Investor concern over whether AI spending can generate adequate returns is weighing on technology and infrastructure stocks. Michael Burry said on November 19 that he had warned about risks involving Nvidia, CoreWeave and Nebius Group, but had not urged investors to short them. By that date, the three stocks had fallen 39%, 76% and 66% from recent highs, and the VanEck Semiconductor ETF was down 33%. A separate market selloff erased about $3 trillion from the S&P 500’s value in early November, while the Nasdaq had its biggest one-day drop in 19 months. The declines raise questions about returns, but do not by themselves show that companies are weakening.
Daniel Loeb of Third Point estimates that hyperscalers and leading AI developers may invest $5 trillion to $7 trillion from 2025 through 2030. To justify that outlay, he says the industry would need $2 trillion to $3 trillion in new annual AI revenue—33% to 50% of today’s roughly $6 trillion global IT market. He cites higher borrowing costs for data-center projects, while saying Third Point sees strong AI demand and does not forecast an industry collapse. Microsoft, Amazon, Google and Meta are among the companies committing heavily to AI infrastructure; Nvidia supplies processors used to train and run models. The articles distinguish demand for equipment from proof that the broader industry can earn lasting returns.
AI infrastructure companies face their own valuation and financing tests. Hut 8 ranked eighth among 10 AI stocks short sellers viewed as overvalued, with a forward price-to-earnings ratio of 137.9 and price-to-sales ratio of 21.59. Its third-quarter 2026 revenue fell 14.9% to $71.1 million, while adjusted EBITDA rose to $33 million from $30.5 million a year earlier. It reported $26.6 billion in expected aggregate base-term contract value for its AI data-center portfolio, but investors are watching for recurring revenue and sustainable free cash flow. Cipher Digital, shifting from Bitcoin mining to AI and high-performance computing, lost more than half its market value in 2026; short interest reached 18%, and net debt was about $3.08 billion as of September. Its shares traded at about 18.24 times sales, with construction delays also a concern.
Infrastructure economics are changing as inference grows: it now accounts for two-thirds of AI workloads, shifting some spending toward smaller sites that can be built quickly. Power, cooling and memory availability influence where companies generate AI tokens. Positron AI uses commodity memory rather than costly, constrained high-bandwidth memory. The company says a server uses about 4,500 watts and a full rack of Titan servers fits within 40 kilowatts; it says modular sites could move from securing land and power to earning token revenue in 90 to 120 days. Separately, an EPR Research audit found that public information let researchers estimate costs at three monthly interaction volumes for just 13 of 25 AI customer-service products. Twenty products disclosed a clear charge trigger, but seven lacked sufficient detail for full estimates. The study assessed pricing transparency, not product quality, and the products’ billing units made direct cost comparisons inappropriate.
Google is adding Create images, Ask about files and Brainstorm buttons to its desktop homepage for signed-in users. It says AI searches, uploaded files and image requests may be used to train its models; users can turn this off in Gemini Apps Activity settings. Google’s integration of AI Overviews into Search is also prompting school-district pushback. The company plans to disable AI tools for K–12 student accounts unless administrators enable them, starting early next year, but says Search and AI are deeply integrated. Chicago Public Schools wants Search without AI and may seek another service; New York City and Los Angeles are weighing the policy, while Miami-Dade’s position is unclear.
Congressional scrutiny is also rising. A bipartisan Senate proposal would require the SEC to issue AI disclosure rules within 180 days for public companies involved in AI. Proposed disclosures include whether a company has an AI governance policy or board committee, details of certain AI-related financing and business arrangements, and a standard AI-risk section in Form 10-K covering reliance on outside models and computing providers, training costs, data rights and foreign AI models.
Anthropic is at the center of two contrasting stories about confidence and accountability. One Washington Post brief argues that the company has struggled to meet promises and relied too heavily on existing technology rather than new ideas, contributing to doubts across the AI industry. Another report says Anthropic raised a record $20 billion in an IPO, intensifying questions about whether its safety mission can withstand commercial and shareholder pressure. CEO Dario Amodei says the company will take the time needed to ensure its systems are safe and remain under human control. As a public benefit corporation, its board must weigh shareholder interests against its social mission. Its 5,300 employees hold an estimated $232 billion in combined equity, and non-founder employees could sell up to 10% of their holdings on the first trading day. Critics cite defense work and rapid AI development as tests of its safety commitments. A separate report says an $81 billion Anthropic IPO did not halt weakness in AI-related stocks.
Other developments span security, publishing and global access. Sakana AI announced an independent Security Research Group on November 20, 2026, led by Akihiro Umegai and CEO David Ha, to study AI-based cyber defense and AI-system safety. It plans to publish research, reproducible benchmarks, guidance and defensive playbooks, and reported Fugu-Cyber scores of 86.9% on CyberGym and 72.1% on CTI-REALM; the claims were not independently verified, and its exploit-code release policy is undecided. Meta is revising terms for its AI publishing platform after publishers raised ownership and control concerns, while Google and Anthropic have expanded media partnerships to cover more content types and formats. Meanwhile, the United States received about 72% of the world’s new AI computing capacity from 2022 through June 2026. Countries including India, Canada, France and the UAE are developing sovereign models, but reliance on US- or China-made open-weight systems, language gaps and limited safety testing raise questions about local fit and oversight.
Key Takeaways
- AI-related stocks face pressure as investors question whether heavy spending on chips and data centers can produce sufficient returns; the declines alone do not establish that companies are weakening.
- Daniel Loeb estimates $5 trillion to $7 trillion in AI investment from 2025–2030 would require $2 trillion to $3 trillion in new annual AI revenue to be justified.
- Hut 8 reported $71.1 million in third-quarter 2026 revenue, down 14.9%, and $26.6 billion in expected base-term AI data-center contract value; investors are watching cash generation.
- Cipher Digital’s shares lost more than half their value in 2026; the company had about $3.08 billion in net debt as of September and 18% short interest.
- Inference accounts for two-thirds of AI workloads. Positron says its 40-kilowatt Titan rack and modular-site approach could support token revenue within 90 to 120 days.
- Google is adding three AI buttons to its desktop homepage and says user AI searches, files and image requests may train models unless users change Gemini Apps Activity settings.
- Chicago Public Schools wants Google Search without AI; New York City and Los Angeles are reviewing Google’s policy for K–12 accounts.
- A bipartisan Senate bill would direct the SEC to set AI disclosure rules within 180 days, including governance, financing and risk information.
- Anthropic’s reported $20 billion IPO raises questions about balancing safety with shareholder demands; its CEO says safety and human control remain priorities.
- Only 13 of 25 AI customer-service products had enough public pricing information for EPR Research to estimate costs at three usage levels; the products’ billing measures were not directly comparable.
Michael Burry Says AI Selloff Proves He Warned Investors
Michael Burry said on X on November 19 that he had warned investors about risks tied to Nvidia, CoreWeave and Nebius Group, though he said he had not told them to short the stocks. By November 19, the shares had fallen 39%, 76% and 66%, respectively, from recent highs, while the VanEck Semiconductor ETF was down 33%. The broader selloff reflects investor doubts that heavy spending on AI chips and data centers will earn enough to cover costs, including financing and equipment depreciation. The article notes that the falling share prices do not by themselves prove the companies are weakening, and that Microsoft and Amazon face different financial risks.
NVIDIA’s Selloff Raises Questions About AI Investment Returns
NVIDIA shares began falling before major media reports about CEO Jensen Huang appeared on October 29, according to the article, though that timeline alone does not explain the decline. The article argues that the selloff may reflect wider concerns about whether AI companies can earn enough to justify large investments in data centers, GPUs and other computing equipment. Microsoft, Amazon, Google and Meta have committed major sums to AI infrastructure, while NVIDIA has benefited by supplying processors used to train and run AI models. The article distinguishes selling equipment from proving that the full AI industry can generate lasting economic returns.
Hut 8 Faces Questions Over Its AI Growth Valuation
Hut 8 ranked eighth on a list of 10 AI stocks viewed as overvalued by short sellers, as investors questioned whether AI infrastructure spending will produce enough returns. The company’s shares traded at a forward price-to-earnings ratio of 137.9, a price-to-sales ratio of 21.59 and an enterprise-value-to-sales ratio of 48.24. In the third quarter of 2026, revenue fell 14.9% to $71.1 million, while adjusted EBITDA rose to $33 million from $30.5 million a year earlier. Hut 8 reported $26.6 billion in expected aggregate base-term contract value across its AI data center portfolio, but investors are watching whether it can turn its growth plans into recurring revenue and sustainable free cash flow.
Cipher Digital’s AI Expansion Faces Valuation and Funding Risks
Cipher Digital shares lost more than half their market value in 2026 as investors weighed its AI data center plans, financing needs and valuation. Short interest reached 18%, as investors bet the company may struggle to produce enough earnings and cash flow to support its share price. Cipher is shifting from Bitcoin mining to AI and high-performance computing, but its net debt was about $3.08 billion as of September, adding to concerns about debt costs and future funding. The stock traded at about 18.24 times sales, and the article also warns that delays in data center construction could add pressure.
Daniel Loeb Says AI Spending Needs Trillions in New Revenue
In Third Point’s third-quarter investor letter, billionaire Daniel Loeb described the challenge of earning returns on the AI industry’s enormous spending. Forecasts point to $5 trillion to $7 trillion in investment by hyperscalers and leading AI developers from 2025 through 2030. Loeb estimates that the industry would need $2 trillion to $3 trillion in new annual AI revenue to justify the spending, equal to 33% to 50% of today’s roughly $6 trillion global IT market. He also cited rising borrowing costs for data center projects, while noting that Third Point sees strong AI demand and does not predict an industry collapse.
Short Sellers Target AI Stocks as Valuations Face New Pressure
A broad U.S. stock market sell-off has raised fresh concerns about high AI valuations and whether large investments in AI infrastructure will pay off. In early November, the decline erased about $3 trillion from the S&P 500’s market value, while the Nasdaq had its biggest single-day drop in 19 months. Investors including Michael Burry and Stanley Druckenmiller have voiced caution, and Barclays strategist Julien Lafargue warned that financing the data centers and computing systems needed for AI is becoming a growing challenge. Anthropic’s IPO raised $81 billion but did not stop the wider weakness in AI-related stocks. The article introduces five AI stocks that short sellers consider overvalued, but the provided text does not name them.
Google Adds Three AI Buttons to Its Homepage
Google is adding three buttons to its desktop homepage: Create images, Ask about files and Brainstorm. They can help users generate images, work with documents and plan tasks, and they appear for users signed in to Google. Google says AI searches, uploaded files and image requests may be used to train its AI models, though users can turn this off through Gemini Apps Activity settings. The company is also pushing AI into Search, a move that has drawn concern from school districts that want students to use Google Search without AI features.
Three Major School Districts Push Back on Google’s AI Search Policy
Google says schools that want students to use Google Search must accept AI Overviews, which summarize information at the top of search results. The company plans to turn off its AI tools for K-12 student accounts unless school administrators enable them, starting early next year, but says Search and AI are deeply integrated. Chicago Public Schools wants to offer Search without AI and may look for another search service if Google will not allow it. New York City and Los Angeles are also weighing the policy, while Miami-Dade’s response remains unclear. The article says the country’s four largest school districts have taken different positions on AI and Google’s policy.
Bipartisan Senate Bill Would Require New AI Disclosures
A bipartisan group of senators plans to introduce a bill requiring public companies involved in AI to disclose details about their AI operations and oversight. The proposal would direct the SEC to issue rules within 180 days, including requirements for companies to disclose whether they have an AI governance policy or a board committee focused on AI. It would also require certain companies to report information about AI-related financing arrangements, such as ownership stakes, debt, major business partners and contracts. Another section would add a standard AI risk section to Form 10-K, covering topics such as reliance on outside foundation models, computing providers, training costs, data rights and foreign AI models.
Only 13 of 25 AI Customer Service Tools Had Clear Public Pricing
An EPR Research audit on November 21, 2026, found that public pricing details allowed researchers to estimate costs for 13 of 25 AI customer service products at monthly volumes of 1,000, 5,000 and 10,000 interactions. The audit measured whether public terms made pricing possible, not product quality, and the products use different billing measures, so their estimated totals are not directly comparable. Twenty products had a clear public charge trigger, but seven still lacked enough details for full estimates because fees, limits or overage terms were unclear. The study used steady-state web-text support scenarios and excluded taxes, human labor, trials, optional integrations and negotiated discounts. It also showed that billable interactions can mean different things, such as a resolution, a session, multiple replies or credits.
Sakana AI launches independent security research group
Sakana AI announced an independent Security Research Group on November 20, 2026. Led by Akihiro Umegai, Chief of Security Research, and CEO David Ha, the team will study AI based cyber defense and the safety of AI systems. It plans to publish research, reproducible benchmarks, practical guidelines, and defensive playbooks, and to work with government, industry, and academia. Sakana reported that its Fugu-Cyber model scored 86.9% on CyberGym and 72.1% on CTI-REALM, but it provided no comparison benchmarks and those claims were not independently checked. The group has not yet set its policy for releasing exploit code.
Why Anthropic has not rescued the struggling AI trade
A Washington Post AI and Tech Brief examines why Anthropic has not been able to restore confidence in the AI industry. The article says the company has struggled to meet its promises and has relied too much on existing technology rather than new ideas. It argues that the lack of progress is affecting the wider industry, not just Anthropic. The brief says the industry needs to reconsider its approach to AI development.
Anthropic IPO puts its AI safety mission to the test
Anthropic raised a record $20 billion in its IPO, bringing new investor and shareholder pressures as critics question whether it can keep safety ahead of commercial success. CEO Dario Amodei said the company would take as long as needed to ensure its AI systems are safe and remain under human control. Anthropic is a public benefit corporation, so its board must weigh shareholder interests against its social mission. The listing gives the company’s 5,300 employees an estimated $232 billion in combined equity, and non-founder employees could sell up to 10% of their holdings on the first trading day. Critics also point to Anthropic’s defense work and the rapid pace of AI development as challenges to its safety commitments.
The global costs of a US China AI race
As Washington and Beijing compete for AI leadership, most other countries rely on systems they have little power to build, test, or govern. The United States received about 72% of the world’s new AI computing capacity between 2022 and June 2026. Several countries are developing sovereign models, including Sarvam in India, Cohere in Canada, Mistral in France, and Jais in the United Arab Emirates. But some models adapt open weight systems made in the United States or China and may carry their biases. The article also warns that language gaps and limited safety testing make it hard to ensure these systems reflect local cultures and needs.
Meta revises publisher terms as Google and Anthropic expand deals
Meta is revising terms for its AI publishing platform after publishers raised concerns about ownership and control of their work. The platform launched in September and lets publishers create and distribute content using AI. Google and Anthropic have expanded partnerships with media companies to cover more types and formats of content. The deals aim to give publishers more control over the final product while supporting AI powered content. The changes come as technology companies and publishers seek closer cooperation.
Inference now drives two-thirds of AI workloads
Inference now accounts for two-thirds of AI workloads, shifting spending from large training clusters to smaller sites that can be built quickly. Power, cooling and memory supplies are shaping where companies generate AI tokens, alongside chip performance. Positron AI Inc. uses commodity memory instead of high-bandwidth memory, which is costly and in limited supply. A Positron server uses about 4,500 watts, and a full rack of Titan servers fits within 40 kilowatts, making the system suitable for some modular data centers. Positron says these sites could help inference companies move from land and power to earning money from tokens within 90 to 120 days.
Sources
- Michael Burry Says ‘I Told You So’ as Wall Street Caught Off Guard by AI Selloff
- Beyond the Headlines: What NVIDIA’s Selloff Reveals About the AI Investment Cycle
- Is Hut 8 Corp. (NASDAQ: HUT) Priced for Too Much AI Growth?
- Is Cipher Digital Inc. (CIFR) Among the Most Overvalued AI Infrastructure Stocks?
- Billionaire Daniel Loeb’s ‘Bear Math’: AI’s $7 Trillion Spending Boom Faces a $3 Trillion Reality Check
- 5 Most Overvalued AI Stocks According to Short Sellers
- Even the Iconic Google Home Page Can’t Survive the Push for More AI
- Google told schools to accept AI or lose Search. Now 3 of America’s 4 largest districts are pushing back
- Securities Regulation Daily Wrap Up, AI NEWS—A bipartisan group of senators plans to introduce a sweeping securities disclosure bill, (Nov 20, 2026)
- AI Customer Service Agent Cost: 13 of 25 Vendors Priced
- Sakana AI Security Research Group: 86.9% Claim Checked
- AI & Tech Brief: Why Anthropic isn’t saving the AI trade
- Anthropic says it will put humanity first. Will its IPO change that?
- America, China, and the Costs of the Bipolar AI Race
- AI companies make concessions to publishers
- Inference drives two-thirds of AI workloads, changing chip deployment
Comments
Please log in to post a comment.