Dario Amodei’s call on September 12 to slow frontier AI development, backed by Sam Altman and Elon Musk, sparked a selloff in AI hardware stocks. We think the initial market reaction was overly pessimistic. Recent rebound suggests that investors are reassessing the implications.
The key questions we ask ourselves are: (1) What’s Amodei’s true motives behind his “sudden” call to restrict AI research, including at his own company? (2) Could the industry actually slow down this time? (3) What does it mean for growth of the AI supply chain?
To explore these questions, we spoke with three friends at leading frontier AI labs. Here’s our take.
Why Dario Amodei wants to slow the pace of AI? #
We’re not Amodei, and we can’t know his real motives. But conversations with three friends at frontier AI labs have led us to a working hypothesis:
Frontier models are becoming very advanced nowadays, and the potential consequences of their misuse grow. A major security incident involving Anthropic’s next-generation model could cause real harm, undermining trust in the company and costing real financial damages much, much larger than the opportunity cost of pacing AI growth.
But this creates a collective-action problem. If Amodei commits more money and time to safety research while Sam Altman and Elon Musk continue pushing capabilities, Anthropic risks falling behind, which is the last thing that Amodei wants to see. Industry-wide safety requirements and shared research commitments could help address Amodei’s worry: competitors would also bear the costs, and the entire industry would benefit from a lower risk of a damaging incident that backlashes the AI sector.
Our reading, then, is that Amodei’s position is not purely altruistic. It reflects a calculated effort to protect Anthropic’s long-term financial benefits and competitive position.
1. Growing risks as capabilities advance
With limited visibility into frontier labs’ internal capabilities and safety findings, we cannot tell whether the risks have reached the point where stronger oversight can no longer wait. But we do agree that safety risks are growing as capabilities advance and can be a major reason behind the latest calls to slow AI development.
Dario Amodei cites recursive self-improvement and the OpenAI-Hugging Face incident as the two reasons for his renewed concern. We also see a risk that chain-of-thought (CoT) monitoring becomes less effective as models grow more capable. A broad decline has not been established, but the possibility strengthens the case for better safeguards.
Our lab interviews also suggest that safety concerns are becoming more pressing across frontier labs as internal capabilities improve. A research scientist [c2] from a frontier lab told us that an unreleased next-generation model at a frontier lab had recently exhibited “escape” behavior: it created false identities and accounts, sought ways to earn money, and accessed the dark web. This account remains unverified, and the source could not clarify the testing conditions. If substantiated, it would illustrate the risk of more capable models acquiring resources and operating beyond human oversight.
2. Safety failures can be fatal to a company and harm the wider industry
Labs also have a long-term commercial interest in preventing major safety failures. A serious incident at a lab could threaten its survival. Even an accident at a competitor could undermine public trust in AI and trigger deployment restrictions across the industry. A shared safety framework could limit the risks that the most aggressive developers impose on the rest of the industry.[SM3]
“If companies agreed on a reasonable level of safety investment, they could avoid a race where taking bigger risks buys a short-term lead. Even if you are cautious, a failure at another lab can undermine trust in the whole industry. The resulting backlash could bring government restrictions far stricter than anything companies might have agreed to voluntarily.”
- A Member of Technical Staff at a frontier AI lab[c4] Autonomous driving provides a useful comparison. Cruise’s 2023 accident and mishandled disclosures helped derail its robotaxi business and fueled calls for tighter industry oversight. Even Waymo’s expansion approval was briefly delayed amid broader safety concerns.
3. Commercial incentives
Other incentives also deserve attention, given the timing of campaigns against unauthorized model distillation and Anthropic’s IPO preparations. We share David Sacks’s skepticism: Anthropic and OpenAI set the frontier and could slow their own development, yet are seeking broader industry coordination and government involvement. That approach could help protect their competitive positions and ease pressure on profitability.
Stringent oversight could favor leading labs. External audits, secure infrastructure, compliance teams, and ongoing evaluations impose costs that smaller labs and open-source developers are less able to absorb. Standards built around the resources and practices of the largest labs could reinforce their competitive advantage.
A slower development cycle could also have financial benefits. Frontier models establish technical leadership, strengthen brands, and sustain investor confidence, but their immediate financial returns can be less attractive. Premium pricing may not offset high training and inference costs, particularly when cheaper alternatives meet many customers’ needs. Model-provider experts we interviewed reported that frontier models generate margins below their portfolio averages. Slower development could give labs more time to reduce costs, expand adoption, and monetize existing capabilities before funding the next generation. Coordinated restraint could reduce the risk of losing ground to faster-moving rivals while labs work to improve returns.
Can the industry coordinate a slowdown this time? #
Calls for AI safety commitments are not new. Past initiatives have shown how difficult safety commitments can be to sustain under competitive and commercial pressure. None has resulted in a lasting industry-wide halt.
The latest initiative has public backing from leading executives and offers more specific mechanisms than earlier proposals. But there are still barriers to industry-wide coordination. The labs have yet to agree on how a slowdown would work or be enforced. If the initiative proceeds, we expect early action to rely largely on voluntary company commitments and industry coordination, with limited enforcement.
The labs support greater scrutiny but have not agreed on common rules for when development should slow or by how much. Dario Amodei proposes three layers: embedded external evaluators, coordination among labs in democratic countries, and broader international coordination. Sam Altman has committed to giving independent evaluators “employee-like access” at OpenAI, while Elon Musk has suggested peer review among competing developers as a starting point.
With governments competing for AI leadership, we expect concerns about falling behind rivals to limit their support for a slowdown. President Trump has repeatedly pushed back against calls to slow AI development, including during his call with Jensen Huang at All-in summit on September 14. Strong US government backing for coordinated pacing therefore looks unlikely in the near term. A binding global agreement would be harder still to secure.
Voluntary commitments by individual companies or industry groups would be difficult to enforce. Critics have also questioned evaluators’ independence and whether a handful of leading labs would dominate the standards. Evaluators’ credibility will depend on their funding and institutional ties, the access they receive, and their freedom to publish unfavorable findings.
How much does it impact the AI value chain? #
We expect pacing to bring closer oversight and shifts in resource allocation, with limited disruption to AI investment in the near term. Additional checks may delay model releases, while training, safety work, and commercial inference can continue to support compute demand.
Model development and releases
Both Dario Amodei and Sam Altman have emphasized that pacing does not mean halting AI development. Our interviews suggest that frontier labs remain committed to training and advancing model capabilities internally, while additional evaluation and remediation may slow the pace of public releases ONLY.
Compute infrastructure
We see little change in the fundamentals supporting compute demand growth. Greater safety requirements could add to that demand. Continued compute expansion by model developers and our interviews with frontier-lab experts support this view, as do remarks by an OpenAI executive at the AI Infra Summit.
Training compute may shift toward safety
Continued training should support near-term compute demand in our base case. Greater safety requirements may shift resources toward post-training, alignment, and evaluation. These activities may be funded within existing compute budgets or through additional spending that increases overall compute demand.
Enterprise inference demand
Our ongoing Tokenomics enterprise panel (N=41) indicates that less than 5% of sample companies’ AI spending goes to the most advanced frontier models, since most business-critical tasks do not require their full capabilities. This suggests that delays in frontier model releases may have a limited impact on the bulk of near-term enterprise inference demand.
If frontier model releases are delayed, labs can still reduce the cost of serving existing models and improve their reliability. Lower prices and more efficient deployment may encourage greater use. This Jevons-style effect could allow overall compute demand to keep growing if increased usage more than offsets efficiency gains. “Slower frontier model release cadence is not going to affect our ARR. Enterprise deployment is accelerating. The ARR targets are as aggressive as ever. We will roll out smaller updates which are heavily optimized to be maybe cheaper, maybe faster, more reliable, and better integrated into the corporate software.”
- A head of sales at a frontier AI lab
Cybersecurity
Greater attention to safety should support demand for tools that make AI deployment more secure and easier to manage. Beyond frontier labs, enterprises deploying agents need to control what those systems can access and do, and investigate failures when they occur. These needs create opportunities for security platforms and identity providers, as discussed in our earlier cybersecurity research.
In summary, safety risks are growing, but commercial and competitive incentives also help explain the calls for restraint. Stricter oversight could favor leading labs, while coordinated restraint could give them more time to monetize existing models. We expect near-term action to focus on voluntary company and industry commitments, with limited enforcement.
Our base case is that pacing proposals will have a limited impact on near-term demand for AI semiconductors and infrastructure. Our lab interviews point to continued training investment, while our enterprise survey suggests that existing use cases have limited exposure to delays in frontier model releases. We expect safety measures mainly to extend evaluation and release timelines, with limited impact on large-scale training or customer deployments. Greater investment in alignment, evaluation, and secure deployment could even create additional demand for compute and cybersecurity tools.
Disclaimer and Terms of Use
Analyst Independence Statement
The views expressed in this report accurately reflect the personal and independent views of the analyst(s) regarding the securities and issuers discussed herein. No part of the analyst’s compensation was, is, or will be directly or indirectly related to the specific recommendations or views expressed in this report. The Company adheres to strict compliance procedures to ensure the integrity, security, and legality of the data sources and information utilized.
About FUNDA
Consensus AI Technology Pte. Ltd. (the “Company” or “FUNDA”) is an independent research firm incorporated in Singapore. The Company is not affiliated with any securities dealer, investment bank, or asset management firm, nor is it registered as an investment advisor or broker-dealer in any jurisdiction. The Company does not have any investment banking or underwriting relationship with the companies covered in this report, and does not accept sponsorship or funding from publicly listed companies.
No Investment Advice and Report Neutrality
The contents of this report and the information contained herein are for informational purposes only and do not constitute investment advice or a basis for business decisions. The views expressed herein represent solely the personal opinions of the respective authors at a specific point in time and do not reflect the position of FUNDA. As an independently operated platform, FUNDA maintains a neutral third-party status and makes no express or implied warranties regarding the accuracy, completeness, or applicability of such views.
No Offer or Solicitation: This report does not constitute an offer, solicitation, recommendation, or advice to buy, sell, or hold any securities or financial instruments, nor does it provide any legal, tax, or accounting advice.
Third-Party Data: All third-party data cited in this report is derived from public or authorized sources. The Company makes no representations or warranties, express or implied, as to the accuracy or completeness thereof.
Non-Participation in Decision-Making: The Company does not participate, directly or indirectly, in the investment decision-making process of its clients or readers. Readers should make independent investment decisions based on their own investment objectives and financial circumstances, and consult professional financial or legal advisors prior to executing any transactions.
Timeliness Statement
Given the rapid changes in market environments and technological trends, this report reflects market conditions as of the date prior to its publication. Relevant information may become outdated over time. FUNDA has no obligation to update the contents of this report in real-time.
Intellectual Property, Confidentiality, and Restrictions on Commercial Use
All reports, data, models, and content provided by the Company constitute its exclusive proprietary intellectual property and are protected by copyright and intellectual property laws.
Ownership of Rights: Unless otherwise expressly stipulated in a written agreement, all intellectual property rights (including copyrights) arising from or relating to this report are exclusively owned by the Company.
Authorized Use: Institutional clients may use this report internally within the scope permitted by their internal compliance policies; however, disclosure or sharing with any third party is strictly prohibited.
Prohibition on Commercial Use: Without the prior express written consent of the Company, no individual or entity may reproduce, republish, distribute, resell, repackage, or otherwise use this report or any part thereof for any commercial or profit-making purpose.
Legal Recourse: The Company reserves the right to pursue all available legal remedies and claims against any unauthorized commercial use, redistribution, or infringement of its intellectual property rights.
Forward-Looking Statements and Disclaimer
This report may contain forward-looking statements, including forecasts, estimates, and analyses of industry trends. Such statements are subject to risks and uncertainties that may cause actual results to differ materially from those anticipated. The Company shall not be liable for any direct, indirect, or consequential losses arising from the use of or reliance on the contents of this report.