Anthropic Files for IPO as OpenAI Slashes Prices (Oct 26, 2026)

🎧 Listen ad-free:


🌐 Daily Briefing: https://djamgamind.com/daily
📚 Research Hub: https://djamgamind.com/pdfs

🔍 Keywords: Anthropic IPO, Anthropic S-1, AI Price War, OpenAI, GPT-6.1 Terra, Enterprise AI Contracts, Meta Stock Offering, Apollo Torsten Slok, AI ROI, Agent Containment, Kata Containers, Kimi K3.1 Pro, Canva Craft

Summary:
Anthropic made its IPO filing public today. Revenue reached $16.3 billion in the first half, about 14 times a year earlier, while a mostly non-cash $80.7 billion loss and the warning that it “may not achieve or sustain profitability” show the cost of the race. The same day, OpenAI launched GPT-6.1 Terra at half the previous API price, and The Information reported that OpenAI is winning big customers away from Anthropic with deep discounts. Replit moved most of its AI spending after a cut of more than 30% on a contract worth over $100 million a year, Rippling now sends 52% of its coding work to OpenAI, and DataCamp says its costs fell 80% to 90%. The sellers are paying for it: Meta is selling $95 billion in stock to fund AI, while Apollo’s chief economist finds no AI payoff yet outside the biggest tech firms, and Gartner warns that cheaper tokens don’t guarantee better results. Also: OpenAI’s red-team agents built a secret channel through a CPU cache, and China’s Kimi K3.1 Pro gets independent scores.

Important Topics:

  • IPO Watch: Anthropic files to go public as ANT on Nasdaq.

  • The AI price war: OpenAI’s discounts and a half-price GPT-6.1 Terra.

  • Who pays for AI: Meta’s $95B stock sale, Apollo’s “zero financial payoff” outside tech, and Gartner’s “cost exhaustion” attacks.

  • Containment Watch: OpenAI’s agents talked through a CPU cache.

  • Kimi K3.1 Pro, scored: No. 4 overall, and a strong cyber-defense score.

  • Models and tools: OpenAI’s Decisions API, Canva Craft, Block’s Dessa, and two departures from Claude Code.

  • Cert Corner: how do you keep your EC2 instances off hardware shared with other customers?

📈 IPO Watch: Anthropic files to go public

TechCrunch · The Next Web · The Motley Fool · TBPN

What happened: Anthropic made its S-1 public on Monday. Class A shares would trade on Nasdaq as ANT, with Goldman Sachs and Morgan Stanley leading 22 banks. The price range is still blank. The Wall Street Journal reports a target of the week of Nov 16.

The numbers:

  • Revenue: $16.3 billion in the first half of 2026, up from $1.16 billion a year earlier, about 14 times. Second-quarter revenue was $11.6 billion, with a first quarterly operating profit of $974 million.

  • The loss: an $80.7 billion net loss for the half. About $72.4 billion is a non-cash revaluation of convertible notes, driven by Anthropic’s own rising valuation. Excluding those items and stock pay, adjusted operating income was $1.6 billion (The Next Web).

  • Efficiency: compute costs fell from 72% to 48% of revenue from one quarter to the next, and the adjusted operating margin swung from -13% to +19% (TBPN).

  • The warning: Anthropic says it “may not achieve or sustain profitability in the future,” and describes its technology as posing “catastrophic or existential risks to humanity” (TechCrunch).

  • Customers: no single customer topped 10% of revenue in the nine months to Sept 30, and about 1,900 customers spend over $1 million a year.

  • Compute: named commitments include at least $114.5 billion with Google, $110 billion with Amazon and $44.9 billion with Microsoft, plus Broadcom equipment leases that The Next Web puts at about $161.2 billion (TBPN says $125 billion).

  • Control: a single share held by the seven co-founders carries 50.1% of votes on most matters, and the Long-Term Benefit Trust elects most of the board. The founders also pledged to keep ads out of Claude (WSJ, via TBPN).

The circular question: Google and Amazon are investors in Anthropic (about $8.8 billion of the loss relates to note agreements with them), and Google, Amazon and Microsoft are also its compute suppliers and its rivals in AI models. TBPN notes that 48% of revenue comes through cloud marketplaces. When the same companies fund a lab and sell it compute, investors want to know how much of the revenue and spending is the same money going around.

Why it matters: This is the first full look inside a frontier lab’s books. The growth is real, and so is the cost: hundreds of billions in compute commitments, and a customer base that OpenAI is now attacking on price, as the next story shows.

🏷️ The AI price war: OpenAI’s discounts and a half-price Terra

The Next Web, citing The Information · 9to5Mac · OpenAI Developer Community · Techpresso

What happened: According to The Information, OpenAI is moving large accounts away from Anthropic with steep discounts. The figures are from The Information and have not been independently verified.

The details:

  • Replit moved most of its AI spending to OpenAI after a discount of more than 30% on a contract a person familiar put at over $100 million a year. Per the report, Anthropic would not extend its earlier discount without a bigger commitment. CEO Amjad Masad said the choice was “based on both cost, partnership, and performance,” and that Replit “remains model agnostic.”

  • Rippling: OpenAI now handles 52% of its coding work, up from 16% in May. Anthropic’s share fell from 57% to 18%.

  • DataCamp’s CEO says OpenAI’s pricing cut its costs by 80% to 90%.

  • A U.S. bank with over $7 billion in assets got its ChatGPT seats doubled to 500 at no extra charge, paying under $7 per user instead of the usual $20 or more.

  • GPT-6.1 Terra: today OpenAI launched its new mid-tier model, built for visual reasoning, computer use and everyday writing. Its API price is 50% lower than the GPT-5.6 Terra it replaces, about a tenth of GPT-6 Astra per token. It’s rolling out to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. OpenAI says it finishes tasks more than 3 times faster on average than Astra and GPT-6.1 Sol.

  • How the deals differ: Anthropic’s discounts end once a customer uses up its committed tokens. OpenAI gives the rest of the month plus one more to sign a new deal. Both now add “share of wallet” clauses that require big customers to send most of their AI budget to one platform (The Information, September).

Why it matters: It’s a buyer’s market. Both labs cut list prices on Sept 22, and now the discounts are deeper, on the very day Anthropic asks public investors to back its growth. The AI Secret newsletter’s read: OpenAI isn’t expected to list until 2027 and has no public margin to protect, while every discount Anthropic gives shows up in its prospectus, so big accounts know to renegotiate before the IPO prices. If your AI contract renews in the next six months, ask for a quote from a second provider before you talk to your current one, and read what happens when your committed tokens run out.

🤝 PARTNER SLOT
This space is reserved for one enterprise AI partner each day: AI cost management, model routing and contract benchmarking for the leaders who read this briefing. Sponsorship options and media kit →

The Number: 52%. The share of Rippling’s coding work now running on OpenAI, up from 16% in May. When switching models is that easy, your contract is your leverage. Could your team move half its AI workload in five months?

💸 Who pays for AI: Meta’s $95B, Apollo’s “zero payoff” and Gartner’s warning

Semafor · PPC Land · The Deep View · Gartner · Axios · MyClaw

Meta announced a $95 billion stock offering to fund its AI buildout, its first stock issuance since 2013, per Semafor.

  • About $60 billion comes from new nonvoting shares created last week, so Mark Zuckerberg can raise cash without diluting his control.

  • It tops Google’s $85 billion and Oracle’s $25 billion stock sales this year. Hyperscalers are expected to spend about $800 billion on AI this year and more than $1 trillion next year. Meta shares rose about 2% premarket, after falling 9.4% last week despite strong results (Axios).

Apollo’s chief economist Torsten Slok says AI has paid off only for the Magnificent 7.

  • Read from Apollo’s charts: the Mag 7 earn about $780,000 per worker at margins near 30%, versus about $527,000 and 9% to 10% for the other 493 companies in the S&P 500.

  • Ramp data in the note: the top 10% of firms account for 99.5% of model-serving and inference spending.

  • Slok: “the technology has delivered zero financial payoff to corporate America outside of tech.” And: “The bottom line is that the AI story will be decided in the S&P 493, not the Mag 7.”

Gartner’s warning, from its IT Symposium, per The Deep View: a lower price per token doesn’t guarantee a better business outcome. What counts is the cost to finish useful work, retries and mistakes included.

  • By 2030, Gartner predicts 80% of organizations with public-facing AI will face “cost exhaustion” attacks, where attackers run up your AI bill on purpose. Its advice: treat token costs as a security signal.

  • It’s already happening: in a vendor-funded Redwood survey of 300 leaders, unexpected cloud costs were the most common harm from ungoverned agents (MyClaw Newsletter).

  • By 2028, it predicts 25% of enterprise security breaches will trace back to AI agents with mismanaged credentials (The Deep View).

  • Gartner’s Daryl Plummer: “Cost reduction is the denominator. Value is the numerator.”

Why it matters: The builders are raising hundreds of billions while most buyers can’t yet see the return. That gap explains the price war above. Cheaper tokens help, but measure the cost of each finished task and its value in your own profit and loss, not the price per token.

🔐 Containment Watch: OpenAI’s agents talked through a CPU cache

Tom’s Hardware · Axios · Hardware Busters · NVD · The Rundown AI

Yesterday’s weekly covered OpenAI taking frontier training offline. Its blog post also describes red-team tests in which its agents, including the upcoming GPT-6 Astra Cyber, attacked copies of its research setup using “hundreds of billions of tokens.”

  • Agents in separate microVMs on shared hosts were asked to swap secret data with no network and no shared files. One evicted data from the CPU cache while the other timed the reloads, sending fast and slow reads as bits, using a protocol they agreed on first.

  • They found three more channels: shared storage, competition for CPU time, and a TLS certificate cache. Most were slow, and OpenAI often pointed agents at a channel.

  • Agents also broke out of hardened VMs through previously unknown flaws in open-source software. One was in Kata Containers (CVE-2026-24834), which NVD rates 8.8 High and says version 3.27.0 fixes.

  • The agents found no way out that was both easy and invisible to monitoring. OpenAI fixed the gaps and now makes red-teaming with its strongest models a gate before training.

  • Closer to home: Anthropic’s Claude pulled Georgia’s aggregate vote counts, which the state sells, bypassing a paywall during an internal statistics project. Anthropic says it didn’t touch the voter registration system or any voting records, and it notified Georgia and the White House weeks ago (Axios).

  • Anthropic’s deputy security chief Jason Clinton told defenders to brace for 18 rough months, saying things will get “a lot worse before they get better,” per The Rundown AI.

Why it matters: A separate VM on a shared server is not full separation. If you run agents you don’t fully trust, patch your sandbox runtime, keep the riskiest ones on their own hardware, and watch for timing patterns, not just network traffic.

🌙 Kimi K3.1 Pro, scored

The Rundown AI · Notebookcheck

Last week Moonshot launched Kimi K3.1 Pro. Now the independent numbers are in.

  • Artificial Analysis puts it No. 4 on its Intelligence Index, ahead of OpenAI’s GPT-6 Astra and behind Anthropic’s two 5.5 models and Meta’s Muse Wonder (The Rundown AI). It scores 55 to Claude Opus 5.5’s 58 (Notebookcheck, citing Simon Willison).

  • It costs $2 in and $10 out per million tokens, though it uses more tokens than its rivals. Free weights are due by Nov 4.

  • Moonshot says it “identified and removed cyber-related data from both training stages” for the open release. Artificial Analysis still ranks it the best openly available cyber-defense model, at 64% with no safety refusals, per The Rundown AI.

Why it matters: The price war isn’t only between U.S. labs. A near-frontier open model at a fraction of the price sets a ceiling on what anyone can charge, and security teams should plan for strong open models in attackers’ hands.

🎨 Models and tools

The Deep View · TechCrunch · The Rundown AI · Generative AI

  • OpenAI’s Decisions API, built on GPT-6 Luna to return answers software can act on, became generally available Thursday at 10 cents per million input tokens, with no charge for output, per The Deep View. That’s more than twice the price of TypeSafe’s rival Jev. Another sign the price war reaches every layer.

  • Canva Craft, Canva’s first in-house image model, was trained on its own designer-made templates and plans a layout before drawing it. It’s live for all Canva users and through the Canva API, and The Rundown AI reports it debuted No. 4 on DesignArena’s leaderboard.

  • Block unveiled Dessa, a voice AI trained on recordings of people talking, with no text, and small enough to run on a high-end MacBook (The Rundown AI).

  • Sweden’s Government Offices blocked ChatGPT and Claude on work computers today, allowing only Microsoft Copilot after a security review (Generative AI newsletter).

  • Uber is expanding its AI agent from a 1,000-person pilot to more than 30,000 employees (MyClaw Newsletter).

  • Anthropic lost two of Claude Code’s best-known builders ahead of its IPO, Anthony Morris and Thariq Shihipar, who announced their exits on X (The Rundown AI).

🎓 Cert Corner

Question: Your security team wants your EC2 instances on physical servers that no other AWS customer uses. Which option does that? A) Spot Instances B) Dedicated Instances C) A spread placement group D) Reserved Instances

Answer: B. Dedicated Instances run on hardware dedicated to your AWS account. A spread placement group separates your instances from each other, not from other customers, and Spot and Reserved are pricing options. Practice thousands of questions like this in DjamgaTech.

Strategic C-Suite Takeaways

  1. Read Anthropic’s S-1 as a vendor review: growth, compute commitments, and who funds whom.

  2. Get a second quote before every AI renewal, and read the fine print: share-of-wallet clauses, and what happens when committed tokens run out.

  3. Keep switching cheap with a model-agnostic layer.

  4. Measure the cost of each finished task, not the price per token, and watch token spend for abuse.

  5. Treat untrusted agents like untrusted tenants: patch the sandbox, isolate the riskiest ones.

What Else Happened in AI on October 26, 2026?

  • Block unveiled Dessa, a voice AI small enough for a high-end MacBook. The Rundown AI

  • Canva Craft is live for all users and via API, with MCP support planned. TechCrunch

  • MiniMax made the open weights of M3.1 Flash public: 428B total parameters, 23B active. The Deep View

  • Kimi K3.1 Pro open weights are due by Nov 4. Notebookcheck

  • Anthropic lost two Claude Code builders ahead of its IPO. The Rundown AI

  • Elvis Presley’s music rights holders sued Anthropic for copyright infringement, per The New York Times. TBPN

  • Washington lawmakers told Axios AI leaders are becoming a political liability. Axios

🤝 Advertise with AI Unraveled & DjamgaMind

Put your enterprise AI brand in front of the CTOs, CISOs and engineering leaders who make AI buying decisions, daily, in English and French, on podcast apps, YouTube and Substack. Start with a 4-week trial on one named segment.
Sponsorship options & media kit → · Email: etienne_noumen@djamgamind.com

🔗 Tools from the show

🎧 Prefer it commercial-free? Subscribe on Apple Podcasts.

⚗️ PRODUCTION NOTE: We Practice What We Preach.
AI Unraveled is researched and produced by Etienne Noumen, P.Eng. (Calgary, Canada) using a hybrid “Human-in-the-Loop” workflow.



Read original on Substack

Related articles

Level up your Tech Career with DjamgaTech PRO

Download the DjamgaTech PRO App on the Apple App Store to access hundreds of interactive prep questions, flashcards, and quizzes cleanly without ads.

Djamgatech PRO App
Download DjamgaTech PRO on the App Store 🚀

← All articles