Kulicke and Soffa Industries is advancing fluxless Thermo-Compression Bonding tools specifically for AI data centers, targeting silicon photonics and high-bandwidth networking. Analysts project the company will reach $1.8 billion in revenue and $500.1 million in earnings by 2029, driven by a 22.8% yearly revenue growth rate. The firm also confirmed a quarterly dividend of $0.205 per share payable on October 7, 2026.
In the nuclear sector, NuScale Power plans to launch nuclear-specific AI tools with partners Nuclearn and NPX to accelerate engineering cycles. The company aims to shift from a current loss of $415.7 million to $442.0 million in revenue and $49.7 million in earnings by 2029. However, NuScale faces cash burn and dilution pressure from a $750 million at-the-market offering, while a major power purchase agreement with TVA through ENTRA1 Energy remains critical for its future.
The U.S. Pentagon is considering a $5 billion loan to AI cloud startup Fluidstack to support the domestic data center supply chain. These funds, sourced from the Office of Strategic Capital, would boost manufacturing capacity for components rather than funding a new AI facility directly. Meanwhile, the Department of Justice is investigating whether Nvidia improperly handled its licensing deal with AI chip startup Groq, though a spokesperson argues the arrangement promotes innovation.
Security firms are also adapting to AI risks. CrowdStrike and Wipro launched a joint CISO Command Center to provide AI-driven cyber resilience, combining Falcon threat protection with Wipro's intelligence solutions. Gravwell released version 5.10 of its platform, introducing environment-aware AI agents that run inside customer deployments to automate security workflows. Additionally, the NSA, CISA, and FBI warned that China-based AI companies are using knowledge distillation to steal capabilities from U.S. models via proxy networks.
Talent and infrastructure shifts continue as Elementum hired Phil Duggan as chief revenue officer to lead global sales. Duggan brings 20 years of experience, and Elementum plans to increase its sales team fivefold in 2026. The company, an AI-native platform automating workflows inside existing data clouds, lists Snowflake among its major customers alongside Danaher and Sanofi. Researchers also introduced the RD-Forget framework, a training-free method helping AI agents manage memory by separating archived information from active query responses.
Key Takeaways
- Kulicke and Soffa targets $1.8 billion revenue and $500.1 million earnings by 2029 with new AI packaging tools.
- NuScale Power aims to reach $442.0 million revenue by 2029 despite current losses and a $750 million capital offering.
- The Pentagon is discussing a $5 billion loan to Fluidstack to bolster the U.S. data center supply chain.
- The DOJ is investigating Nvidia's licensing deal with Groq for potential antitrust violations.
- CrowdStrike and Wipro launched a joint AI security hub to address frontier AI risks for enterprises.
- Gravwell 5.10 introduced environment-aware AI agents to automate security workflows and reduce human error.
- The NSA, CISA, and FBI warned of China-based AI attacks using knowledge distillation to steal U.S. model capabilities.
- Elementum hired Phil Duggan as CRO and plans to expand its sales team fivefold in 2026.
- Elementum lists Snowflake as a major customer for its AI-native workflow automation platform.
- Researchers released the RD-Forget framework to help AI agents manage memory and prevent outdated facts from misleading answers.
Kulicke and Soffa advances AI packaging tools
Kulicke and Soffa Industries announced progress on fluxless Thermo-Compression Bonding tools designed for AI data centers. The company confirmed a quarterly dividend of US$0.205 per share payable on October 7, 2026. Analysts forecast revenues of US$1.8 billion and earnings of US$500.1 million by 2029, expecting a 22.8% yearly revenue growth rate. This new technology targets silicon photonics and high-bandwidth networking builds to meet AI infrastructure needs.
NuScale Power faces cash risks despite AI tools
NuScale Power plans to launch nuclear-specific AI tools with partners Nuclearn and NPX to speed up engineering cycles. The company aims to reach US$442.0 million in revenue and US$49.7 million in earnings by 2029, a shift from a current loss of US$415.7 million. However, the firm faces ongoing cash burn and dilution pressure from a US$750 million at-the-market offering. A major long-term power purchase agreement with TVA through ENTRA1 Energy remains a critical factor for future success.
Pentagon considers loan to Fluidstack for data centers
The Pentagon is in talks to lend about US$5 billion to AI cloud startup Fluidstack to support the U.S. data center supply chain. The funds would come from the Pentagon's Office of Strategic Capital and are intended to boost manufacturing capacity for data center components. This loan would not fund a new AI facility directly but would help shore up the supply chain for related parts. The Wall Street Journal reported these discussions on Thursday, citing people familiar with the matter.
DOJ investigates Nvidia licensing deal with Groq
The U.S. Justice Department is investigating whether Nvidia improperly handled its licensing deal with AI chip startup Groq. The inquiry began in December after the deal was announced and has led to a formal request for information from the agency. An Nvidia spokesperson stated that the American system promotes innovation and benefits consumers through such arrangements. The Justice Department could fine Nvidia if it finds the company mishandled the deal, though it is unlikely to unwind the transaction.
CrowdStrike and Wipro launch AI security hub
Wipro and CrowdStrike Holdings launched a joint CISO Command Center to provide AI-driven cyber resilience for large enterprises. The new hub combines CrowdStrike Falcon's threat protection with Wipro's intelligence solutions to support security operations. This initiative targets frontier AI risks and reinforces CrowdStrike's Project QuiltWorks coalition across multiple industries. CrowdStrike focuses on helping enterprises detect and respond to digital threats as AI deployments grow globally.
Gravwell 5.10 launches environment-aware AI agents
Gravwell released version 5.10 of its full-stack data analytics and security platform. This update introduces environment-aware AI agents that run directly inside a customer's deployment. These agents perform multi-step security workflows to provide real-time insights and automate operations. The new features aim to reduce human error and speed up responses to security incidents. Gravwell 5.10 is now available for customers to deploy immediately.
NSA CISA and FBI warn of China AI attacks
The National Security Agency, CISA, and the FBI released a joint advisory on September 10, 2026. They warned that China-based AI companies are using knowledge distillation to steal capabilities from U.S. models. These attackers use methods like premium subscriptions and proxy networks to bypass restrictions since late 2024. The agencies advise U.S. companies to detect malicious prompts and share intelligence across organizations. Immediate actions include altering responses for suspected attacks and monitoring anomalous network behaviors.
Elementum hires Phil Duggan as chief revenue officer
Elementum named Phil Duggan as its new Chief Revenue Officer to lead global sales. Duggan brings about 20 years of experience in enterprise technology and go-to-market strategies. He previously helped scale the international business for Astronomer. Elementum is an AI-native platform that automates workflows inside existing data clouds without needing complex integrations. The company plans to increase its sales team fivefold during 2026 to meet growing global demand. Major customers include Danaher, Elevance Health, Sanofi, Snowflake, and Under Armour.
New RD-Forget framework helps AI agents manage memory
Researchers introduced RD-Forget, a training-free framework that helps persistent language agents manage their stored information. This system separates what an agent keeps in its archive from what it uses to answer specific questions. It uses a frozen language-model curator to extract relevant evidence and group facts into semantic slots. The framework prevents old facts from misleading current answers while still allowing access to historical data. Tests show this method improves accuracy in conversational memory and long-context reasoning tasks.
Sources
- Is AI Packaging Progress Altering The Investment Case For Kulicke And Soffa Industries (KLIC)?
- Should AI Driven Nuclear Interest Require Action From NuScale Power Stock Investors?
- Pentagon in talks to lend $5 billion to AI cloud startup Fluidstack, WSJ reports
- DOJ probes Nvidia's licensing deal with AI startup Groq, NYT reports
- CrowdStrike (CRWD) Unveils CISO Command Center For AI Driven Cyber Resilience
- Gravwell 5.10 Introduces Environment-aware AI Agents for Security Operations
- CISA Releases Advisory on China-Based Attacks Targeting AI Companies
- Elementum Names Phil Duggan As Chief Revenue Officer As AI-Native Enterprise Platform Scales Global Sales
- What Should an Agent Forget? Separating What Is Stored from What Is Used
Comments
Please log in to post a comment.