Newsletter · · Ashutosh Agarwal

Anthropic Races Toward a Record IPO as the Capital Markets Window Cracks Open - Capital Markets: IPOs, M&A and Exchanges - Week of August 18, 2026

Capital-markets newsletter for the week of August 18-23, 2026. Anthropic is racing toward what could be the biggest IPO ever, and the podcast tape traced the machinery underneath it: the debt financing the AI build, prediction markets becoming Robinhood's biggest business, and banks buying their way into ETFs.

Capital Markets: IPOs, M&A and Exchanges

Week of August 18–23, 2026: Anthropic Races Toward a Record IPO as the Capital Markets Window Cracks Open

The traditional Wall Street rule is that the market for new stock offerings wakes up right after Labor Day. This year the alarm is going off early, and it has one name attached to it: Anthropic, the maker of the Claude chatbot. Almost every finance and tech podcast this week circled back to the same question: how big can this get, and does it drag the rest of the IPO market open behind it?

Underneath the headline, the more interesting story is the machinery: who is quietly financing the AI boom with borrowed money, why prediction markets (websites where you bet on real-world events) suddenly became the hottest business in retail trading, and why the big banks are buying their way into the ETF business instead of building. Let's walk through it.

1. The IPO Desk

Anthropic: The Biggest Listing Anyone Has Ever Tried

Here is the plain version of what was reported this week. Anthropic is preparing to go public and wants to raise as much as, or more than, SpaceX did in June, and SpaceX raised a record $75 billion. On Bloomberg Tech's "Anthropic Preps for Blockbuster Public Listing" (Aug 21), reporter Shireen Ghafari laid out the two numbers that will define the whole deal:

"We know that Anthropic is expecting to have an IPO that could match or exceed the size of SpaceX's record-setting IPO... We also know we found out a few more details about their losses, that their net losses in 2025 were nearly $42 billion. And that's a five-fold increase from 2024."

So: a possibly-historic fundraise, sitting on top of losses that are also historic. That tension, enormous demand for AI versus the enormous cost of providing it, is what the whole market is waiting to see resolved when Anthropic files its S-1 (the detailed financial document a company must publish before going public). Bloomberg's deals reporter Liana Baker explained that the company is already on file confidentially, so the public version could appear fast (analysts are saying this filing could hit as soon as next week), and that the open question is whether it prices at its last private valuation of $965 billion or pushes through $1 trillion or even $1.5 trillion (Bloomberg Tech).

Why rush to go public when you already have plenty of cash? Because, as Baker put it, an IPO gives you "a repeatable process to raise large amounts of capital", and training a single cutting-edge AI model "can cost billions of dollars." When you need to keep raising, a public stock is the most reliable tap.

The revenue growth behind the hype is genuinely hard to believe. On an earlier Bloomberg Tech (Aug 18), correspondents pegged Anthropic's annualized revenue at $65 billion as of July, up from $47 billion in May (Bloomberg Tech). On ARK Invest's "For Your Innovation" podcast (Aug 19), analyst Frank Downing gave the comparison that makes your head spin:

"Salesforce is a roughly $45 billion run rate revenue business... and they are over 25 years old. And Anthropic added more than a whole Salesforce in run rate revenue in less than a year."

Downing's math on valuation is worth sitting with, because it cuts against the "this is insane" reaction. If Anthropic is at a $70-80 billion revenue run-rate (up from $9 billion at the start of the year) and reaches roughly $100 billion by year-end, then a $2 trillion IPO would be about 20 times run-rate revenue, "a lower multiple than a lot of things," he noted, and actually below public AI darlings like Cloudflare and Palantir, which trade in the 30s on next-year sales. The genuine debate is about 2027: on the All-In podcast, investor Gavin Baker floated $400-500 billion of revenue next year; investor Brad Gerstner called that too high and said $200-250 billion. Either number would be extraordinary (FYI - For Your Innovation).

The most useful color came from an actual early investor. On Squawk on the Street (Aug 21), Menlo Ventures' Matt Murphy, who put money into Anthropic early, described a business unlike anything he'd seen:

"When have we ever seen a company grow 10x at this kind of scale, have great margins, be operating cash flow positive? Just a tremendous business."

His most telling detail was about how much companies are willing to spend on AI: "You look at the number of companies that are spending 1 or 10 million a year on AI... usually when these waves take off, somebody's spending 50 or 100K a year on the new technology, not millions." But Murphy also explained why the rest of the IPO pipeline is stuck behind Anthropic and OpenAI: private markets are so flush with cash right now that most companies would rather stay private and mature than face public-market scrutiny, "it really is fine to stay private longer" (Squawk on the Street).

Why it matters: Anthropic is not just one deal. It's the test of whether public investors will fund an AI lab that loses tens of billions a year. If the S-1 shows a path to profit (there were even rumors of a positive operating quarter), it blows the IPO window wide open. If it shows uncontrolled cash-burn, it could slam shut just as fast.

Unitree: China's Robot IPO That Went Vertical

While America waited on Anthropic, China delivered the week's actual fireworks. Unitree, a humanoid-robot maker, listed on Shanghai's STAR Market and its shares roughly quintupled on day one. The exact figures varied by show, but the shape was consistent: the company raised around $904-905 million and its first-day pop ran anywhere from +460% (per The Rundown) to +542% (per Morning Brew Daily), vaulting its market value to roughly $50-60 billion. Morning Brew called it one of the most oversubscribed IPOs in the market's history: demand exceeded shares available by thousands of times, meaning most retail applicants got nothing.

What makes the number eye-watering is the underlying business. On Rich Habits (Aug 21), the hosts noted Unitree earns only about $250 million in revenue and $40 million in profit, so the market is paying for the story (Beijing treating humanoid robots as a national priority, plus Morgan Stanley's projections for Chinese humanoids), not the earnings (Rich Habits). Treat it as a sentiment gauge for "physical AI," not a valuation you'd want to defend.

Lintris (LYNX): The US IPO That Actually Happened, and Had to Shrink

Here is the reality check under all the trillion-dollar talk. The one real US company that rang the bell this week had to cut its deal to get it done. Defense-tech firm Lintris debuted on the NYSE under the ticker LYNX, raising about $298 million, down from what it hoped. It had marketed 24 million shares at $19-22; it ended up selling 17 million at $17.50. CEO Brian Morrison was candid on Squawk on the Street about why:

"I think the downsize just indicates that the sector has been under a little bit of stress. Candidly, we are delighted by the book of investors that we have in today... We're here to drive long-term value, not one-day share price."

The company is a "merchant supplier" to defense primes (it makes cooling technology for high-power radar and antennas), it grew its backlog 94% in the first half and 30% organically before acquisitions, and it plans to cut leverage from roughly 6x earnings down to about 2.4x by the close of the offering, using most proceeds to pay down debt (Squawk on the Street). The read-through: for everyone who isn't a frontier AI lab, the window is open but choosy. Deals are getting done at a discount, not a premium.

A Refresher on How IPO Pricing Really Works

If you want to understand why Lintris cut its price rather than push, FinPod's "IPO Timing and Pricing" (Aug 18) is the best plain-English primer of the week. The counterintuitive point: companies deliberately price below what demand would bear, leaving money on the table, to engineer a healthy first-day "pop" and goodwill. As the hosts put it, "The goal on pricing day isn't actually to extract the absolute maximum dollar for every single share. The goal is a successful, sustainable deal and a healthy aftermarket."

Their case studies double as recent history: Uber priced aggressively at $45 in May 2019 into a nervous, trade-war-rattled market and fell 7-8% on day one, the worst first-day dollar loss for a US IPO up to that point. Arm Holdings priced at $51 in September 2023, floated only ~10% of the company to manufacture scarcity, and popped 25%. CoreWeave, in a jittery March 2025, swallowed its pride, priced at $40 (below its $47-55 range), closed flat, and then rallied for months. The lesson Lintris clearly took to heart: "A completed deal at a realistic clearing price is infinitely better than a shattered deal at an ambitious price" (FinPod). And a caution for anyone chasing IPO pops: on Metcalf Money Moment (Aug 19), the host walked through SpaceX's own June debut: it opened at $135, ran to $211, then fell back to $114 within a month (Metcalf Money Moment).

2. The M&A Desk

Dealmaking was busy this week across very different corners of the market. A quick tour:

  • Kirin buys Jamieson Wellness for ~$1.4 billion. On the Joshua Schall Audio Experience (Aug 18), the acquisition was framed as Japanese brewer Kirin executing its "Innovate2035" strategy, pivoting away from a shrinking domestic beer market toward a goal of $3 billion in global health-science revenue by 2035, with Jamieson (vitamins and supplements) giving it a North American anchor (Joshua Schall).
  • Silver Lake circles Workday. Reports of private-equity firm Silver Lake in take-private talks with software maker Workday pushed the stock up sharply (a roughly 19% reaction) and lifted Workday's value past $51 billion (The Compound and Friends). On The Chad & Cheese Podcast (Aug 21), the host was scathing about what PE ownership tends to do to enterprise software (cut R&D, load on debt, "nickel-and-dime captive customers"), while conceding the takeover interest itself reflects a "buying opportunity" after this year's software sell-off (The Chad & Cheese Podcast).
  • The AI land-grab: Stripe/OpenRouter and SpaceX/Cursor. Payments company Stripe agreed to buy OpenRouter (which routes AI requests to different models) for roughly $7 billion, a huge markup from its $1.3 billion valuation just three months earlier in May. On Limitless (Aug 19), the hosts noted OpenRouter processes about 100 trillion tokens a month with only ~$50 million of annual revenue and a 5.5% take rate, so Stripe is paying up for infrastructure and strategic position, not current profit (Limitless). In the same conversation on RiskReversal Pod (Aug 19), investor David Waltcher tied it to a broader pattern of "hyperscale acquirers" consolidating AI assets, including SpaceX reportedly paying $60 billion (in stock) for the coding tool Cursor (RiskReversal Pod).
  • Two closed deals that repriced real companies. Telltales' "Weekend Update" (Aug 23) walked through two transactions that landed this week. Radiopharmaceutical firm Curium agreed to buy Lantheus outright for $102.50 a share in cash plus a contingent payment worth up to $12 more (a total of roughly $8 billion, expected to close in the first half of 2027), and the agreement landed just five days after the FDA approved Lantheus's new Alzheimer's imaging agent, TauClarify. Separately, Charter closed its $34.5 billion acquisition of Cox Communications and Liberty Broadband, creating the largest internet and cable company in the country after California regulators signed off on August 13. In an unusual twist, the buyer is taking the target's name: Charter is rebranding as Cox (Telltales).
  • Bank M&A keeps grinding. The Banker Next Door (Aug 19) rounded up three: Home Trust Bank buying Blue Ridge Bank for $448 million in stock (closing Q1 2027, taking Home Trust to $7 billion in assets and 60 locations); HBT Financial buying Tri-County Financial for about $204.6 million (closing Q1 2027, adding 19 branches and pushing HBT past $8 billion, its 12th acquisition since 2007, a genuine serial buyer); and Citibank picking up rewards-fintech Card Financial for undisclosed terms the bank called immaterial against its $2.9 trillion balance sheet (The Banker Next Door).
  • Goldman buys its way into ETFs. One of the more strategically interesting deals: on Bloomberg's Trillions (Aug 20), the hosts detailed Goldman Sachs acquiring two ETF issuers, Innovator and Neos, within a couple months of each other, spending somewhere in the $10-15 billion range to buy "really, really sticky" assets in buffer and defined-outcome funds (products aimed at older investors who want downside protection). The framing: Goldman gave up on ETFs years ago, watched JPMorgan run away with the category, and is now paying a premium to catch up because "getting assets is hard." Expect more ETF consolidation: "any number is in play" now (Trillions).

3. Antitrust and Regulation

The classic FTC-blocks-a-merger story was quiet this week, but a notable investigation surfaced. On TechCrunch's Equity podcast (Aug 21), the hosts reported that the Department of Justice is investigating venture-capital giant Andreessen Horowitz (a16z) over holding board seats at competing companies, including Databricks (where Ben Horowitz sits) and Fivetran (where partner Martin Casado sits). The probe has reportedly been running for nearly a year, and the hosts flagged the unusually long duration as a hint there may be more to it than simple board-seat conflicts. Tellingly, the famously outspoken firm has stayed publicly quiet, a contrast with its loud complaints during the prior administration (Equity). Why it matters: if the DOJ treats overlapping VC board seats as an antitrust problem, it changes how the whole venture industry governs the companies it funds.

4. Exchanges and Prediction Markets: The Fastest-Growing Corner of Trading

If you only track one structural shift this year, make it this one. "Prediction markets", regulated venues where you buy and sell contracts that pay out based on real-world events, have gone from a niche curiosity to a genuine business line at the center of retail finance.

Kalshi is scaling fast. On the CoinDesk Podcast Network (Aug 19), Kalshi's head of crypto John Wang confirmed the company just raised at a $40 billion valuation, and detailed something new: Kalshi Pro, a CFTC-regulated perpetual-futures platform ("perps" are a leveraged derivative popular in crypto, previously off-limits to US traders). The traction was immediate:

"You guys did like a billion in the first week. I think it was like five billion in the first month. And so clearly, once you unlock that, people really started to flood in."

Wang said crypto is now the second-largest category on Kalshi, catching up to sports, and that the platform's user base is about half the size of crypto venue Hyperliquid's (CoinDesk Podcast Network).

The exchanges want in, carefully. On Bloomberg Talks (Aug 20), Intercontinental Exchange (ICE) chairman and CEO Jeff Sprecher explained why the owner of the New York Stock Exchange invested $1.6 billion in Polymarket rather than building its own event market: it wanted the blockchain technology and to help Polymarket fix its US compliance, not to run a prediction market itself. He was frank that constant "listing and delisting of contracts... is really not how my universe, Wall Street, operates." Sprecher also confirmed ICE is working with the SEC to tokenize the NYSE and extend its trading hours, and, separately, is exploring futures on AI compute power, timed alongside a CFTC industry survey (Bloomberg Talks). (On that compute-futures front, TechCrunch's Equity also noted CME plans to launch "Silicon Data" compute futures on October 5, pending approval, Equity.)

The plumbing story: Robinhood and Rosera. The most revealing scoop came from The Information's TITV (Aug 20). Robinhood took a 45% stake in Rosera, a small Chicago-based prediction-market exchange launched in June, founded by financial-infrastructure veteran Thomas Chippas (who previously built crypto exchange ErisX, later Cboe Digital). The reason is a lesson in market structure: brokers usually don't own the exchange they trade on (you don't trade NYSE-owned stock through the NYSE), because owning both creates conflicts, heavy regulation, and fragmented liquidity. So instead of building its own venue, Robinhood took a controlling-ish stake in a pure "behind-the-scenes" exchange it can route orders to. The stunner in the reporting:

"The revenue for Robinhood coming from prediction market is now higher than the revenue they generated from either equities trading or crypto trading."

Rosera runs on about 50 people and is already a top-five prediction-market exchange by volume, mostly on Robinhood's order flow (The Information's TITV).

The fight over who regulates all this. Not everyone is cheering. On The Paul Barron Crypto Show (Aug 21), the discussion centered on Kalshi's COO defending prediction markets at a CFTC hearing that "turned into a brawl", sparring with CME chief Terry Duffy over manipulation concerns (The Paul Barron Crypto Show). On Crypto Curious (Aug 19), the hosts laid out the core legal tangle: the federal CFTC treats these as commodity contracts, while some states (New York) treat them as gambling, a conflict that "will likely require court resolution", and noted Kalshi's backers include the New York Stock Exchange (Crypto Curious). The Wolf Of All Streets (Aug 18) added that Charles Schwab is adding prediction markets and Gemini launched its own marketplace: the field is crowding fast even as the rules stay unsettled (The Wolf Of All Streets).

5. Retail Flow: Robinhood's Quiet Takeover of Options

Zoom out from prediction markets and Robinhood's broader numbers are striking. On The Iced Coffee Hour (Aug 23), CEO Vlad Tenev said the company hit multiple records in its most recent quarter (including equities trading volume above the 2021 GameStop mania peak) and, most notably, crossed into first place in retail options:

"Options trading, all time high. We're now the top options trading platform in terms of retail market share. So we've surpassed all of the incumbents who keep just combining with each other."

He also described the leading edge of retail behavior: over 100,000 customers have set up "agentic trading" accounts, connecting AI agents (like Claude or Codex) to Robinhood's tools to execute trades automatically. Tenev, a former algorithmic trader, framed it as democratizing something that used to require serious programming. And he offered an unintentional froth indicator: the flood of customer requests for Korean stocks. "Every question I would get in an interview was, when are you going to add Korean stocks?... the fact that we were hearing from so many people at the same time does raise the alarm bells a little bit" (The Iced Coffee Hour).

6. The Plumbing Under Everything: Who's Actually Paying for the AI Build

This is the story that quietly ties the whole capital-markets picture together, and it got its clearest explanation on Facts vs Feelings (Aug 19). The short version: the AI build-out has outgrown the cash flows of even the richest tech companies, so it is now being financed with an enormous, newly assembled wall of debt, and that changes the math for every other deal in the market.

The centerpiece is a plan for a $500 billion compute-financing platform, announced as a memorandum of understanding on August 11 by five of the biggest names in finance: BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR. As co-host Sonu Varghese explained, the platform will lend to a new category of borrower: "NeoClouds", non-traditional cloud operators that exist only to build and run AI systems. Those operators borrow money, buy mostly NVIDIA chips, and the debt sits in a special-purpose vehicle collateralized by the chips themselves and the customer contracts to lease that computing power. The clever, and slightly unnerving, part is how it moves risk around:

"The market sort of likes this because it's like, okay, we're taking this tail risk off of NVIDIA's balance sheet. We're just shoving it on everybody else. And that's the financial engineering."

You can see it in the price of insuring these companies' debt. NVIDIA's credit-default-swap spread (a market gauge of default risk) jumped to 82 basis points in late July when it looked like NVIDIA was on the hook for everything, then eased to about 70 once the risk was spread to outside investors, with NVIDIA's own backstop capped at roughly 25% of any deal, a ceiling of about $125 billion (Facts vs Feelings).

The same structure showed up in this week's Broadcom news. On Bloomberg Intelligence (Aug 21), reporters described Broadcom raising more than $60 billion in debt through a special-purpose vehicle that buys chips and leases them to Anthropic, letting Anthropic avoid carrying that cost on its own balance sheet ahead of the IPO (Bloomberg Intelligence). On Squawk on the Street (Aug 21), CNBC's Christina Partsinevelos put the scale in context: Apollo and Blackstone are arranging $70-100 billion in vendor financing for Broadcom chips destined for Anthropic and other AI labs, creating "an avalanche of investment-grade paper" that is competing directly with the US Treasury for capital (Squawk on the Street).

That competition for capital is why the government's bond-market moves matter for dealmakers. On Forward Guidance (Aug 20), the panel argued the Treasury and Fed are actively suppressing long-term yields (through buyback programs and yield-curve management) specifically to keep corporate borrowing cheap enough to fund the AI build via debt (Forward Guidance). And on Excess Returns (Aug 20), macro strategist Andy Constan flagged the real risk: with more than a trillion dollars of AI financing hitting the market, capital markets could temporarily "seize" if too many deals try to close before the spending is fully funded (Excess Returns).

Why it matters: every IPO, buyout, and bond deal now competes for money against the AI build. The Anthropic listing, the Broadcom debt, the Treasury's yield management, and the private-credit boom are not separate stories, they are the same pool of capital being fought over. When financiers like Blackstone and KKR are underwriting the AI chips instead of NVIDIA, and when financial-sector stocks (the XLF ETF) have risen a record 11 weeks in a row, it tells you where the money, and the fees, are flowing.

7. Data and Index Franchises: The Retail Door Opens to Private Markets

Beyond Goldman's ETF shopping spree (Section 2), the notable structural shift is private markets coming to ordinary investors. On Invest Like a Billionaire (Aug 18), the hosts dug into new funds from Wellington, Vanguard, and Blackstone offering retail access to private-market investments with $2,500 minimums, daily pricing, and limited quarterly liquidity. Their verdict was cautious: they like the access but worry that Blackstone's sheer size makes it harder to generate outperformance, and they flagged layered fees (1.39-2.28% for the blended fund) on top of underlying management fees and carry (Invest Like a Billionaire). Meanwhile, on Morningstar's Investing Insights (Aug 21), the message on the flood of new ETF launches was blunter: proceed with caution (Investing Insights).

The Bottom Line

This was a week where the capital-markets narrative pointed in one direction: toward AI, and toward the increasingly creative ways it's being financed. Anthropic's coming IPO is the marquee event, but the more durable stories are structural: prediction markets becoming Robinhood's biggest business, the big banks buying rather than building in ETFs, and a half-trillion-dollar private-credit machine being assembled to fund the chip build-out. Watch three things next: whether Anthropic's S-1 actually appears in the coming days and what its losses look like in daylight; whether the Lintris-style "priced-to-get-done" discount spreads to bigger US listings after Labor Day; and whether the wave of AI debt starts crowding out ordinary corporate borrowers. The window is open. The question is who gets through it before the capital runs thin.