language behind much of the web, and that Wiz independently found the same flaws. OpenAI led a coalition of maintainers, researchers, and security providers including Cloudflare to fix and disclose them, working with the White House's Super Intelligence Force, ONCD, and CISA. It committed $1 billion in subsidized Daybreak model access for under-resourced defenders, launched patchtheplanet.org, published a five-stage AI-era coordinated-disclosure framework, and is offering free Codex-for-OSS subscriptions to open-source maintainers.
Why it matters
Frontier AI is now finding internet-scale bugs faster than attackers do, and the industry is writing a new disclosure playbook around that fact. OpenAI calls it the "Defender's Window": it only exists while labs stay ahead of attackers, and OpenAI expects similar findings to increase sharply in the coming months. For security founders, this is both a threat and a market: the same models that find the bugs can be sold as the defense.
Story 04OpenAI
OpenAI updated its Usage Policies, effective November 20, 2026, requiring services that give direct access to its models, such as model routers or AI gateways, to get OpenAI's review and approval before offering that access. Partners must keep safeguards consistent with OpenAI's own, investigate violations, and coordinate enforcement across services. The post also introduces SafetyIDs, a way to tell end users apart in abuse investigations without learning their identities, and names GPT-6 Astra as the new model expanding what developers can build.
Why it matters
OpenAI is tightening control over how its models reach the market even as it widens distribution through the clouds. Gateway operators are no longer neutral pipes: they need OpenAI's blessing and must police their own users to OpenAI's standards. For founders building on top of OpenAI through third-party routes, the cheap anonymous proxy is going away; plan for compliance costs and direct relationships instead.
Story 05Sequoia Capital
Sequoia is leading Whop's Series D, with Shopify Ventures participating and Doug Leone joining the board; the round size and valuation were not disclosed. Whop is at half a billion dollars in run-rate revenue, with 10,000 new businesses created on the platform daily, including student founders at Stanford, MIT, Harvard, Princeton, and Columbia. The platform bundles payments, delivery, and marketing in one place and is shipping an AI-powered decision engine built on its transaction data.
Why it matters
Sequoia is betting that AI is driving a new wave of one-person businesses, and that the winners are platforms that make starting up a matter of minutes, not months. Shopify Ventures joining is a tell: the ecommerce giant sees Whop as the next layer of online entrepreneurship. For founders, the lesson is that transaction data is the moat; once you own the payments, the AI features build themselves.
Story 06The Information
a16z's crypto arm, nearly $10 billion across five funds, told LPs it will rename itself "a16z frontier" and invest beyond crypto. GP Ali Yahya, the team's first full-time member since 2017, is leaving to start his own company. The latest $2.2 billion fund raised in May was less than half the 2022 vintage of $4.5 billion; recent non-crypto bets include Kalshi and Ornn (AI compute markets). Chris Dixon has said crypto is in its "financial era" of stablecoins and agentic payments after the Clarity Act failed in Congress, and peers Haun Ventures and Paradigm broadened their mandates earlier this year too.
Why it matters
The crypto-native fund model is officially over; the surviving funds are becoming generalist tech investors. Stablecoins and agentic payments are the crypto bets that still excite top firms, and everything else is fading. For founders, a16z frontier means crypto startups will now compete for capital against AI and fintech on equal terms: the dedicated crypto premium is gone.
Story 07Sequoia Capital
On Sequoia's Training Data podcast, Mecka AI founder and CEO Josh Gao described how the company manufactures robot training data: custom sensor caps and tactile gloves, since nothing off the shelf captures force data, 100-gigabit upload lines, and hundreds of petabytes of data handled. The industry's converging recipe is roughly a million hours of egocentric human video, tens of thousands of hours of robot data, and a few hours of in-domain teleoperation. Gao argues there will be no single GPT-3 moment for robotics, just a steady unlock of commercial tasks starting with constrained industrial work. Sequoia led Mecka's $60 million Series B.
Why it matters
Robotics has no free internet to train on, so training data has to be manufactured at industrial scale, and that is the real moat. A million hours of human video plus force-sensing gloves is the new pretraining corpus. For investors, the play is the data pipeline and the sensors, not the robot body; for founders, the unlock is constrained tasks where the data problem is actually solvable.
Story 08The Information
Utopai Studios, a four-year-old Mountain View film startup, closed a $60 million round at a $1 billion valuation and already has term sheets for a follow-on at $2-3 billion. It has more than 20 projects in the works including two TV series and three full-length films debuting next year, targeting about 3,000 North American theaters. CEO Cecilia Shen, ex-Google X, says its customized MiniMax H3 video model ranks second on Artificial Analysis's text-to-video leaderboard and cuts scene-planning and CGI costs by 80-90%, while the company still pays directors and actors. For context: Higgsfield quadrupled to a $5.4 billion valuation in August and crossed $1 billion in annualized revenue, and Kuaishou's Kling AI raised about $3 billion at $15 billion.
Why it matters
AI video has crossed from tools to studios: the winning model is not selling software to filmmakers but making the films yourself. Cutting CGI costs by 80-90% while keeping human directors and actors is the formula that makes the economics work. For founders, the open question is who captures the savings: the model makers, or the studios that own the audience.
Story 0920VC
This week's 20VC roundtable covered Modal raising $750 million at a $15.75 billion valuation as inference demand booms, panic from the CoreWeave sell-off spreading across neocloud names, Anduril priced at $90 billion, Town at $1 billion, General Atlantic eyeing an IPO, and talk that Anthropic's IPO valuation could slip from $2 trillion to $1.5 trillion. All claims are show-discussed, not confirmed announcements.
Why it matters
The private markets are repricing the same AI trade in two directions at once: inference companies raise at eye-watering multiples while debt-heavy neoclouds get punished. Modal at $15.75 billion says investors still pay up for real compute demand; Anthropic's rumored IPO haircut says they want a discount for the privilege. For founders, the message is that growth still commands premiums, but leverage now commands discounts.