Anthropic Adjusts Profit, Billions Still Lost.
Today's market narrative for frontier AI is a study in calculated ambiguity, where the definition of "profitability" hinges precariously on what, precisely, is being excluded. The tension between immense developmental outlays and the pressure to present immediate financial viability is not merely an accounting exercise; it is reshaping investor perception and the foundational economics of this pivotal sector.
TL;DR
- Anthropic's Adjusted Profit Claim: At its pre-IPO investor day, Anthropic projected a second consecutive quarter of adjusted operating profit, a stark contrast to earlier reports of significant losses.
- The Caveat of "Adjusted": This profitability metric omits the substantial costs associated with training future AI models, revealing a projected 2025 net loss of ~$42B against ~$4.6B revenue, with an $8.06B operating loss.
- Ambitious Valuation Target: The company aims for a $1.8–2 trillion valuation before Thanksgiving, with reported gross margins exceeding 80%.
- Safety as a Strategic Lever: Anthropic positioned open models as carrying "more risk," while Nvidia's Jensen Huang indicated he would restrict chip supply to developers of unsafe AI, a stance that simultaneously reassures and hints at potential control.
- Persistent Hardware Investment: Semiconductor stocks saw mixed movement ahead of TSMC's earnings, yet ASML raised its 2026 guidance to €44–45B, and both Shield AI ($12.7B) and Legora ($550M) secured significant new funding, signaling continued capital flow into the infrastructure layer despite broader market skepticism.
Lead Story: Anthropic Adjusts Profit, Billions Still Lost
Anthropic convened its closed-door pre-IPO investor day today in San Francisco, offering institutional investors their first comprehensive look into the company's quarterly financials.
The key disclosure, designed to anchor investor confidence, was Anthropic's projection of a second consecutive quarter of positive adjusted operating income for Q3, alongside gross margins reported above 80%. This marks a notable pivot from the loss-heavy narratives that had circulated throughout the week leading up to the event. The firm reported crossing into adjusted operating profit in Q2, with an estimated $570M on $11.6B in revenue, and indicated a widening of this margin in Q3.
However, this metric's critical qualifier lies in its definition: "adjusted" purposefully omits the monumental expenditures associated with training future AI models—the very research, development, and compute power that underpin the company's core mission and future valuation. Incorporating these costs reveals a starker reality: a projected 2025 picture of ~$4.6B in revenue against a ~$42B net loss and an $8.06B operating loss.
This is the financial proposition presented for a potential $1.8–2 trillion valuation: an enterprise deemed profitable by a metric that discounts its most substantial operational cost, yet on a projected run rate exceeding $65B and potentially reaching $100–200B by 2028.
The "safety" narrative, a cornerstone of Anthropic's brand, served a dual strategic purpose. Concurrently, an Anthropic sales executive stated in Dublin that open-source alternatives pose "more risk"—reinforcing the "safe, controllable frontier" positioning crucial for both product differentiation and investor confidence. Conversely, the company's own research, which warned of models capable of self-replication across devices, simultaneously serves as a capability flex and a potent liability risk factor. The public prospectus is anticipated around October 25, with the roadshow commencing the week of November 9, targeting a Nasdaq listing before November 26.
In Other News
Semiconductor sector mixed ahead of TSMC earnings; ASML raises guidance. US semiconductor stocks showed divergence on October 14, with Nvidia climbing ~1% to $228.52 and AMD up 1.9%, while Micron dropped 3.6%, ASML lost 1.9%, and TSM slid ~2% intraday. Despite this, ASML, the lithography monopolist, upgraded its 2026 revenue guidance to €44–45B, an increase from its previous "at least €43B." This sets the stage for TSMC's Q3 report on Thursday, October 15, which will provide a critical real-time assessment of the AI infrastructure buildout that market skepticism continues to challenge.
Nvidia's Huang pledges to restrict chips for unsafe AI development. Jensen Huang, CEO of Nvidia, which commands approximately 90% of the AI training chip market, informed Bloomberg that he would cease supplying chips to any customer developing unsafe AI—qualifying it as "a hypothetical, but yes"—while simultaneously asserting that the current level of public alarm is disproportionate (Investing.com). This posture from the industry's primary enabler, who is also committing $250M for UK AI safety compute, mirrors the nuanced messaging Anthropic employs regarding its own safety narrative and commercial strategy.
Significant capital continues to flow into AI ventures. The sustained influx of capital is evident as Shield AI secured a $1.5B Series G round, achieving a $12.7B valuation, and legal-AI specialist Legora successfully raised a $550M Series D. These follow recent substantial rounds from Fireworks AI ($1.5B at a $17.5B valuation) and China's Manus ($500M+). Investment capital is conspicuously directed towards defense, core infrastructure, and model development, showing no signs of abatement despite broader market "bubble" discussions.
Open-weight models maintain performance leadership in key benchmarks. Even as Anthropic advances its argument for increased risk with open models, GLM-5.3 demonstrates superior performance in open-weight coding benchmarks, including DeepSWE and Terminal-Bench 2.1. Similarly, DeepSeek V4-Flash recorded impressive scores of 88.1 on GPQA Diamond and 79.0 on SWE-Bench. This persistent closing of the capability gap by open-source alternatives directly challenges the "safety risk" framing employed by closed-source proponents and demands a more granular strategic response.
X / Social Pulse
- Yann LeCun continues his pointed critique of Dario Amodei's more alarmist "doom framing," characterizing it as "completely deluded" and "the worst marketing campaign you can possibly imagine." Such direct commentary from a leading figure gains sharper traction in the context of Anthropic's investor day.
- Sam Altman's earlier assertion that "right now would be an ill-advised moment to go public" is being widely recirculated, implicitly serving as a subtle counter-narrative to Anthropic's imminent listing. This highlights the ongoing, often indirect, competition between the major AI players.
- The headline detailing Anthropic's "profitable, actually" claim is garnering attention from outlets like Futurism, with critics incisively noting that Anthropic "would be profitable if you ignore how much it costs" to actually develop AI. This precisely targets the nuance of "adjusted" profitability that the market is currently evaluating.
One to Watch
Anthropic's public prospectus. Due around October 25, this document will be the critical juncture where the "adjusted operating profit" narrative confronts the rigorous disclosure requirements of the SEC. The key will be to observe how the 80%-gross-margin, profitable-quarter framing stands against the GAAP net-loss figures, and whether the planned November 9 roadshow remains on track should TSMC's upcoming Q3 report inject fresh volatility into the semiconductor market.
Quick Hits
- TSMC's Q3 earnings: Scheduled for Thursday, October 15, following a preliminary report of a record NT$1.494T quarter.
- Google's research access: The company extended access to its advanced models—AlphaFold 3, AlphaGenome, WeatherNext—to 200 European researchers.
- Concord Music v. Anthropic: A cross-motion hearing is set for October 21, making copyright exposure a tangible IPO risk factor.
- Mistral Large 4 release: Open weights are still anticipated for October 31, sustaining momentum in the open-versus-closed model debate.
- xAI's NYC Council testimony: Elon Musk's xAI is slated to testify before the NYC Council on AI safety on Friday, October 16.
The underlying thread through today's intelligence was safety—manifesting as Anthropic's core IPO pitch, Huang's conditional chip supply threat, and the ongoing capability struggle between open and closed models. Yet, the very notion of profit underpinning much of this strategic positioning exists solely by excluding the monumental, defining costs of developing the next generation of AI. The market now awaits clarity: is this strategic adjustment a sustainable moat, or merely a financial footnote that the upcoming prospectus will be forced to unceremoniously expose?
Sources
- IPO & financials: FinanceFeeds, Bloomberg Law, Yahoo Finance, Yahoo (two exit doors), Bloomberg, Futurism
- Markets & chips: TradingKey, Yahoo Finance (TSMC), GuruFocus (ASML), Bloomberg (Huang), Investing.com (Huang)
- Funding: mean.ceo
- Models & safety: Thundercompute, TechXplore (Google/Europe), Fortune (LeCun)
Lock in. M. mazen@thorterminal.com