Anthropic S-1 Shocker, Greatest Business Ever? Google’s $500B Wipeout, Midterm Predictions
In a Nutshell
Anthropic's IPO filing reveals what could be the most capital-efficient business in history, with $16.3B revenue in H1 2026 (14x YoY), 85% gross margins, $100B cash reserves, and projections to reach $100B ARR by year-end—potentially surpassing the iPhone, Google search, and Amazon combined. The AI complex now represents 40-50% of capital market flows and 30-35% of Treasury issuance, creating systemic risk as these companies compete with governments for capital. Google lost $500B in market cap after missing search revenue targets, demonstrating the existential threat to legacy tech franchises while enterprise AI adoption remains at just 10-30% of white-collar workers, implying massive remaining runway.
These notes were generated by AI and may contain inaccuracies.
The All-In Podcast episode 295 features the return of Sachs, with the hosts discussing fake news reports claiming they had stopped using the term "super intelligence" in favor of AI. The hosts clarify Sachs was not present on the October 23rd episode and reaffirm their commitment to the term. The episode introduces David Freeberg, described as an organic farmer who recently visited Kenya testing true potato seed with local farmers, and Chamath Palihapitiya, referred to as an enterprise sales executive from the 8090 Corporation, who reports Q4 is off to a good start following a poor Q2 performance.
Anthropic filed their S-1 document for an IPO led by Goldman Sachs and Morgan Stanley, targeting November 16th before Thanksgiving. The company aims to raise $100 billion, which would be the largest IPO ever, surpassing SpaceX's $85 billion raise in June. Poly Market assigns a 78% probability that Anthropic raises at least $95 billion, with valuation estimates between $1.5-2 trillion and a 31% probability range of $1.75-2 trillion.
The revenue figures show $16.3 billion in the first half of 2026, representing 14x year-over-year growth. Q2 revenue is projected at $11.6 billion, a 30x increase over the previous five quarters. The company claims a total addressable market of $40 trillion for tokens and services, with projections of $100 billion in ARR by year end. The ownership structure features seven founders holding 50.1% voting power through a single class F share, with $112 billion in cash on the balance sheet.
Sachs addresses the widely reported $80 billion loss figure, explaining it as a phantom number resulting from convertible notes issued to infrastructure partners like Amazon. As Anthropic's valuation increased, these notes were recorded as losses on the balance sheet, though they represent no actual cash loss. The company has raised approximately $110 billion but spent only $9-10 billion, retaining roughly $100 billion in cash equivalents.
Gross margins have improved to approximately 85%, with 10x year-over-year improvement in cost-effectiveness for given tasks. The company can reduce token costs by 50% while capturing greater efficiency gains in margins. This efficiency stems from what Anthropic calls "compute multipliers" - algorithmic improvements, better data curation, and architectural enhancements that enable delivering the same output at one-tenth the cost.
Freeberg presents data showing Anthropic's customer growth: 200 customers spending $100K or more by end of 2024, 1,500 by end of 2025, and 11,900 customers by end of September. He estimates the addressable customer base at approximately 1 million businesses globally where $100K annual spend represents less than 1% of their total annual spend, suggesting Anthropic has achieved roughly 1% customer penetration.
Freeberg presents an AI value chain analysis showing profit margins across the ecosystem. Nvidia achieves 75% margins, turning every dollar of COGS into $4 of revenue. xAI generates 83% margins on chip spend, with $10 billion invested in Colossus yielding $1.25 billion monthly revenue. Anthropic demonstrates 85% incremental gross margins on each additional dollar of revenue.
The analysis shows a dollar spent by Nvidia ultimately generates $18 in revenue for Anthropic selling to customers. Freeberg notes that customers appear to be generating 2-10x returns on their AI spend, as evidenced by continued spending growth and retention rates.
Chamath presents Microsoft earnings data showing revenue growth from $198 billion to $332 billion over four years while adding only 2,000 employees. The company offered voluntary exit packages to up to 60,000 employees, with potential headcount remaining under 200,000. This demonstrates productivity gains enabling revenue growth without proportional headcount increases.
Chamath conducted analysis of Anthropic's S-1 risk factors, creating a word cloud showing the distribution of concerns. Compute represents the largest risk category, specifically mentioning data centers, power, chips, grid, electricity, and hardware. Regulation ranks second, followed by privacy/IP concerns, AI safety, monetization, competitors, and governance issues.
Within compute risks, data centers emerge as the primary concern, followed by financing commitments and power availability. The analysis indicates Anthropic rarely mentions customer traction or employee retention as risks.
The discussion addresses how Anthropic's data center and power requirements create political dynamics. Democrats could potentially slow data center and power development to create recessionary conditions heading into 2028, while Republicans would need to accelerate power and data center infrastructure to maintain economic momentum. The hosts note the tension between traditional Democratic positions on AI regulation versus potential influence from new Anthropic-linked wealthy donors.
Sachs argues the Democratic agenda on super intelligence includes either outright bans with 20-year jail sentences for developers or draconian regulatory frameworks featuring federal AI review boards. He suggests this regulatory approach serves to sabotage the Trump economy, which he claims is driven by the super intelligence buildout representing approximately 3% of GDP.
Anthropic has approximately 40 million individuals and small businesses spending under $100K annually. There are about 10,000 customers spending between $100K and $1 million. Approximately 1,650 customers spend between $1 million and $10 million. Only 250 customers spend $10 million or more annually.
The business model enables negative churn where existing customers can grow into larger spending tiers while simultaneously attracting new customers at high rates. This structure suggests potential for hundreds of billions in revenue within a 3-7 year timeframe, assuming 25-30% market penetration with reasonable negative churn assumptions.
When companies reach larger scales, budget allocation becomes distributed across director, manager, and VP levels, with millions of individuals holding million-dollar budgets. This enables extensive experimentation with multiple AI tools.
Transitioning from sub-million to multi-million dollar deals requires more rigorous ROI demonstration and total cost of ownership analysis. Cross-functional executive approval becomes necessary. Approximately 1,900 companies have successfully made the ROI case, though a significant portion are AI-native, venture-funded entities. The remaining enterprise customers likely number in the low hundreds of Global 2000 companies.
Anthropic's CFO Krishna Raalo has secured $125 billion in cash entering June, with total net burn of only $9-10 billion to reach this position. This represents exceptional capital efficiency relative to market cap.
The company needs to raise an additional $80-120 billion to reach $250 billion total capitalization. IPO pricing estimates range from $1.3 trillion on the low end to $1.7 trillion on the high end, with a midpoint around $1.5 trillion. Post-IPO, including an $80 billion debt facility, the company would have over $300 billion in available capital.
Alphabet maintains substantial cash reserves. Meta shows negative $1.2-2 billion in net cash position, with $80 billion in cash and short-term investments against $82 billion in debt.
The seven or eight major AI companies represent the core of the United States AI complex. These highly profitable, cash-generative businesses have initiated massive capital expenditure programs, shifting from cash accumulation to aggressive spending.
The AI complex collectively owes $340-400 billion to debt markets. Through 2027, these companies need to raise approximately 30% of total Treasury issuance for the same period, creating direct competition for capital.
Microsoft under Satya Nadella has halved its debt while peers have dramatically increased leverage.
Current token usage represents only 1-2% of corporate America adoption. When 100-200 million white-collar workers subscribe to AI tools with multiple agents per user, token usage could increase 500x.
Regulatory approval processes, utility staffing shortages, and capital equipment availability create bottlenecks preventing adequate compute buildout. The country cannot scale infrastructure fast enough despite willingness to invest.
AI companies are competing with government organizations for capital resources. Governments may respond by attempting to regulate compute capacity allocation, determining who receives access to what amounts of compute.
Anthropic has built significant government affairs operations, hiring former Biden administration AI officials and subsequently former Trump administration officials. The company created PACs, nonprofits, and other entities for political influence.
Anthropic leads in recursive self-improvement (RSI) efforts. As of August, 26% of tasks for creating next-generation models utilized RSI. The company has achieved algorithmic improvements that enable offering the same capability at one-tenth the price through compute multiplier effects.
The pacing letter published by Dario may function as preemptive alibi creation, acknowledging rapid RSI pursuit while claiming attempts to slow industry pace. A potential split between Anthropic and doomer communities may emerge, as maintaining leadership in superintelligence requires continued frontier advancement.
At $145-160 per share, the IPO appears fairly priced as an entry point into future growth. Two primary risks require assessment:
-
Revenue composition risk: The timeline for converting 10,000 sub-million customers to single-digit and tens of millions in spending depends on developing defensible use cases and white papers that enable multi-person executive sign-off.
-
Market risk: The AI complex's size creates systemic competition with US Treasuries. If Treasury auctions require higher rates to compete, broad-based borrowing costs increase across the economy. If AI companies must pay higher rates, this creates negative reflexive cycles within the sector.
The AI complex represents 40-50% of capital market equity and debt flows and 30-35% of Treasury issuance. This concentration creates a complex that may be considered too big to fail, introducing systemic rather than isolated risks.
Anthropic is projected to reach 100 billion, then 300 billion, and eventually 600 billion dollars in revenue. The company is already generating about 100 billion in revenue this year, representing roughly one-third of Google's revenue and one-quarter of Apple's. The cumulative revenue in the first seven years of America's best franchises—Amazon, Google search cost-per-click, Freaker, Alma, and the iPhone—will show that Anthropic's business could be bigger than the top three franchises in business history combined: the iPhone, Google cost-per-click search, and Amazon retail. This represents an unprecedented scale that has never been seen before.
This unprecedented scale means global investors have no choice but to buy American debt. For global participants seeking to invest capital, alternatives like France or China are not viable options.
The question was raised about whether open models would eat Anthropic's lunch, as previously predicted. The response is that open models are lowering prices, but the enterprise opportunity remains so great that Anthropic has not been dramatically impacted. While startups and enterprises are exploring open models anecdotally, these numbers show that open models are not slowing Anthropic down yet. Revenue traction and business quality are not being debated—it's acknowledged that revenue risk remains the biggest risk factor, but it's not materializing currently.
There are moments of systemic risk in markets where everyone goes risk-off. This hasn't been seen in the last 15-20 years, with the last close example being the Global Financial Crisis. A narrow part of the economy now represents essentially all capital market activity, requiring everyone to pay significantly more attention regardless of whether they are long or short the AI trade. This is a structural observation: approximately 10 companies collectively funding one thing will be competitive with global governments on borrowing, creating implications that have never been observed before.
Anthropic's quarter-over-quarter revenue is decelerating on a large base. While Q3 data isn't fully available, scuttlebutt on X suggests things slowed significantly in Q3, partly attributable to Anthropic voluntarily eliminating 99% of distillation to prevent copycat models from stealing their technology. This was a voluntary revenue hit to stop distillation. Additionally, Anthropic recently cut token costs considerably, and demand has since reaccelerated, resuming previous growth trajectory after Q3 hiccups.
Anthropic's ability to cut token costs is not due to giving up pricing power or losing margin—they are actually gaining margin through algorithmic improvements that make models more efficient. This applies across model tiers: Sonnet and Opus (tier 2 and tier 3 models) are getting cheaper. Enterprises are finding these lower-tier models sufficient for their needs, solving more problems without requiring the latest models. For specialized applications like science or mathematics, the frontier models may still be necessary.
If Anthropic can continue making second and third-tier models 10x cheaper each year, the Jevons Paradox becomes more pronounced as enterprises use AI more extensively. Token cost reduction drives increased usage across all levels, with demand growing as costs fall. New applications consume exponentially more tokens—agents use significantly more tokens than prompting, and cyber defense requires continuous monitoring of agents and attack surfaces rather than periodic pen tests, potentially requiring 50-100 times more tokens than previous use cases.
The biggest risk factor for Anthropic is open source models potentially commoditizing intelligence at a low price point, particularly for coding and other market segments. A second-order risk involves recursive self-improvement (RSI): to stay ahead and deliver algorithmic efficiencies, Anthropic must increasingly use their latest models for recursive self-improvement to discover compute multipliers and algorithmic improvements. This requires pushing the limits of what they themselves call risky activity. The question is whether this becomes unsustainable, potentially creating a breach with those concerned about RSI that could slow development and allow open models to catch up.
Nick Gallup's survey shows daily versus frequent AI users at work (defined as a few times per week) at approximately 10% and 20% respectively. Major enterprises are not allowing unconstrained AI usage or connections to email, knowledge bases, Notion, or Slack due to cybersecurity concerns. Once cyber threats are mitigated, daily usage could expand from 10% to 60-70-80% of employees. Uber recently purchased OpenAI ChatGPT access for every employee, suggesting this expansion may be the next development. When every employee has an agent, significant growth remains untapped.
Sax predicted this would be the greatest business ever on multiple programs. Two points were made: first, creating the best business ever and an AGI monopoly; second, pursuing regulatory capture, which should be prevented given the dominant position. The IPO will clarify this dynamic—as Anthropic appears as a big tech monopoly or part of a duopoly, arguments for regulatory measures to increase competition, break up Anthropic, or separate infrastructure from tokens and subscriptions will gain traction. Dario reportedly claimed they were building the last business ever, with a lengthy response that didn't deny the statement.
Google was down 10-15% on Thursday, wiping out approximately 500 billion in market cap. For two consecutive quarters, Google missed on search revenue despite massive search growth—the issue is that people clicking on ads is not growing proportionally. Google is experiencing negative free cash flow despite cloud revenue doubling year-over-year to 30.7 billion. Overall revenue growth is 25% year-over-year, but the core cost-per-click advertising network faces disruption as ChatGPT, Anthropic, and similar services provide direct answers and potentially solve problems through agents. Search revenue grew 12% year-over-year, down from 17%, showing deceleration but continued growth.
Google trades at roughly 21 times operating income based on reliability and growth of future cash flows. Google Cloud generated 3.5 billion in operating income last year for the quarter, now exceeding 11 billion—similar to SpaceX's AI infrastructure value proposition. The question is whether Google has competitive models to run on this infrastructure. Core Google services grew from 33.5 billion to 39 billion, potentially driven by changes in ad pricing algorithms that may be over-earning on clicks, which could cause advertisers to reduce spend over time. The key question is whether consumer AI business can replace declining search revenue and how quickly that transition occurs.
Internally, there is turmoil about Google's identity across business segments—whether Google Search is AI or search, whether Google Cloud is models or compute rental. Jeff Dean and others recently left Google, potentially because computer scientists building models aren't receiving hoped-for resources, as Google can make more money renting compute to others. Alphabet was restructured with the creation of other bets to diversify from search dependency. YouTube, Android, and Google Cloud are all surging while the search franchise slows.
Max Anderson's viral post (5 million views) detailed changes in Google's auction model, claiming Google abandoned the Vickrey auction where the winner pays the second-highest bid price. The allegation is that Google is now over-earning on clicks through pricing algorithm changes, and if advertisers detect this, they may reduce spend. These concerns have been circulating since July without clear response from Google IR. If this narrative gains momentum, suggesting the franchise is pulling money from the future to smooth current earnings, market reaction will be negative.
Google Search revenue has fallen below 50% of total Alphabet revenue for the first time, though absolute search revenue increased from 56 billion to 63 billion year-over-year—a 7 billion increase. Google Cloud doubled to 30.7 billion, driving the relative share decline. Other growing segments include YouTube and G Suite subscriptions. The diversification away from search dependency is viewed positively, though the transition requires moving from investors seeking single-stock search bets to those comfortable with a conglomerate structure. Alphabet owns 6% of SpaceX, 15% of Anthropic, and 85% of Waymo.
Current polling shows confusion with significant misinformation. The prediction is for a Democratic "spritz" rather than Republican gains. Early data suggests Republicans may keep the Senate while Democrats secure a narrow House majority. A Democratic mandate would involve creating FUD through subpoenas, disclosure, and document releases heading into 2028, plus benefiting from any economic slowdown. Attention should focus on governorships and state-level races rather than federal outcomes, except for FUD creation.
Multiple polling firms have admitted to releasing fake polls that were included in polling averages. Some polls are outright fraudulent, while others are well-known for methodological issues.
Three-quarters of polls across the last three elections showed bias favoring Democrats according to Patrick Rafini's study. The 2024 election cycle featured Ann Seltzer, considered the gold standard pollster, predicting Trump would lose Iowa by approximately 10 points when the actual result proved entirely different. Pollsters demonstrate recurring patterns of bias through oversampling progressives, whether intentional or otherwise. This bias has been documented across multiple election cycles rather than representing isolated incidents.
Current economic indicators show GDP growth at 2.5% for the last quarter, tracking toward 3.7% presently. Median household income has reached an all-time high of nearly $90,000. Poverty rates stand at approximately 12%, representing the lowest level ever recorded. Real wage growth is occurring in real terms for average households, contrasting with the Biden administration period where this did not occur. The recovery extends beyond K-shaped patterns to include broad-based improvements across income levels.
Poly Market data indicates Republican midterm prospects, though the platform was incorrect about Bolsonaro's election and the Net and Yahoo outcomes. These prediction markets reflect public sentiment among participants willing to place bets rather than guaranteed outcomes. The platform serves as one data point among others rather than definitive forecasting.
Republicans face structural challenges winning midterm elections when holding the White House. Bill Clinton lost the 1994 midterms, resulting in Republicans gaining House control for the first time in 40 years through the Contract with America. Barack Obama experienced what he termed a "shellacking" during his midterm elections. The out-of-power party typically performs well in midterm contests as a standard political pattern.
Current media portrayals describe the midterm as a potential tsunami or bloodbath for Republicans. The actual outcome may align more closely with typical midterm patterns rather than representing an extreme result. Media operations appear designed to create voter depression and suppress turnout among Republican voters.
Poly Market currently shows Democrats with a 68% chance of winning the Senate and 92% probability of taking the House. Trump's net approval rating has reached an all-time low. Democratic voters are expected to turn out in force motivated by anti-Trump sentiment. Moderate voters may not participate at previous levels, having lost the "Trump curious" motivation from the previous election cycle.
The America First faction within the Republican Party feels the Trump administration failed to deliver on key promises. Specific grievances include unresolved inflation, unaddressed wars, and initiation of additional conflicts. This internal division creates conditions favoring Democratic midterm success according to the analysis.
If Democrats gain congressional control, impeachment proceedings against Trump are expected to resume through the standard House and Senate process. This would create a chaotic two-year period with extensive legislation designed to force Republican opposition votes. The strategy aims to damage JD Vance's political prospects through forced voting records.
A Democratic midterm victory would likely prompt the Republican Party to move beyond the Trump era and associated figures including JD Vance and Marco Rubio. New candidates would emerge to champion a more consistently populist agenda that addresses grievances about unfulfilled campaign promises. The analysis suggests this shift would occur regardless of personal political preferences.
Democratic control would likely involve extensive subpoena campaigns targeting any perceived instances of grift, whether substantiated or not. This approach would encompass both valid concerns and politically motivated investigations across a broad spectrum of targets.
The All-In Podcast will host a midterm live stream from 8-10 PM on YouTube, X, and Poly Market featuring guests Shane Copeland, Tim Dylan, Rahm Emanuel, and Taylor Lorenz. The event is expected to be chaotic given the anticipated election outcomes.
Taylor Lorenz is described as being on a redemption arc with improved performance on certain topics recently. Her participation in the upcoming midterm coverage represents a notable development for the podcast.
The Democratic Party has not fundamentally changed following the 2024 election loss. The party has not conducted meaningful analysis of electoral defeat factors. Rather than moderating positions, the party has become more extreme with the DSA faction gaining influence, as acknowledged by Gavin Newsom. This continuity suggests Democrats would revert to previous patterns including lawfare and woke policies if returned to power.
American politics operates on a pendulum principle where extreme positions generate counter-reactions. If Democrats regain congressional control, their unchanged approach would likely remind voters of previous governance patterns, potentially affecting future electoral outcomes.
Both Republican and Democratic parties will likely see new champions emerge who cannot currently be identified. The post-election period may create opportunities for previously unknown figures to rise within both parties based on demonstrated ability to address voter concerns more effectively than current leadership.
Keep All-In Podcast in your library
Save the videos and channels worth coming back to, and find them again in one place.
