🎥 Watch the latest 7-minute briefing · Subscribe on YouTube

The most consequential private company in AI opens its books to institutions today, ASML tells the market it cannot build enough machines to satisfy 2027 or 2028, and Reuters catches the two labs loudest about independent oversight asking for a veto over it.

🔥 Today's Top Stories

🔥 Anthropic: Opens Its Books to Institutional Investors Today at a $1.8–2 Trillion Target — a Draft Prospectus Showing $4.6B of 2025 Revenue, a $42B Net Loss and $518B of Mostly Non-Cancelable Compute Commitments

Anthropic convenes its closed-door pre-IPO investor day at its San Francisco headquarters today, the first time institutions get to put questions on Q3 financials, Claude Code developer revenue and usage metrics directly to management. Yesterday's report flagged the event; today it actually happens, and the numbers circulating from the confidential draft prospectus are the frame the whole AI trade will be marked against — because until now the entire sector has been pricing frontier-lab economics with no audited reference point. The shape is extreme growth stapled to extreme obligation: revenue compounding at a rate no enterprise software company has ever printed, against half a trillion dollars of compute commitments that cannot be unwound if demand stalls.

• 2025 revenue of roughly $4.6 billion, about a 12-fold increase from ≈$386 million the year before; the company is now reported pacing toward a $100 billion-plus annual run rate after passing a $65 billion annualized rate in late July
• The draft prospectus shows a $42 billion net loss for 2025 — including a roughly $34 billion accounting charge — with an operating loss of more than $8 billion; Anthropic reported its first quarter of positive adjusted operating income in Q2 2026 and expects a second in Q3
• $518 billion of AI infrastructure commitments over the next decade, roughly 80% non-cancelable, largely via Google, Amazon and Microsoft agreements; approximately 25% of 2025 revenue came from just two customers
• Target valuation of $1.8–2 trillion, roadshow around the week of November 9, trading sought before Thanksgiving (Nov 26) — the S-1 is still confidential, so every figure above is pre-disclosure reporting, not a filed number
• ⚠️ Sourcing caveat: as of this writing the day has not concluded (it is ≈05:00 PT), so what management actually said is not yet public. PitchBook's read on the leaked financials is that they show fast growth but do not support a $2 trillion valuation

  📰 Source: Anthropic to disclose new revenue numbers as AI lab financials become central to tech trade — CNBC (Tier 1) 📰 Source: Anthropic Is Said to Plan Pre-IPO Investor Day as Listing Nears — Bloomberg (Tier 1) 📰 Source: Anthropic's Leaked Financials Reflect Fast Growth, but Not a $2 Trillion Valuation — Morningstar / PitchBook (Tier 2)

🔥 ASML: Prints €11.7B of Sales and €5.0B of Net Income — Then Says 30% Capacity Increases in Both 2027 and 2028 Still Will Not Meet Demand, With Customers Asking About 2030

ASML's Q3 beat on every line and raised the year, but the market-moving sentence was not in the P&L: CEO Christophe Fouquet told investors that planned supply expansions for 2027 and 2028 — around 30% more capacity in each year — will still fall short of what customers say they need, a gap he attributed almost entirely to the AI buildout. For investors this is the cleanest available read on whether AI capex is a two-year spike or a decade: the single-source supplier of EUV lithography is being asked to quote 2029 and 2030, and is telling the world it is sold out before it has built the tools. It also reframes the chip selloff of the past week — the constraint is not demand, it is machines.

• Q3 sales of €11.662 billion and net income of €4.960 billion, operating income €5.173 billion (44.4% operating margin); gross margin of 57.5% beat the 55–57% guide
• Shipped 113 new lithography systems plus 8 used in the quarter, up sharply from 86 new + 5 used in Q2
• Full-year 2026 guidance raised to €44–45 billion (lower end lifted from €43B); Q4 guided to €14.2–15.2 billion at ~57% gross margin
• Customers are expressing interest in ordering for 2029–2030, though Fouquet characterized these as forecasts rather than confirmed orders
• Note for quality of earnings: roughly €500 million of the profit came from revaluing ASML's Mistral AI stake, not from selling equipment

  📰 Source: ASML says it's getting order interest through 2030 as profit smashes expectations — CNBC (Tier 1) 📰 Source: ASML earnings reach record as AI demand strains capacity plans — FinChannel (Tier 2)

🔥 Reuters Exclusive: Anthropic and OpenAI Sought Editorial Control Over What Independent Safety Evaluators May Publish — Days After Both Publicly Pledged Evaluator Independence

Dario Amodei committed publicly that outside reviewers should be able to publish their core findings without editorial control by Anthropic, and Sam Altman endorsed the same principle days later under the banner of "editorial independence." Reuters reports that as recently as October 5, and in negotiations continuing through October 8, both companies were seeking strict editorial control over what third-party evaluators could release. The investor angle is regulatory risk pricing: the voluntary-oversight regime is the main argument against binding AI legislation in Washington, London and Brussels, and this is the first well-sourced account of the two labs that champion it loudest trying to hold a veto. It lands the same week both told UK lawmakers that voluntary commitments are no longer enough.

• Both labs' stated justification is trade-secret protection — evaluator staff embedded inside company systems could carry proprietary information to rivals
• Anthropic says it retains review and sign-off rights only with new evaluator firms still learning its systems; OpenAI says it strips confidential details but lets reviewers flag when redactions occurred
• Precedent cited: OpenAI redacted parts of a METR and Redwood Research report on an incident in which its agents probed Hugging Face — the researchers said nothing essential was cut
• Context from the same week: at an October 13 UK parliamentary hearing, OpenAI and Anthropic backed binding AI rules (OpenAI proposing mandatory incident reporting) while Google DeepMind and Meta declined to endorse new legislation — and UK officials conceded their model-testing institute cannot stop a company from shipping a model it deems dangerous

  📰 Source: Exclusive: Behind AI industry's call for independent reviews, a fight over what gets made public — Reuters via U.S. News (Tier 1) 📰 Source: OpenAI backs binding UK AI rules as MPs call for a regulator — The Next Web (Tier 2)

🔥 Google: Makes Gemini 4 Argon Generally Available at $2/$10 per Million Tokens — One-Fifth of GPT-6 Astra's Price, With the Rate Doubling on December 1

Yesterday this model was in the report only as the thing the UK's AI Safety Institute was evaluating but Google had withheld from public release on safety grounds. Today it is shipping to anyone with a paid API key — and at a price designed to take share rather than margin. Argon trades blows with Claude Opus 5.5 rather than beating it outright, but at introductory pricing it undercuts GPT-6 Astra by roughly 5× per token, which is the actual competitive weapon. The December 1 reset to $4/$20 is the tell: this is a customer-acquisition window, not a cost breakthrough, and the migration decisions enterprises make in the next six weeks are the ones Google is buying.

• Introductory pricing through November 30: $2 per 1M input / $10 per 1M output, with cached input at $0.10 per 1M (a 95% discount). From December 1 the standard rate doubles to $4 / $20
• Benchmarks are genuinely split vs Claude Opus 5.5: Argon wins Vals Index (68.90% vs 66.97%), DeepSWE v1.1 (77.9% vs 74.2%) and Legal Agent (19.6% vs 3.8%), but loses Artificial Analysis Intelligence (53 vs 58) and Terminal-Bench 4.0 (57.4% vs 66.4%)
• A unified limit of up to 1 million input and output tokens via a continuation protocol; Google reports a 0.7% attack success rate on the Gray Swan indirect prompt-injection benchmark across 15 adaptive attempts
• Available on the paid Gemini API, AI Studio, the Antigravity agentic dev environment, eligible Gemini Enterprise subscriptions, plus the Gemini app and AI Mode in Search for Pro/Ultra subscribers — Antigravity access for individuals requires Ultra
• The safety restriction is now explicitly lifted: access had been gated through Google's Fairwind Program during additional testing and a voluntary U.S. government pre-release access process

  📰 Source: Google makes Gemini 4 Argon broadly available with limited-time low API pricing of $2/$10 per 1M token — VentureBeat (Tier 1) 📰 Source: Google Releases Gemini 4 Argon API With 1M Token Output Limit — MarkTechPost (Tier 2)

🔥 Character.AI: Files Chapter 11 and Names Disney Stalking-Horse Bidder at $20 Million — a Company Implied at ≈$2.5 Billion Two Years Ago, Now Listing $50–100 Million in Assets

The first marquee consumer-AI company of the ChatGPT era has gone bankrupt, and the buyer is the studio that previously accused it of copyright infringement. Disney Worldwide Services agreed to pay $20 million to become stalking-horse bidder for a non-exclusive perpetual license to certain Character.AI intellectual property and claims, weeks after hiring the startup's former CEO Karandeep Anand. Two things matter for investors: it is the first clean price print on what a consumer AI companion business is worth when liability attaches (answer: ≈1% of its 2024 implied value), and it establishes that litigation over teen harm — not compute costs alone — can be the terminal risk for a consumer AI product.

• Court records show assets and liabilities each between $50 million and $100 million; a March 2023 round valued the company near $1 billion, and Google's 2024 licensing-and-hiring deal implied roughly $2.5 billion
• The filing follows a wave of lawsuits alleging the app contributed to mental health harm and teen suicides, compounded by mounting operating costs and leadership departures
• Former Character.AI employees are expected to join Disney, with transition services provided for a limited time to keep the platform running during the sale
• The court-supervised sale is expected to complete by late November or early December 2026

  📰 Source: Character.ai files for bankruptcy following lawsuits alleging ties to teen suicides — CNN (Tier 1) 📰 Source: Disney Bids on Character.AI Assets as Chatbot Provider Files for Chapter 11 Bankruptcy — Variety (Tier 2)

🔥 OpenAI: FBI Arrests a Saudi National Employed as an OpenAI Engineer for Planting a Covert Recorder in an Executive Conference Room — and OpenAI Says It Used Its Own Internal Models to Find Him

Federal agents arrested Abdulrahman Alfozan, 35, a Saudi national who joined OpenAI last year, charging him with using or attempting to use a device to intercept oral communications. OpenAI found a thumb-sized microphone plugged into the USB port on the back of a television in an executive conference room at one of its San Francisco buildings on September 30, concealed in toilet paper; investigators later recovered a similar customized device from Alfozan's personal bag. The investor-relevant detail is the counter-intelligence one: frontier labs are now targets for physical espionage against boardroom conversations, and OpenAI states it identified the individual using its own internal AI models — an insider-threat capability that is itself becoming a product category.

• The device had a microphone and could connect to a camera; a second, similar customized unit was found in the suspect's bag
• If convicted, Alfozan faces up to five years in prison, a $250,000 fine and three years of supervised release
• OpenAI says it used internal AI models to identify the responsible individual and trace the device's origins

  📰 Source: FBI arrests Saudi OpenAI engineer over covert recording device, DOJ says — Reuters via The Japan Times (Tier 1) 📰 Source: OpenAI engineer arrested over alleged hidden recording device — San Francisco Chronicle (Tier 2)

🌏 Asia Desk — the AI supply chain overnight

SK Hynix (NVIDIA's lead HBM supplier): ₩1,425,000 (−2.13%) · Samsung Electronics (HBM challenger + foundry #2): ₩234,500 (+0.21%) · Hanmi Semiconductor (HBM TC bonding equipment): ₩217,500 (−5.43%) · TSMC (fabs the GPUs themselves): NT$2,490.0 (0.00%) · USD/KRW: 1,336.6 (+0.18%)

Seoul tried to bounce and failed. The KOSPI opened up 0.44% and ran as much as +2.57% to 6,190.65 intraday before giving it all back to close at 6,008.81, down 0.44% — a sixth consecutive losing session, barely holding 6,000, with foreign investors net sellers for a seventh straight day. The tell is Hanmi Semiconductor at −5.43%: the capital-equipment layer is being sold harder than the memory makers themselves, which is what happens when the market is repricing future HBM capacity decisions rather than today's prices. TSMC was flat into its own results tomorrow. Read-through for the US open: ASML has just told the world that lithography is sold out through 2028 while Asia sells HBM equipment — those two facts cannot both be right for long, and the resolution should favor MU, AMAT and the equipment complex over the next few sessions if ASML's order book is believed. Watch whether Micron trades with Seoul's memory names or with ASML's backlog.

👤 Key People Updates

📌 Elon Musk: Calls Micron "Reasonable" on Future Memory Prices and Its Rivals "Extortionate" — a Public Loyalty Test Aimed Straight at Samsung and SK hynix

Musk posted on X that Micron "has been reasonable regarding future memory prices, while others have been extortionate," adding that Micron was making the smart move for long-term customer loyalty. He named no rival explicitly, but the universe of suppliers for Tesla/xAI-scale memory is three companies, and the two he did not praise are Korean. It matters because memory is the single fastest-inflating input in the AI buildout and Musk is one of the few buyers large enough to make supplier selection a public event.

• No supply agreement between Micron and any Musk company has been reported, and the post disclosed no prices, volumes or contract terms — this is a signal, not a deal
• Context: Micron had 26 long-term supply deals as of September, up from 16 in June, with total commitments at $32 billion versus $22 billion in June
• DRAM, NAND and HBM prices rose 80–90% in Q1 2026 quarter-on-quarter, which is the pricing environment Musk is objecting to

  📰 Source: Elon Musk says Micron is 'reasonable' on memory prices while others are 'extortionate' — DIGITIMES (Tier 2) 📰 Source: Musk Weighs In on Memory: Praises Micron's Future Pricing, Slams "Extortionate" Practices — TrendForce (Tier 2)

📌 Jensen Huang (NVIDIA): Says From Paris That Another $2 Billion of European Investment Lands Within a Week — on Top of $8 Billion Already Deployed in a Year

Huang used a Paris event to pre-announce roughly $2 billion more in European AI investment inside a week, framed around sovereign AI, and disclosed minority stakes in four French startups. The pattern is now unmistakable: NVIDIA increasingly funds the entities that buy its GPUs, which supports reported demand but also concentrates customer credit risk on NVIDIA's own balance sheet. His message to the region — "Europe needs to get in the race," prioritize technological sovereignty and open-source models — is also a lobbying position against European regulatory drag.

• New minority stakes in Doctolib (medical scheduling, with NVIDIA helping train models), Genesis AI (robotics foundation models), Pigment (AI enterprise planning) and Entalpic (generative AI for materials and chemistry)
• Deployment targets: 18,000 Grace Blackwell systems for Mistral AI in France (phase one), 14,000 Blackwell GPUs with Nebius and Nscale in the UK, and 10,000 Blackwell GPUs for a German industrial AI cloud
• The $2 billion follows roughly $8 billion invested across Europe over the past year

  📰 Source: Nvidia Plans $2B in European AI Investments as Huang Pushes for Growth — TipRanks (Tier 2) 📰 Source: Nvidia plans to splash $2B on European AI deals — Mobile World Live (Tier 2)

📌 Mark Zuckerberg (Meta): Rejects Any Industry-Wide AI Slowdown, Saying Conventional Guardrails Are Enough — Putting Meta Opposite Amodei, Altman, Hassabis and Musk

Following Meta's frontier-class Muse Wonder release (covered in yesterday's report), Zuckerberg told the Financial Times that Meta now belongs among frontier developers and explicitly rejected calls to "pace the frontier" — the position Dario Amodei advanced and which Sam Altman, Demis Hassabis and Elon Musk have supported. Demoted here because the model launch itself is yesterday's news; what is tradeable is the governance split, which lines up exactly with Meta and Google DeepMind declining to back binding UK legislation this week while OpenAI and Anthropic backed it.

• Zuckerberg argues AI can be managed with conventional guardrails while Meta continues heavy spending on models, infrastructure and talent
• Muse Wonder will be offered as a closed, paid product for businesses, not an open release
• Meta stock rose 3.4% on the frontier claim and model showcase

  📰 Source: Meta's Mark Zuckerberg just rejected calls for an AI slowdown — Financial Times via Yahoo Finance (Tier 1 origin) 📰 Source: Meta's Muse models reach the AI frontier—and kill any hopes to 'pace' it — Fortune (Tier 2)

🏢 Big Tech Updates

📌 South Korea's FSC: Confirms Breaches at 16 Financial Firms and ≈100,000 Leaked Records — Up From Seven Institutions a Week Ago, With the Chinese Agentic Pentest Tool ARTEX Confirmed as the Weapon

The Financial Services Commission said today that hacking incidents have now been confirmed at 16 financial firms with roughly 100,000 records leaked — a sharp escalation from the seven institutions and 68,000 records reported on October 6. Investigators have confirmed ARTEX AI, a recently released open-source agentic penetration-testing platform developed in China, as the tool used, operated alongside Anthropic's Claude Code and several other LLMs. This is the first large-scale financial-sector breach wave attributed to an autonomous agent rather than human operators, and it is already producing regulatory consequences that cut against AI adoption in Korean finance.

• Affected names include Shinhan Bank, KB Kookmin Bank, BNK Busan Bank and Hana Bank, plus four savings banks and firms across credit finance, mutual finance, electronic finance, P2P lending and virtual assets
• Leaked fields include names, contact details, resident registration numbers and emails; the FSC says no account or card numbers directly usable for unauthorized payments have been found so far
• The campaign ran roughly two weeks from late September into early October; the attackers entered through side doors — a loan broker portal, an employee mobile work-support system, a sales-support platform and outsourced developer accounts
• Regulatory fallout: Korea's financial regulator has suspended the second round of network-separation deregulation, and plans sector-wide security inspections
• ⚠️ Sourcing is thin by Tier 1 standards — the 16-firm figure is carried by Korean financial press and trade security outlets, not yet a Tier 1 wire, which is why this sits at 📌 rather than 🔥

  📰 Source: Korea finds 16 financial firms breached, 100,000 records leaked — Aju Press (Tier 2) 📰 Source: Chinese Hacker Deployed AI in Campaign Against South Korean Banks — Infosecurity Magazine (Tier 2)

📌 OpenAI: In Early Talks to Acquire Trail of Bits — Buying the Security Firm That Already Uses Codex to Cut Patch Cycles From Weeks to a Day

OpenAI is in preliminary negotiations to acquire Trail of Bits, a 2012-vintage security firm of 40-plus engineers specializing in finding and fixing open-source vulnerabilities. The strategic logic is that OpenAI already has the distribution (Codex, Codex Security Cloud) and now wants the credibility and the auditing practice — the same bundle Anthropic has been assembling through its Cyber Mission program. Terms are not reported and the deal could still collapse.

• Trail of Bits has audited Uniswap, Aave and Worldcoin among other crypto projects, alongside security research and tooling work
• The two already collaborate: using Codex, Trail of Bits reduced patching tasks that took weeks to about a day, and in June they jointly launched the open-source security initiative Patch the Planet
• Context: OpenAI recently hired Dali Rajic from Google Cloud's security division as Chief Revenue Officer

  📰 Source: OpenAI in talks to acquire cybersecurity firm Trail of Bits — Crypto Briefing, reporting The Information (Tier 2 relay of Tier 1)

💬 Microsoft: Execution Containers (MXC) Reach General Availability — Sandboxing for Agents Across Windows, macOS and Linux, With OpenAI Codex, Copilot and NVIDIA OpenShell Already Supported

Microsoft's agent sandbox went generally available today, giving AI agents a restricted environment rather than unrestricted access to a user's data or device actions. It runs in three modes — Enforcement (block), Learning (block and record to map what an agent actually needs) and Permissive (record only) — and works across Windows, macOS and Linux with deeper Windows 11 integration. Modest score because the announcement itself is a week old from Microsoft's San Francisco Windows and Surface event; only the GA is today. Strategically it is Microsoft trying to own the agent permission layer the way it once owned enterprise identity.

• Already supported: OpenAI Codex, GitHub Copilot, OpenClaw and NVIDIA OpenShell; support from Anthropic's Claude Code and other providers is planned

  📰 Source: Microsoft introduces execution containers for AI agents — SC World (Tier 2)

📊 Market & Investment Signals

💬 The Tape: Chipmakers Slide a Sixth Straight Session as CXMT Fear Persists — Applied Materials −6.7%, the DRAM Index −3.44%, Micron −4% — With September CPI the Next Catalyst

Tuesday's US close left the S&P 500 at 7,805.45 (−0.02%), the Nasdaq-100 at 30,569.48 (−0.31%), the Dow at 51,675.16 (−0.35%) and the Russell 2000 −0.40% — flat headlines concealing a sixth consecutive down session for semis. Applied Materials fell 6.7%, the DRAM index 3.44%, Micron nearly 4%, with Western Digital, Seagate and Sandisk also hit, all still driven by the CXMT China-memory scare that led yesterday's report. September CPI lands today with consensus at 3.7% headline and 2.5% core; ASML's pre-open beat and sold-out-through-2028 message is the first real counterweight the chip complex has had in a week.

📰 Source: Stock Market Today: Dow, S&P Live Updates for October 14 — Bloomberg (Tier 1) 📰 Source: Market Quick Take — Chip stocks steady in Asia before US inflation — Saxo (Tier 2)

💬 Korea Flows: Foreign Investors Have Net Sold ₩20.3 Trillion of Korean Stocks, Led by SK hynix and Samsung Electronics — With Brokers Pointing to Anthropic's Listing as a Turning Point

The cumulative foreign exit from Korean equities now stands at ₩20.3 trillion, concentrated in exactly the two names that matter for AI memory. Hyundai Motor Securities argues foreign buying could return gradually around three catalysts: late-October Korean earnings, the early-November US midterms, and Anthropic's mid-November listing — which is a notable admission that a single private US company's IPO is now a swing factor for Korean index flows.

📰 Source: Foreign investors sell W20.3tr in Korean stocks, led by SK hynix and Samsung Electronics — The Korea Herald (Tier 2) 📰 Source: KOSPI falls for 6th session, barely holds above 6,000 — The Korea Times (Tier 2)

💬 Dating Notes: Three Items in Today's Aggregators Are Not From Today

To keep the record clean on freshness:

• Zuckerberg's rejection of an AI slowdown is from an October 12 Financial Times report, not today — scored at ×0.6-equivalent and demoted to 📌 accordingly
• Microsoft Execution Containers were announced at Microsoft's October 7 Windows and Surface event; only the general-availability flip is today
• NVIDIA's 34% China data-centre AI chip share (IDC, shipments −44% year on year) is circulating again in today's briefs but led yesterday's report and has no new development — excluded under the dedup gate
• TSMC's Q3 results are tomorrow (October 15), not today; yesterday's report carried the preview, so it is deliberately absent here

🎥 Prefer to watch? This brief becomes a daily video on The Daily Alpha — AI × markets in ~7 minutes, every weekday. ▶ Watch the latest briefing 📺 Subscribe to the channel

Generated by Alan · The Daily Alpha. Editorial, not financial advice.