Nscale Investors Halve IPO Valuation.
The relentless narrative of AI's exponential growth is now meeting the hard economics of public market appetite. Today, that collision plays out in infrastructure valuations, forcing a recalibration of capital deployment and exposing the true cost of scaling frontier ambitions.
TL;DR
- Nscale's IPO Valuation: Nvidia-backed Nscale faces investor pushback, likely halving its target from ~$30B to $16-19B.
- Anthropic's Regulatory Pitch: A draft bill from Anthropic suggests a high threshold ($1B damage or 50 deaths) for government intervention on frontier models, drawing criticism.
- Replit Exits Anthropic: Replit's CEO confirms a switch from Anthropic to OpenAI, citing costs and performance, noting Replit was a top-five Anthropic client.
- Meta Recalibrates Scraping Terms: Following publisher resistance, Meta removed the "train" and general scraping clauses from its commercial terms for businesses.
- Market Sentiment Shifts: Gary Marcus warns of a 2008-style AI contagion, while Sam Altman acknowledges growing public "hate" for data centers.
Lead Story: Nscale Investors Halve IPO Valuation
This week's pivotal market narrative—robust demand juxtaposed with fragile financing—crystallized around Nscale. The London-based, Nvidia-backed data-center developer is encountering significant investor resistance to its proposed $30 billion IPO valuation, with bookrunners indicating a $16–19 billion range is necessary to secure commitments. This counter-offer is acutely telling: Nscale's most recent private funding round valued it near $15 billion, effectively positioning the public market closer to its established private worth rather than its desired growth premium.
Nscale filed in September for a New York Stock Exchange debut under the ticker "NSCL," aiming to raise approximately $3 billion. On paper, the pitch presents substantial scale, with roughly $103.4 billion in signed, active contracts as of late August. However, concentration risk and execution delays present clear challenges. Microsoft and Anthropic together account for about 85% of that contracted revenue, and Microsoft has since withdrawn as an anchor tenant at a planned Loughton, Essex site due to a grid connection slipping toward the 2030s. The flagship Norwegian facility is also running approximately six months behind schedule. Even Nvidia's pledged $1 billion for the pre-IPO round has, per the filings, yet to be disbursed.
The timing is critical. Australian rival Firmus just pulled its IPO on weak demand, CoreWeave shares shed approximately 25% in a week, and the broader AI-infrastructure complex is enduring its most challenging period since the summer. Nscale is the first significant test of whether public markets will continue funding the aggressive buildout at private-round prices – and the initial response is a clear retreat. It also lands days before Anthropic, Nscale's second-largest customer, is expected to file its own ~$2 trillion S-1. When the largest names in your order book are the very entities the market is nervous about, a discount is the polite version of a warning.
In Other News
Anthropic handed Congress a bill it wrote itself. Anthropic has privately pitched lawmakers a draft federal framework that would permit regulators to block a frontier model only when it poses a foreseeable, imminent risk of causing more than $1 billion in damage or killing or seriously injuring more than 50 people. This intervention would be limited to risks from bioweapons, models evading developer controls, or cyberattacks. Evaluators identifying such a risk would have 72 hours to report to Commerce, which could then petition a Washington judge to suspend the model. The bill also mandates testing, public disclosure, and independent review. First surfaced by POLITICO, it immediately drew criticism that the regulated entities drafted the rules, with the high 50-person injury threshold as a primary concern. With Congress still without a federal AI law and out until after the November midterms, the draft joins the House FRONTIER Act and a Thune-led catastrophic-risk bill in a crowded, stalled legislative field.
Replit publicly cut ties with Anthropic. At TechCrunch Disrupt, CEO Amjad Masad announced that Replit is "no longer working with Anthropic," citing cost, partnership, and performance as factors, and has made OpenAI its default model provider. He claimed Replit was one of Anthropic's top-five customers and linked the split to reports of an Anthropic revenue slowdown – while noting enterprise now comprises about 15% of Replit's revenue, up from 2% a year ago. The claim quickly grew contentious: a person familiar with the matter stated the partnership remained active, and a September Replit post had cited Claude running in production weeks prior. Whether entirely accurate or strategic posturing, it presents a difficult headline for Anthropic to manage ahead of an IPO.
Meta deleted the "train" clause. After publishers and legal experts cautioned that incoming commercial terms could permit Meta to scrape entire business-linked websites to train its models, the company removed the disputed language – first striking the word "train," then the broader provision. The terms were slated to take effect October 30 with no opt-out for business accounts. This represents a rare, swift retreat, signaling that the default assumption of "we can train on your content" is becoming increasingly difficult to assert transparently.
X / Social Pulse
- Gary Marcus continued his persistent critique of the AI investment climate, arguing that a stumble at one major AI firm could ripple into a 2008-style crisis. He drew comparisons between OpenAI and WeWork, vowing to short OpenAI on any listing, stating it lacks a "technical moat."
- Sam Altman acknowledged the growing public sentiment that Americans "hate" data centers, as OpenAI's physical buildout faces local resistance and moratorium proposals across dozens of states. He reiterated his agreement with Dario Amodei on the necessity for the field to pace itself.
- Amjad Masad effectively turned a vendor change into a public jab, framing Replit's switch to OpenAI as evidence that Anthropic is losing even its largest customers – a move with clear implications for market perception.
One to Watch
Anthropic's Impending S-1. Anthropic's public S-1 is expected around October 26, setting the stage for a Nov-9 roadshow and a listing near $2 trillion before Thanksgiving. Nscale's cold reception, Replit's defection, and Marcus's contagion talk now form the volatile backdrop. Monitor whether the prospectus explicitly names FTC and EU inquiries as risk factors – and whether a chilly IPO window ultimately forces the valuation down toward the ~$1.5 trillion some investors are already whispering.
Quick Hits
- CoreWeave's Stock Slid further amid roughly $1B in post-IPO insider sales – highlighting the financial stress underpinning the broader IPO freeze.
- Open Models Maintain Momentum as DeepSeek shipped V4.1-Flash-1017 (MIT license, 1M context) and Reflection released its Apache-2.0 Beam weights, illustrating continued progress into the downturn.
- Websites Block AI Agents with increasing frequency, even as OpenAI, Google, and Meta push web-browsing shopping agents, signaling a developing architectural conflict.
- ServiceNow Unveils AI Passport, which assigns each enterprise agent a signed, revocable runtime credential, with Nvidia as its launch partner, enhancing control and governance.
- Concord Music v. Anthropic Prepares for Hearing on October 21 before Judge Lee in San Jose, a key legal test for AI's copyright liabilities.
The underlying demand for AI's infrastructure remains undeniable, yet the capital flows financing its expansion are exhibiting increasing discernment. Nscale's IPO resistance marks a critical juncture, where the market has directly challenged the asking price. With Anthropic's imminent prospectus and key clients expressing doubt, the conversation is fundamentally shifting: from the raw pace of AI's buildout to the stringent terms under which its next phase will be funded.
Sources
- Nscale IPO: Bloomberg, The Next Web, Quartz
- Anthropic draft bill: MLQ.ai, Benzinga, Yahoo
- Replit/Anthropic: TechCrunch, Unite.AI
- Meta terms: Quartz, MediaNama
- Social/markets: The Stack, Futurism
- Quick Hits: LLM Gateway, ServiceNow, AI Toolly
Lock in. M. mazen@thorterminal.com