The Information
14:56 PT
Exclusive: Anthropic IPO Waiting Game Puts Wall Street on Edge
So what
Anthropic is pre-announcing margin compression before the roadshow, so price the listing on compute commitments rather than the June quarter print.
Cory Weinberg's scoop: Anthropic's unit economics flipped positive just in time for IPO meetings, and management is already telling prospective investors the profit won't last.
- Operating spend went from roughly $2.30 per $1 of revenue in spring 2025 to a slight operating profit in the June quarter
- CFO Krishna Rao and other executives met groups of prospective public investors last month
- Executives told them Anthropic will use its growth and cash reserves to aggressively sign data center deals, raising compute spend and compressing operating margin
- Investors left with the understanding the profitability window is deliberate and temporary
McKinsey & Company
13:42 PT
Why transformations stall—and where only CEOs make the difference
So what
The 6 percent EBIT figure is the number to quote when a portfolio CEO defends an AI program on adoption metrics alone.
A podcast promo arguing that stalled transformations are a human coordination problem, not a strategy problem, with only the CEO positioned to break the deadlock.
- Kurt Strovink and Mathew Lee identify five collective-action problems: playing it safe, protecting turf, trusting insiders, waiting change out
- The claim is these behaviors are individually rational and collectively fatal
- Buried in the same email: McKinsey's State of AI finds only 6 percent of companies see significant EBIT impact from AI
- Ties to a September 22 McKinsey Live session on weighing agentic AI against the fully loaded cost of the work it replaces
Acquired
13:32 PT
Acquired in the WSJ: How Home Depot Became the Highest-Returning Stock of the Last 45 Years
So what
Worth ten minutes as a reminder that boring retail compounding beat almost every technology name you have ever underwritten.
Ben Gilbert and David Rosenthal turned their Home Depot episode into a Wall Street Journal column, with an ungated gift link for listeners.
- The claim: Home Depot is the highest-returning stock of the last 45 years
- Companion piece to the new Acquired episode; past WSJ columns collected at acquired.fm/wsj
- Pure distribution play — podcast research repackaged as mainstream financial press
The Information
12:30 PT
Inside Anthropic's IPO, Policy Fight and Rivals
So what
Three of Anthropic's largest enterprise logos are restricting Claude while it markets a $2T listing — that vendor risk belongs in your diligence, not the prospectus.
A promotional roundup that is actually the most complete picture anyone has of Anthropic's position: a possible $2 trillion listing built on $517 billion of compute commitments, with safety politics and customer defection running underneath.
- $517 billion in compute commitments signed in 11 months, plus a possible $10 billion Nvidia check at roughly $2 trillion valuation
- A $13.7 billion compute deal with a Trump-linked group; Anthropic may let existing shareholders sell in the IPO, breaking from the SpaceX playbook
- Anthropic split from Google and OpenAI over a state AI safety bill, even as the three quietly discussed a joint safety standards body
- Nvidia, Palantir and Booz Allen have restricted Claude use over data policy fears; Glean says Anthropic customers overpay by 80 percent; developers are running Claude Code on non-Anthropic models
- A researcher's extinction warning went viral mid-IPO-marketing, and price cuts are complicating the pitch
Every
11:24 PT
Vibe Check: Is Astra a Breakthrough for Indie Filmmakers?
So what
The VFX specialist line item is now optional for rigid-geometry shots; budget for one only when the effect touches skin, animals or physical interaction.
Afterimage's Cyrus Duff and Josh Lee ran GPT-6 Astra against real production footage and found a model that can drive Nuke and Blender competently enough to replace a VFX hire — for rigid objects.
- Astra beat Sol on masking (Robust Video Matting via self-written Python) and tracking a moving pan, producing usable masks where Sol did not
- It assembled the final composite inside Nuke via the Python scripting API; the filmmakers never opened Nuke themselves
- Successful shots: a 3D building on a Brooklyn street, a transparent wineglass with correct refraction, a Snoopy float outside a window using SAM 3.1 plus Runway Aleph 2.0
- Failures are consistent and structural: scene interaction (eyes on fingertips never merged into skin) and organic forms (cows in a drone shot)
- Earlier baselines: Opus 4.6 could write deterministic removal scripts but not complex changes; Gemini Omni was impressive but not controllable
McKinsey & Company
10:12 PT
Last chance to register: Improving the economics of agentic AI
So what
Adopt the fully-loaded-cost-of-work-replaced framing for your next agent budget review; per-seat and per-token comparisons hide the actual return.
Last-call reminder for the September 22 McKinsey Live on agentic AI economics, framed around the spend-value gap.
- Only 6 percent of companies see significant EBIT impact from AI, per McKinsey's State of AI
- Tanguy Catlin and Lari Hämäläinen on evaluating agents against the fully loaded cost of the work they replace
- Explicit framing: business leaders who delegated AI decisions to technology teams are now demanding returns
- Argues traditional IT budgeting was never built for this and governance is the new frontier
Ben Thompson (Stratechery)
10:01 PT
Doomforce (This Week in Stratechery)
So what
Salesforce just conceded that UI is not a moat — reprice any SaaS position whose defensibility rests on the interface layer.
Thompson's week was dominated by the AI doom debate, which he and Andrew Sharp both judge to have gone off the rails, plus a sharp read on Salesforce conceding UI as a moat.
- Thompson on Dario Amodei's pacing-the-frontier proposal: unrealistic, and mostly geared toward political control of AI
- Sharp's Some of All Fears argues the last ten days of AI anxiety discourse were absurd and irresponsible; solve real problems when they are real
- Salesforce's Dreamforce zag: Benioff climbs down from AgentForce, integrates Anthropic and OpenAI chatbots as the preferred UI, charges a premium — at the risk of eroding the SaaS beachhead for everyone
- ChatGPT ads are working and solve Amazon's biggest problem with chatbots; Walmart finally accepts Apple Pay
- Sharp China with Bill Bishop on why a US-China AI pacing deal is unlikely, and why the CCP's history of controlling technology feeds its confidence on AI risk
Every
09:03 PT
All Access Office Hours: A sneak peek at Every Agent
So what
Watch whether Every Agent's shared-in-Slack framing beats per-seat assistants — that's the live question for your own team deployments.
Membership pitch for a September 25 Office Hours previewing Every Agent ahead of public launch.
- Douglas Brundage and Brandon Gell with host Micah Rich; All Access upgrade required to attend
- Thesis: Slack agents are the mechanism for making a team AI-native, because they are used collectively rather than individually
- Most of the hour is Q&A; no coding experience or existing agent setup required
The Information
09:00 PT
Share your perspective: 2026 AI infrastructure outlook
So what
Note the Nebius sponsorship tag before you treat whatever gets published from this as neutral infrastructure demand data.
A five-minute pulse survey on AI infrastructure and cloud provider consideration, with aggregate-only reporting.
- Covers priorities, perceptions and what shapes vendor consideration
- Campaign tagging suggests a Nebius-sponsored lift study
- Results will surface later as published infrastructure sentiment data
AI with Allie
08:31 PT
How business professionals build AI agents in 4 weeks
So what
The addressable market for agent training is now bowling alley owners and Fidelity analysts in the same cohort — that's your distribution signal.
Allie Miller's sales email for Cohort 3 of the AI Agent Mastermind, positioned entirely around non-technical business professionals shipping working agents in four weeks.
- October 19 to November 12; 4 weeks, 8 live sessions, Slack opens October 13; $500 off until September 22
- Cohorts 1 and 2 both sold out; alumni included a bowling alley owner, a Cirque du Soleil employee, and non-technical builders from Netflix, Fidelity, Meta, Adobe, Charles Schwab and Novartis
- Guest session with Dominik Kundel from OpenAI; additional team seats at $997
- The pitch: stop asking AI for answers and build the system that runs at your goals unprompted