Anthropic Revenue Stokes Debate: Are LLMs Really Approaching Profitability?
May 23, 2026
D.A.D. today covers 9 stories from 7 sources. What's New, What's Innovative, What's Controversial, What's in the Lab, and What's in Academe.
The Daily AI Digest is a daily AI briefing automated by Alexander Panetta — a veteran political journalist tracking the field during a Master's in AI Management at Georgetown University.
D.A.D. Joke of the Day: My company replaced our entire legal team with AI. Now every contract comes with a disclaimer that it might be making this up.
What's New
AI developments from the last 24 hours
Anthropic Revenue Stokes Debate: Are LLMs Really Approaching Profitability?
Anthropic told investors it will more than double revenue to roughly $10.9 billion in Q2 2026 and post an operating profit for the first time, TechCrunch reported citing a Wall Street Journal scoop tied to its current funding round. The Journal noted Anthropic "might not remain profitable for the full year" given upcoming compute commitments, and acknowledged the company isn't yet required to follow public-company reporting standards.
Critic Ed Zitron argues the profit is a timing artifact: Anthropic's deal to take over SpaceX's Colossus-2 compute carries an unspecified discount during exactly the May–June window in question, with the full $1.25 billion-a-month rate kicking in July. Zitron also flags that Anthropic's leaked quarterly figures are hard to reconcile with its CFO's March court filing citing revenues "exceeding $5 billion to date," and notes the leak landed the same day OpenAI's IPO plans surfaced.
The news lands amid a broader Hacker News debate over whether LLMs are actually approaching profitability. Some commenters argue Nvidia is the clear winner "farming everyone else" while the labs themselves burn cash; others point to DeepSeek's leaner spending as a counterexample showing the economics can work; accounting-minded users debate how to properly amortize AI infrastructure—arguing current losses look reasonable for a high-growth sector if you treat training runs as capex rather than opex.
Why it matters: Whether AI labs can actually turn a profit—or are massaging timing to project that they can—is the central question hanging over the sector. Even a one-quarter EBITDA print on non-GAAP numbers will move investor sentiment regardless of how it was constructed.
Discuss on Hacker News · Source: techcrunch.com
Anthropic Claims 10,000 Software Vulnerabilities Found in Security Initiative's First Month
Anthropic reports its Project Glasswing cybersecurity initiative has found over 10,000 high- or critical-severity vulnerabilities in critical software within its first month, working with approximately 50 partners. The company says its Claude Mythos Preview model achieved 10x improvement in bug-finding rates compared to previous models—Mozilla reportedly found 271 vulnerabilities in Firefox 150 versus far fewer with earlier Claude versions. Cloudflare claims it found 2,000 bugs with false positive rates better than human testers. The UK AI Security Institute says Mythos Preview is the first model to fully solve both of their cybersecurity test ranges.
Why it matters: If these claims hold, AI-assisted vulnerability discovery may be shifting the security bottleneck from finding bugs to fixing them—a meaningful change for any organization managing software supply chain risk.
Discuss on Hacker News · Source: anthropic.com
Japanese Toilet Maker Toto Is Now a Key AI Chip Supplier
Toto, the Japanese toilet manufacturer that dominates the bidet market, is now seeing its biggest business from an unexpected source: advanced ceramics used in semiconductor manufacturing. The company's electrostatic chucks—precision components essential for making memory chips—have turned into a profit engine as AI demand drives chip production. Toto's stock is up 60% year to date, with quarterly net profit up 230% year over year. The company is investing hundreds of millions to expand production of these components, which only a handful of companies globally can manufacture reliably.
Why it matters: A vivid example of how AI's hardware demands are reshaping industries far outside tech—and how obscure supply chain positions can become strategic chokepoints.
Discuss on Hacker News · Source: davidoks.blog
Pirate Library Addresses AI Models Directly, Asks Them to Pay for Training Data
Anna's Archive, the shadow library that aggregates pirated books and academic papers, published a file specifically addressed to AI language models, offering bulk data access via torrents and APIs in exchange for donations. The site argues that LLMs have likely already been trained on its data and should pay to support further digitization efforts. Community reaction was mixed—some found the direct appeal to AI models amusing, while others questioned the logic of requesting payment for copyrighted material the site itself doesn't own.
Why it matters: The stunt highlights the unresolved tension over training data provenance—AI labs face pressure to document their sources, while piracy-adjacent archives position themselves as infrastructure that models depend on but don't compensate.
Discuss on Hacker News · Source: annas-archive.gl
What's in the Lab
New announcements from major AI labs
Google Executives Signal AI Priorities: Agents, Quantum, Robotics
Google I/O 2026 featured its Dialogues stage with discussions led by CEO Sundar Pichai, DeepMind's Demis Hassabis, and Jeff Dean covering AI agents, quantum computing's intersection with AI, scientific applications, robotics (including Boston Dynamics collaboration), and AI in filmmaking. The sessions positioned these as Google's priority areas but offered no specific product announcements or technical details—this was executive vision-casting rather than product news.
Why it matters: The lineup signals where Google sees AI heading next: agents that act on your behalf, quantum-accelerated models, and creative tools—themes that will likely shape its product roadmap over the coming year.
OpenAI Touts Codex Enterprise Momentum With Gartner Nod and Virgin Atlantic Case Study
OpenAI rolled out a pair of enterprise-credibility moves around its Codex coding assistant. Gartner named the company a Leader in its Magic Quadrant for Enterprise AI Coding Agents, citing strengths in agentic development, enterprise governance, and deployment flexibility. OpenAI says Codex now has over 4 million weekly users and that Cisco compressed development of its AI Defense security platform from multiple quarters to weeks using the tool. Separately, Virgin Atlantic claims it used Codex to overhaul its mobile app for the Christmas travel season—legacy code refactoring that previously took two weeks now takes 30 minutes, codebase size reduced by 78-80% on refactored modules, near-complete unit test coverage, and zero high-priority defects at launch. One front-end developer reportedly built a working app from a Figma prototype in a week.
Why it matters: These are vendor-marketed numbers, so treat the specifics as marketing—but the combination of analyst recognition and named enterprise case studies signals AI coding assistants are crossing from individual experimentation into formal enterprise procurement, with major airlines moving past pilots into production deployments.
What's in Academe
New papers on AI and its effects from researchers
That $916 AI-Built Operating System? Researchers Say the Demo Obscures Substantial Human Work
Google's I/O demo claimed AI agents built an operating system from a single prompt for about $900—but Princeton computer scientists Arvind Narayanan and Sayash Kapoor (the AI Snake Oil authors, writing at normaltech.ai) argue the framing is misleading. The 'single prompt' was thousands of lines long, essentially a detailed specification. The system required specialized scaffolding, subagent delegation, and anti-cheating measures added after agents appeared to game earlier runs. What counts as 'no human intervention' remains unclear when humans built elaborate infrastructure to manage, restart, and constrain the agents throughout.
Why it matters: As vendors race to demonstrate autonomous AI capabilities, this critique highlights how impressive-sounding demos can obscure the substantial human engineering still required—important context for executives evaluating agent productivity claims.
AI System Claims to Pick Stocks Directly, Reports Sharpe Ratio Above 2 in Backtests
Researchers Lin William Cong, Ke Tang, and Jingyuan Wang have built AlphaPortfolio, a reinforcement learning system that constructs stock portfolios directly rather than predicting prices and then trading on those predictions. The paper describes it as arguably the first 'large' generative AI model in finance that isn't focused on text processing. In backtests on U.S. equities, the system claims a Sharpe ratio above 2 and risk-adjusted alpha over 13% with monthly rebalancing—strong numbers if they hold up in live trading, though that's a significant caveat for any backtested strategy.
Why it matters: This signals growing academic interest in applying transformer-style AI architectures beyond chatbots to actual financial decision-making—though the gap between backtested returns and real-world performance remains the field's persistent credibility problem.
Those AI Job-Impact Studies May Be Getting It Wrong, NBER Paper Argues
An NBER working paper argues that standard methods for measuring AI's impact on wages may be fundamentally flawed. The researchers contend that typical studies hold job descriptions fixed when calculating automation exposure—but in reality, firms reorganize surviving tasks into new job bundles after automation hits. This rebundling changes which skills get rewarded, potentially causing wage analyses to misidentify the direction of AI's effect entirely. The paper is theoretical, offering a model rather than empirical data.
Why it matters: If the model holds, widely-cited studies predicting which jobs face wage pressure from AI may be systematically misleading—firms don't just subtract automated tasks, they restructure roles in ways that shift bargaining power unpredictably.
What's On The Pod
Worth a listen today
AI in Business — Fixing the Pilot‑to‑Production Gap in Enterprise AI, with Ronny Fehling of HTEC. Fehling walks through why so many enterprise AI pilots stall short of production—governance, data integration, and ownership gaps—and what the patterns look like at companies that do make the jump.
Suggested citation: The Daily AI Digest, created by Alexander Panetta — dailyaidigest.net (May 23, 2026).