Anthropic Nvidia and AMD Shares Fall as AI Returns Face Scrutiny

Several reports point to growing pressure to show that AI spending delivers measurable returns. Palo Alto Networks beat fiscal first-quarter expectations, reporting revenue of $3.44 billion, up 39%, and adjusted earnings of $1.05 per share. It raised its fiscal 2027 forecast to $14.52 billion–$14.57 billion in revenue and $4.26–$4.30 in adjusted earnings per share. Next-generation security annual recurring revenue reached $9.73 billion, up 66%; Prisma AIRS, its AI security product, reached nearly $170 million in annual recurring revenue in five quarters and more than 1,000 customers. CEO Nikesh Arora estimated the wider AI security market could reach $30 billion–$50 billion in annual spending by 2029, and said vulnerabilities are being found nearly seven times more often than last year.

Palo Alto also agreed to acquire AI startup Applied Compute for $4.5 billion in stock, aiming to build its enterprise AI security business toward $1 billion in annual recurring revenue by fiscal 2029. The company plans more than 35 AI-powered security applications, secure sovereign open-weight models and over 100 megawatts of computing capacity over three to five years, with less than $3 billion in capital spending through fiscal 2030. Arora said open-weight models’ share of token use rose from below 5% to above 60% in a year, and warned that organizations may need tens of thousands of virtual patches while permanent updates can take three to nine months. Palo Alto’s Unit 42 team also identified an AI-assisted hacking campaign targeting 19 organizations; attackers accessed systems or data at 10 and exposed more than one million personal records at one institution.

Meanwhile, AI and technology stocks sold off as investors questioned whether heavy AI investment will pay off quickly. Anthropic shares dropped more than 10% in their second day of trading, reaching an intraday low of $86.80 after its record-sized listing; another report put the shares at $87.28, about 10% below Thursday’s close and near the $83 IPO price. The Nasdaq fell more than 1.2% intraday, while Nvidia declined 3.63%, AMD fell more than 6%, and Alphabet overtook Nvidia in market capitalization. A separate report said Nvidia’s market value had fallen to $3.39 trillion from a recent $5.7 trillion peak. NVIDIA Tokenised bStocks (NVDAB) slid 10.73% to $140.41. Bitcoin fell 2.3% to $61,874.40, its lowest in nearly five months, and Ether dropped 3.6% to $1,742.05; spot Bitcoin ETFs recorded $1.12 billion in outflows so far that week. The reports also cited interest-rate concerns and possible tighter U.S. crypto rules.

Companies are also debating how to measure AI’s business value. In a CloudZero survey of 260 finance executives, 71% said they were measuring the wrong metrics. Infosys said AI revenue rose 29% quarter over quarter to 10.5% of sales but lowered its revenue-growth forecast. Snowflake CEO Sridhar Ramaswamy argued that AI use matters when it improves operations and replaces existing costs; Snowflake says customers using its AI to automate tasks see costs fall 30%. Lucidity introduced its AI Action Engine at Microsoft Ignite, using agents to identify and validate cloud savings and implement changes with customer approval.

Research and education initiatives offer other snapshots of AI’s development. A study of Reinforcement Learning from Calibrated Decision Rewards trained Qwen3-1.7B on MATH and raised MATH500 accuracy from 72.2% to 76.8%, near the 76.9% achieved with an answer key. Accuracy on the hardest level rose from 44.8% to 51.7%; the method also scored 0.772 F1 on ProcessBench for locating the first incorrect step, versus 0.722 for a dedicated 7B process reward model. StarTV-Agent, a traffic-analysis system combining visual evidence with spatial and temporal reasoning, improved results by up to 8.0% across three benchmarks while reducing token use by up to 52.1%. Morehead State University received a five-year, $3 million federal grant for 2026–2031, including an AI Center for faculty and student training, responsible-use education, advising tools and science lab equipment. Its endowment challenge will combine $300,000 raised by the university with $600,000 in federal matching funds.

Security teams face a related challenge as AI agents are deployed and retired. Gartner predicted that more than 40% of agentic AI projects would be canceled by the end of 2027, citing costs, unclear business value and weak risk controls. The article warns that agents may leave active credentials behind: agents using a person’s session can be disabled by removing that user’s access, but those with their own API keys, service accounts or tokens need a separate retirement process. Reporting on agent identity frameworks introduced at RSA Conference 2026 found none could confirm that a retired agent had zero remaining credentials.

Key Takeaways

  • Palo Alto Networks reported Q1 fiscal 2026 revenue of $3.44 billion, up 39%, and raised its fiscal 2027 revenue outlook to $14.52 billion–$14.57 billion.
  • Palo Alto’s Prisma AIRS reached nearly $170 million in annual recurring revenue in five quarters; the company plans more than 35 AI security applications and over 100 megawatts of computing capacity.
  • Palo Alto agreed to acquire Applied Compute for $4.5 billion in stock and is targeting $1 billion in Applied Compute-related annual recurring revenue by fiscal 2029.
  • Anthropic shares fell more than 10% after its record-sized IPO; Nvidia and AMD also declined amid a broader technology selloff and doubts about AI spending returns.
  • Bitcoin fell to $61,874.40, while spot Bitcoin ETFs saw $1.12 billion in outflows that week; NVDAB dropped 10.73% to $140.41.
  • A CloudZero survey found 71% of 260 finance executives believed they were measuring the wrong AI metrics. Snowflake says customers using its AI to automate tasks see a 30% cost reduction.
  • Lucidity launched its AI Action Engine to identify and validate cloud savings, with customer approval required before changes are carried out.
  • Calibrated AI judging raised Qwen3-1.7B’s MATH500 accuracy from 72.2% to 76.8%; StarTV-Agent cut token use by up to 52.1% across three traffic-analysis benchmarks.
  • Morehead State received a five-year, $3 million grant to expand AI literacy and establish an AI Center; its endowment challenge aims to combine $300,000 in university funds with $600,000 in federal matching.
  • Gartner forecast that over 40% of agentic AI projects could be canceled by the end of 2027, while reporting found identity frameworks unable to verify that retired agents have no credentials left.

Palo Alto CEO Sees AI Security Market Reaching $50 Billion

Palo Alto Networks beat its fiscal first-quarter guidance, raised its fiscal 2027 outlook, and agreed to buy AI startup Applied Compute for $4.5 billion in stock. CEO Nikesh Arora estimated that companies could spend $30 billion to $50 billion a year on AI security by 2029, based on projected token sales of $1 trillion and security spending of 3% to 5%. Palo Alto’s Prisma AIRS product has reached nearly $170 million in annual recurring revenue after five quarters. Arora’s estimate covers the whole industry, while analysts expect Palo Alto and CrowdStrike to generate a combined $29.2 billion in fiscal 2029 revenue. Palo Alto shares rose 6%, despite already trading at 81 times forward earnings before the rise.

Palo Alto Raises Outlook as AI Security Demand Grows

Palo Alto Networks shares rose after the company raised its full-year forecast. CEO Nikesh Arora said critical software vulnerabilities are being found nearly seven times more often than last year, increasing demand for virtual patching and AI security tools. The company’s Unit 42 research team found an AI-assisted hacking campaign that targeted 19 organizations in August and September; attackers accessed systems or data at 10 and compromised more than one million personal records at one institution. Palo Alto reported fiscal 2026 first-quarter revenue of $3.44 billion, up 39%, and next-generation security annual recurring revenue of $9.73 billion, up 66%. Its Prisma AIRS product generated nearly $170 million in annual recurring revenue within five quarters and had more than 1,000 customers.

Palo Alto Shares Climb After Strong Results and $4.5 Billion AI Deal

Palo Alto Networks shares rose more than 7% after the company reported fiscal first-quarter revenue of $3.44 billion, up 39% from a year earlier, and adjusted earnings of $1.05 per share. It raised its fiscal 2027 forecast to adjusted earnings of $4.26 to $4.30 per share and revenue of $14.52 billion to $14.57 billion. The company also agreed to buy AI startup Applied Compute for $4.5 billion in an all-stock deal. Next-generation security annual recurring revenue reached $9.73 billion, up 66%. CEO Nikesh Arora said companies may need tens of thousands of virtual patches by the end of the fiscal year, while permanent updates can take three to nine months to install.

Palo Alto Plans More Than 35 AI Security Tools and $3 Billion in AI Capacity

Palo Alto Networks shares rose 8% overnight after an earnings beat, as CEO Nikesh Arora described a major opportunity to secure open-weight AI models. He said these models’ share of token use rose from below 5% to more than 60% in a year. Palo Alto bought Applied Compute for $4.5 billion to expand its enterprise AI security business, with a goal of reaching $1 billion in annual recurring revenue by fiscal 2029. The company plans to develop more than 35 AI-powered security applications and provide secure, sovereign open-weight models. It also plans to build more than 100 megawatts of computing capacity over three to five years, with less than $3 billion in capital spending through fiscal 2030.

NVDAB Falls 10.73% as Nvidia and AI Stocks Weaken

NVIDIA Tokenised bStocks, or NVDAB, fell 10.73% in 24 hours to $140.41 on Friday, as Nvidia and other AI-related stocks lost ground. A November 20 report linked the decline to investor concerns about high valuations and AI spending after Anthropic’s November 19 debut. The article says Nvidia’s market value fell to $3.39 trillion from a recent peak of $5.7 trillion, and that weakness in Nvidia shares puts pressure on the token. Bitcoin fell 3.91%, while the total crypto market value dropped about 3.76%; U.S. spot Bitcoin ETFs recorded $2 million in net outflows on November 19. The report says selling pressure could ease if NVDAB holds near $140.41 and Bitcoin stays above $60,000, while a fall below either level could bring further weakness.

Bitcoin falls below $62,000 as tech selloff hits crypto

Bitcoin fell 2.3% to $61,874.40 on Friday, its lowest level in nearly five months, and was down 4.1% for the week. The drop came as investors sold technology stocks and questioned whether heavy spending on artificial intelligence can pay off. Spot Bitcoin exchange traded funds saw $1.12 billion in outflows so far this week, their third straight week with more than $1 billion withdrawn. Investors also faced concerns about higher U.S. interest rates and possible tighter crypto rules under Democrats taking office in 2027. Ether fell 3.6% to $1,742.05, while XRP, Solana, BNB, Cardano, Dogecoin and $TRUMP also declined.

Morehead State gets $3 million grant to expand AI literacy

Morehead State University received a five-year, $3 million Title III Strengthening Institutions Program grant for 2026 through 2031. The U.S. Department of Education awarded the grant, which aims to help the university serve low-income students and improve academic quality, management and financial stability. The university plans to open an AI Center in Camden-Carroll Library to train faculty and staff, teach students AI literacy and support responsible AI use in courses. The grant will also fund student advising tools, lab equipment for the Science and Engineering Building expected to be completed by Spring 2027, and an endowment challenge. Over five years, the endowment effort will combine $300,000 raised by MSU with $600,000 in federal matching funds, for a total of $900,000.

A calibrated AI judge boosts language model math scores

Researchers studied Reinforcement Learning from Calibrated Decision Rewards, or RLCDR, which trains language models using an AI judge that reports how sure it is about its decisions. The method uses an LLM to assess answers and identify errors, with its probability estimates adjusted using about a hundred labels. Training Qwen3-1.7B on MATH raised its MATH500 accuracy from 72.2% to 76.8%, close to the 76.9% achieved with an answer key. On the hardest level, accuracy rose from 44.8% to 51.7%, and the reward stayed better calibrated than a softmax judge on the same 27B backbone. Without step training, the model also beat a dedicated 7B process reward model at finding the first wrong step, with a ProcessBench F1 score of 0.772 compared with 0.722. When trained only on problems the base model mostly failed, RLCDR still improved scores by 1.9 points but fell short of the answer key because its probabilities became overconfident.

AI firms must prove cost savings to keep customers paying

AI companies often point to measures such as increased automated activity, but those figures may not show whether customers are getting useful results or value for their money. Infosys said AI revenue rose 29% from the previous quarter to 10.5% of sales, yet it lowered its revenue growth forecast. In a survey of 260 finance executives by CloudZero, 71% said they were measuring the wrong metrics. Enterprise leaders told Fast Company that disagreements over success can affect prices and contract renewals, especially when automated work creates extra tasks or a cheaper model delivers the same result. Snowflake CEO Sridhar Ramaswamy said higher AI use is worthwhile when it improves operations and replaces existing costs. Snowflake reported that customers using its AI technology to automate tasks see a 30% reduction in costs.

Anthropic shares plunge as AI selloff erases Nasdaq gains

Anthropic shares fell more than 10% in their second day of trading on Friday, reaching an intraday low of $86.80 as investors sold AI and technology stocks. The Nasdaq Composite dropped more than 1.2% intraday, giving up all its gains for the year. Nvidia fell 3.63% and lost its market capitalization ranking to Alphabet, while Micron dropped 6.45%, AMD fell more than 6% and Taiwan Semiconductor Manufacturing’s ADR declined more than 4%. The MSCI All-Country World Index was heading for its seventh straight decline as investors moved toward bonds, pushing the 10-year Treasury yield below 5%. Gold and bitcoin also weakened, while fund managers’ overweight position in tech stocks fell by 25 percentage points, the biggest drop in about 20 years.

Anthropic shares fall in rocky Wall Street debut

Anthropic’s stock fell in its Wall Street debut after the company held the largest listing on record. It was last trading at $87.28, about 10% below Thursday’s close and near its $83 initial public offering price. Investors have grown concerned that spending on artificial intelligence, expected to top $1 trillion this year by some estimates, is not bringing returns quickly enough. Robert Pavlik of Dakota Wealth said investors appeared skeptical of the AI market, while Carsten Hesse of Berenberg said conditions were not good for Anthropic. Options trading is set to begin next week and may add to the stock’s volatility.

How to safely retire AI agents and their credentials

Gartner predicted in mid-2025 that more than 40% of agentic AI projects would be canceled by the end of 2027, citing rising costs, unclear business value, and weak risk controls. The article says canceled projects may leave agents’ credentials active, creating a security risk. Agents that use a person’s session can be retired by removing that person’s access, while agents with their own API keys, service accounts, or tokens need a separate retirement process. Effective tools should confirm that no credentials remain, find the systems and access an agent used, and revoke access automatically. Reporting on agent identity frameworks introduced at RSA Conference 2026 found that none could confirm a retired agent had zero remaining credentials.

StarTV-Agent improves traffic analysis while cutting token use

Researchers introduced StarTV-Agent, a system that combines visual evidence with spatial and time-based reasoning to understand traffic scenarios. It uses policy learning and experience reuse to make decisions more efficiently, with a fast path for tasks that match stored workflows and a ReAct-based path for other tasks. Its training methods include Role-Sensitive Process Optimization and Weakness-Driven Demonstration Learning. The system also uses feedback on task completion, workflow use, and cost savings to improve how it reuses workflows and updates memory. Across three challenging benchmarks, StarTV-Agent improved performance by up to 8.0% relative and cut token use by up to 52.1%.

Lucidity launches AI tool to automate cloud cost savings

Lucidity launched its AI Action Engine at Microsoft Ignite to help manage cloud costs and resource use. The tool uses AI agents to find and validate savings opportunities, then carries out changes with customer approval. Lucidity says the engine aims to close the gap between cost-saving recommendations and putting them into action. The launch reflects an industry effort to address cloud inefficiencies through more streamlined and autonomous tools.

Sources

NOTE:

This news brief was generated using AI technology (including, but not limited to, Google Gemini API, Llama, Grok, and Mistral) from aggregated news articles, with minimal to no human editing/review. It is provided for informational purposes only and may contain inaccuracies or biases. This is not financial, investment, or professional advice. If you have any questions or concerns, please verify all information with the linked original articles in the Sources section below.

AI investment returns AI security Palo Alto Networks Prisma AIRS Applied Compute acquisition enterprise AI cybersecurity agentic AI AI governance AI metrics cloud cost optimization AI research reinforcement learning Qwen3 StarTV-Agent AI education technology stocks Anthropic IPO cryptocurrency market AI agent identity

Comments

Loading...