Comfy Sunday Spaces v37: OAI/Anthropic Revenues, Neocloud volatility and Trading Philosophy

Summary

• Lower revenue expectations are testing an AI trade priced for exceptional growth. Flood contrasts OpenAI’s rumored $70 billion annualized run rate with its subsequently stated $50 billion, while Anthropic’s leaked roughly $16.8 billion Q3 revenue leaves questions around $100 billion ARR expectations. These remain unaudited claims, but the repricing already hit Nebius, CoreWeave and IREN, reportedly down 30–50% over two weeks: extraordinary technology still faces the “laws of financial gravity.”
• Kyle’s central macro claim is that AI’s demand for financing is setting the marginal price of money. The 10-year Treasury reached roughly 5.3%, yet yields subsequently fell alongside AI stocks despite hot CPI and PPI releases. His mechanism: less anticipated AI borrowing frees capital elsewhere; a genuine AI rebound could push yields back toward their highs. Developers who believe they are “building God” will keep borrowing until financing costs constrain them.
• An Anthropic IPO could succeed while draining capital from existing AI holdings. Kyle expects roughly $5 trillion of equity supply over the next year across major offerings and subsequent unlocks, competing for finite risk capital. Circle’s post-IPO surge illustrates his mechanism: investors seeking “the pure AI play” may sell infrastructure proxies to buy the labs themselves, regardless of whether those proxies’ businesses deteriorate.
• Flood’s defensive position is primarily cash, with a willingness to reverse when the thesis changes. He reports roughly 70% cash versus a usual 80% net-long posture without a strong thesis, retaining smaller Bitcoin and Hyperliquid positions. Flood says he has not seen Bitcoin outperform strongly through material equity drawdowns; Kyle adds that he has never seen it stay flat, much less rise, during one. Renewed equity conviction or inflationary support for AI could change Flood’s view. “I’m gonna wait for the next pitch.”
• Crash preparation means accessible capital and rehearsed decisions before prices dislocate. Kyle separates buying forced-liquidation rebounds from accumulating durable holdings, while Flood warns that resting bids 10–20% below market still liquidated leveraged traders when altcoins fell 60–70%. Flood’s illustrative shopping level is HYPE at $40, where he would begin buying incrementally. Their framing: “markets are pop quizzes.”
• Cheap semiconductor multiples may reflect peak earnings rather than overlooked value. Flood describes memory producers earning extraordinary margins while end-user AI economics remain expensive; the market may correctly anticipate that upstream profits will normalize. His valuation test is deliberately uncomfortable: “is every human on Earth gonna spend three thousand dollars a year on AI?” The broader discipline is distinguishing cash flows that justify ownership from stories that merely attract another buyer.
• Higher rates could eventually strengthen the debasement case rather than undermine it. Kyle follows Jeffrey Gundlach’s framework: sufficiently high sovereign debt can turn attractive Treasury yields into a warning about fiscal sustainability, potentially weakening the dollar and supporting gold, Bitcoin and property. AI distress could also invite inflationary intervention framed as strategic competition with China, which he considers “extremely constructive for Bitcoin and debasement trades.”
• Prediction-market growth does not settle either valuation or legal durability. Flood calls Polymarket’s $20 billion valuation excessive, including at $10 billion, and prefers ordinary exchange multiples; Kyle pushes back that regulatory uncertainty makes a confident valuation difficult. Kyle sees gambling businesses as relevant comparables. Flood argues crypto needs protections “codified into law” before a possible political reversal in 2028.

Deep dive

1. Private AI labs create public-market information risk

• Flood’s opening frustration is informational: enormous public-market exposure rests on two private companies whose revenue, spending and balance sheets outsiders cannot inspect. Even some investors lack information rights. Someone who learns that lab revenue is slowing could make informed trades in public beneficiaries before everyone else. His complaint is “a tremendous amount of asymmetric information,” rather than an accusation that either company falsified its numbers.

• Flood floats a forcing function for companies raising outside capital at this scale: tens of billions in revenue and hundreds of billions raised might justify public ownership and disclosure. He also suggests allowing annual reporting to reduce short-term pressure. Kyle’s disagreement is explicit: “completely unbothered.” A trillion-dollar furniture manufacturer, he argues, should retain the autonomy to remain private.

• Kyle’s concern instead follows the technology’s externalities: the labs monetize knowledge contributed across the public web while their products can affect everyone. He cites Anthropic’s proposed donation of up to 20% of its shares to Trump Accounts as an attempt to distribute economic upside without voting rights. Separately, Flood describes cryptography fears, reported hacking, and REA, a reverse-engineering tool he says can use AI to break down compiled programs. Those are Flood’s examples of externalities and potential harms.

2. Annualized run rates leave major questions about revenue and margins

• Flood stresses that the figures under discussion are rumors and unaudited disclosures. OpenAI’s whispered $70 billion ARR became a company-stated $50 billion. Annualization itself deserves scrutiny: $1 million earned on one day can become a $365 million annualized figure without establishing that the next year will produce that revenue. Markets initially shrugged off the revision before AI names weakened.

• Anthropic’s investor-meeting leak put Q3 revenue around $16.8 billion, or roughly $17 billion, against earlier discussion of $100 billion ARR. Flood’s possible reconciliation is approximately $25 billion revenue in Q4, supporting that annualized rate. That would require another roughly 50% sequential increase, similar to Q2–Q3, but the earlier period included “token maxing.” He also distinguishes rising costs for heavy API use from falling prices per token.

• Flood gives the bull case its strongest form: these might be “the best businesses on Earth,” with profitability obscured by pre-training. He cites inference as only 37% of data-center usage; shifting 90–100% toward inference could produce a massive gross-margin business. The question is whether those economics survive when the spending that maintains frontier performance stops.

• The bear case is commoditization: without continued pre-training, users might migrate to free open-source models or subsidized competitors. Flood’s analogy is economy air travel, where passengers largely want the same destination and acceptable delivery rather than a particular carrier. If the “end product, the end state” dominates purchasing decisions, frontier intelligence may struggle to retain premium economics for ordinary tasks.

3. Neoclouds sit inside an amplified spending chain

• Kyle separates outside customer demand from transactions within the AI complex. His estimate is roughly $250 billion in annualized spending by customers outside that complex, amplified into trillions as companies purchase from one another. Nvidia hardware enters CoreWeave facilities, networking comes from Broadcom, and servers consume AMD or Intel CPUs and memory. That “amplification factor” makes many securities sensitive to the same underlying demand.

• Kyle thinks some earlier growth reflected Chinese model distillation that the labs knowingly tolerated because it generated revenue and helped fundraising. In his explanation, blocking those customers subsequently flattens growth and unsettles downstream suppliers. This is his interpretation of the revenue trajectory; the discussion supplies no internal company evidence establishing it.

• Flood reports Nebius, CoreWeave and IREN falling roughly 30–50% over two weeks, even with the broader market still up around 15% year to date. Kyle points to widening credit-default-swap protection on CoreWeave and Oracle as investors identify potential casualties. His concern is what happens if the labs are “only like seventy percent as good as we thought they were.”

• Neither claims a trading edge in those names. Financing structures, construction delays, political obstruction and contractual responsibility require specialist knowledge they say they lack.

4. AI financing may now govern the cost of money

• Kyle recalls the 10-year Treasury reaching approximately 5.3%, alongside a rate hike, while equities initially resisted the pressure. Then hot CPI and PPI arrived, yet Treasury yields fell as AI sold off. His provocative reading is that inflation is not currently the dominant marginal variable; anticipated borrowing for AI is driving the market’s pricing of capital.

• The relevant distinction, in Kyle’s framework, is whether next year requires $500 billion or $1 trillion of AI borrowing. Strong demand makes financing expensive and competes with other borrowers. An AI slowdown reduces that competition, making capital more available to governments and businesses outside the sector. Falling yields can therefore accompany deteriorating expectations for AI rather than better inflation news.

• His conditional test is concrete: if the sell-off proves to have been a fund liquidation and AI stocks recover, Treasury yields should probably return toward their highs. If AI continues unwinding, financing should keep getting cheaper, potentially reaching the two-year Treasury and expectations for no further hikes. Neither outcome is presented as certain.

• Kyle believes developers who think they are “building God” will borrow every available dollar to accelerate, whether motivated by AGI or competition with China. Higher financing costs finally impose a ceiling. That is why he says fixed-income investors are currently “in the driver’s seat”: their pricing of default risk and funding determines what the rest of the trade can sustain.

5. An AI IPO can rally while starving its proxies

• Anthropic’s S-1 remains the major unresolved catalyst. Flood allows that it could reveal “the greatest company on earth” and squeeze bears. Kyle is more suspicious of delays: Polymarket’s probability of an IPO that year fell from roughly 94% in early September to around 70%, with only about 50% for completion by November’s end. He also cites limited information-sharing and renewed safety concerns.

• Flood discusses a rumored $1.5 trillion valuation versus an earlier $2 trillion ambition, with approximately $200 billion sought. Kyle combines potential Anthropic and OpenAI issuance, subsequent unlocks and SpaceX supply into roughly $5 trillion of equity looking for buyers over the next year. These are prospective supply estimates, not completed offerings, but they underpin his caution toward risk assets.

• Kyle’s best analogy is Circle: investors wanted a pure stablecoin business rather than partial exposure through Coinbase or PayPal, and the shares surged within three weeks. Portfolio allocation to a theme could overwhelm valuation. He expects similar demand for “the pure AI play,” potentially making Anthropic perform very well while investors sell correlated infrastructure holdings to fund purchases.

• Flood also watches existing companies issuing equity to finance capital expenditure. Management is choosing to sell ownership rather than borrow more; that counterparty decision deserves attention. He does not treat it as an automatic instruction to liquidate, but as another reason to reconsider sizing while the market works out AI’s eventual economics.

6. Flood’s cash position follows his equity thesis

• Flood reports approximately 70% cash, retaining Hyperliquid and Bitcoin in substantially smaller quantities. The context matters: without a strong thesis, he typically runs around 80% net long because of his underlying enthusiasm for crypto. He also says Bitcoin sold during the reduction delivered almost a $2,000 gain per coin. Cash is his current allocation decision: “I’m gonna wait for the next pitch.”

• Flood says that in nearly a decade and a half in crypto, he has not seen Bitcoin trade exceptionally strongly while equities sold off materially—10%, 15% or 20%. Kyle adds that he has never seen Bitcoin remain flat, much less rise, through such a decline. Its continuous trading and liquidity often make the drawdown larger. Betting on an exceptional decoupling is therefore insufficient reason for Flood to retain his usual exposure.

• A bullish change in equities would likely bring rapid crypto re-entry; highly inflationary preferential financing for AI could also change Flood’s view. Their defensive posture includes low-double-digit-percentage shorts, rather than a huge crash bet. Flood considers a 10% market decline the following week exceptionally unlikely, and Kyle emphasizes that their equity framework operates over approximately six months.

7. A thesis needs an invalidation and a matching time horizon

• Flood wants an identifiable price level, fact or state change that proves a thesis wrong. Otherwise, recognition can arrive after substantial losses. His own sensitivity to rising rates comes from complacency in 2022, when his portfolio was “eviscerated.” He also warns that social feeds reinforce positioning: liking one bearish CoreWeave post caused his feed to fill with similar arguments.

• Winning frequently is a different objective from making money. Flood cites fellow trader Basics TNZ reporting roughly a 30% win rate despite substantial profits; sizing the strongest opportunities can dominate the result. His requirement is emotional flexibility: losses become more damaging when a trader cannot admit error or change exposure.

• Kyle’s timeframe warning is that a six-month thesis can coexist with painful daily reversals, including 3–4% rallies and 10% up weeks. Flood’s example is Hyperliquid: buying in single digits and selling around $30 worked, but approaching $100 required enduring major tests, including Jelly Jelly and competition from Lighter. A short-term trading rule can interrupt a longer-term investment thesis.

• Flood and Kyle manage separate books and can disagree, so one visible wallet does not represent every position or view. Flood also rejects taking trades to defend an online reputation. “You get paid for being fluid and responding to state changes,” rather than letting social pressure choose the trade.

8. Crash preparation starts with capital that can actually move

• Kyle’s first rule is to avoid being incapacitated on the day. Capital behind a seven- or 30-day staking unlock cannot fund an immediate opportunity. Then distinguish two strategies: buying a forced-deleveraging overshoot, perhaps in a leveraged ETF such as SOXL, or accumulating assets intended for holding over one to three years. Those require different selection and exit decisions.

• His rehearsal scenario runs from a terrible Anthropic S-1 through missed spending commitments, distressed neoclouds, trouble at Oracle and Nvidia’s backstops creating contagion. The purpose is to pre-decide a response while calm. He explicitly says such a dramatic sequence “almost certainly won’t happen”; preparation is useful without making the scenario a forecast.

• Flood’s cautionary example is traders leaving bids 10–20% below market before altcoins fell 60–70%, liquidating even 2×-leveraged positions. Cross-margin accounts can also fail during a momentary 90% drop. Venue outages and congested chains matter independently of insolvency: capital on the wrong platform can leave a trader unable to act during the crucial hour.

• Flood’s illustrative plan is HYPE at $40: begin buying a specified dollar amount every minute and welcome lower prices, because he considers levels underneath attractive. A shopping list does not require unattended limit orders. “Markets are pop quizzes”: preparation can determine whether a rare dislocation produces double-digit gains or losses, rather than leaving the trader frozen at the screen.

9. Falling markets force stories back toward cash flows

• Kyle describes ownership fatigue after a stock falls in 18 of 20 weeks: “Why the fuck do I own this?” His framework is that markets oscillate between cash flows and stories. Prices can climb as high as a persuasive future allows, but durable bottoms require an ownership case based on being paid back within a reasonable period and with acceptable risk.

• Zoom supplies his historical example: the story that offices, classrooms and business communication would permanently migrate to one remote platform supported extraordinary expectations. “The stories work until they don’t.” In a prolonged decline, investors may prefer businesses whose valuations need less storytelling, even at the cost of sacrificing upside.

• Flood’s tweet cited companies moving from $50 million to $10 billion valuations within months. He says that implies $9.95 billion of value creation by a relatively small team. Kyle calls it effective storytelling. Flood adds that a VC’s claimed 1× downside protection depends on an acquisition actually occurring. Their preferred reading includes Buffett and Howard Marks for longer-term allocation judgment, rather than weekly market calls.

10. Debt stress could reverse the usual response to higher rates

• Asked whether the 1970s oil shock, rate increases and gold’s more-than-tenfold rise offer a useful parallel, Kyle says historical comparisons will likely fail here. He follows Jeffrey Gundlach’s framework: normally higher Treasury yields attract investors away from non-yielding gold, but sufficiently large government debt can make higher rates look like a sustainability warning instead of an attractive opportunity.

• If debt-service costs consume an alarming share of tax revenue, Kyle thinks the result is “probably dollar down” and probably supportive of gold, Bitcoin and even property. Separately, worsening AI distress could invite extraordinary support framed around beating China. He sees that potential intervention as constructive for debasement trades, which helps explain why he is less anxious about crypto’s longer-term position than about equities.

11. Prediction-market valuations depend on law as well as growth

• Flood considers Kalshi reasonable but is publicly bearish Polymarket at $20 billion and even $10 billion. His comparison is an exchange: he cites Nasdaq around 8× and suggests 5–10× as reasonable multiples, rather than 20×, 40× or 100×. The sector may grow, but implementation and competition leave him broadly neutral to somewhat bearish on current companies at their valuations.

• Kyle’s pushback is “I don’t know that it’s way overvalued,” because legal uncertainty makes the business itself difficult to handicap: the operators might not have a business in two years. Robinhood’s prediction revenue attracted growth enthusiasm, but he sees much activity substituting for exchanges or sports betting. MGM and DraftKings therefore look like relevant comparables, implying lower valuations than private markets currently assign.

• On the midterms, Flood distinguishes ordinary political gridlock, which markets often welcome, from crypto’s need for affirmative legal protection. Without legislation such as the CLARITY Act, a Democratic presidency in 2028 could reverse today’s tolerance; prediction markets are particularly exposed. Flood’s intuition is that Democrats probably win the midterms and will then be favored for 2028, rather than that either outcome is certain.

12. Cheap chip earnings and expensive AI services can coexist

• Asked whether institutions are simply chasing stories like retail traders, Flood says valuation depends on where one looks. He describes SK Hynix and Micron producing extraordinary earnings, citing roughly 90% net margins and enough profit to repurchase their shares at current prices in about three years. Earlier, Kyle noted some chip names trading at approximately 3–4× earnings. Those securities can look strikingly cheap.

• The catch is overearning: circular spending can make supplier profits unusually strong, while low multiples signal that the market doubts their durability. Flood says the market may be right to price chip companies cheaply. Conversely, assessing Google, Microsoft, Amazon or Meta solely on AI revenue against AI spending would make those activities look extremely expensive; their other businesses soften the consolidated picture.

• Flood does expect AI revenue to rise as people learn productive uses over time. His unresolved test is the scale of spending required to justify expectations: “is every human on Earth gonna spend three thousand dollars a year on AI?” Otherwise, as he puts it, “a lot of the math isn’t mathing.”