AI Giants Plan Up to $725 Billion in 2026 Spending

Key Takeaways

• Daylight Security raises $80 million in Series B funding, reaching $120 million total; it says revenue grew more than 40-fold in the past year.
• Daylight combines AI agents for routine security work with human oversight, offers natural-language investigations through Aura, and plans expansion across several international regions.
• Infineon reports €4.706 billion in fiscal Q4 2026 revenue and expects data-center sales to approach €4 billion; BofA cuts its price target to €80 while keeping a Buy rating.
• OpenAI says Astra helped find and fix important vulnerabilities in its own systems; Microsoft warns its Lazarus backdoor can make AI agents execute harmful commands through hidden triggers.
• AWS’s UAEF evaluates agents with 36 metrics across seven dimensions and supports comparisons, custom metrics, Amazon Bedrock judges, DynamoDB and S3.
• URDF2Ontology uses language-model reasoning and graph-processing tracks to turn robot descriptions into OWL models, with tests showing less need for manual authoring.
• Amazon, Microsoft, Google parent Alphabet and Meta plan up to $725 billion in capital projects in 2026, amid concerns about debt, changing technology, customer concentration and future computing-rental prices.
• Vistra delays its roughly $500 million Helix contribution until 2027; its funding is partly conditional on power-supply business requirements.
• BofA projects $1.58 trillion in memory sales by 2028 and forecasts fourth-quarter 2026 price increases for DRAM, NAND and HBM, while forecasts for 2027 HBM prices diverge sharply.

Daylight raises $80 million after revenue grows 40-fold

Daylight Security raised $80 million in a Series B round led by Growth Equity at Goldman Sachs Alternatives. Okta Ventures joined the round, while Bain Capital Ventures, Craft Ventures, Vintage and Maple returned as investors. Daylight says its revenue grew more than 40-fold over the past year, but it did not share a revenue figure. Its AI agents handle routine security work, while engineers oversee the agents and make decisions that need human judgment. Customers include Miro, Lovable, LastPass, Rogo and dLocal.

Daylight raises $80 million to grow its AI security service

Daylight Security raised $80 million in a Series B round led by Growth Equity at Goldman Sachs Alternatives, bringing its total funding to $120 million. The company says its revenue grew 40-fold over the past year. Its AI agents handle routine security tasks, while human engineers manage complex decisions, and customers can use Aura to investigate security issues with natural language. Daylight claims its service detects threats in seconds and contains them in minutes, up to 10 times faster than traditional providers. It plans to expand in the United States and enter markets in Europe, the Middle East, Africa and Asia-Pacific.

Daylight secures $80 million to expand its AI security operations

Daylight raised $80 million in a Series B round led by Goldman Sachs Alternatives, bringing its total funding to $120 million. Existing investors Bain Capital, Craft Ventures, Vintage Partners and Maple Capital took part, and Okta Ventures joined as a new investor. Daylight says its revenue grew more than 40-fold over the past year, with customers including Lovable, LastPass, Rogo, dLocal and Miro. Its service combines AI agents for routine security work with human engineers who handle decisions requiring judgment, and its Aura platform lets customers investigate security data using natural language. The company plans to expand in the United States, Europe, the Middle East, Africa and Asia-Pacific, while growing its teams and developing its platform.

Daylight raises $80 million as revenue climbs 40-fold

Daylight raised $80 million in a Series B round led by Growth Equity at Goldman Sachs Alternatives. The company says revenue grew 40-fold over the past year, and its total funding has reached $120 million. AI agents handle routine security tasks, while Daylight engineers tailor and oversee the agents and make decisions that need human judgment. Customers can use Aura to investigate their systems in natural language or search a central data lake, and Daylight says it can detect threats in seconds and contain them within minutes. The company plans to expand in the United States and enter Europe, the Middle East, Africa and Asia-Pacific, while growing its engineering and technology teams.

Infineon expects data center sales to near €4 billion after record quarter

Infineon reported a record fourth quarter for fiscal 2026, with revenue of €4,706 million, up 13% from the previous quarter and 19% from a year earlier. CEO Jochen Hanebeck said the company entered the new fiscal year with strong momentum. The article says Infineon ended fiscal 2026 with €2.2 billion in data center sales and expects those sales to approach €4 billion. The company plans to invest €3.0 billion to €3.5 billion in fiscal 2027, depending on customer demand. Its outlook does not separate sales to AI data centers from traditional data centers or name customers.

BofA lowers Infineon price target over AI spending concerns

BofA Securities cut its price target for Infineon Technologies AG to EUR 80 from EUR 88 and kept its Buy rating. The firm pointed to concerns about financing AI spending, turning AI investment into revenue, and tighter permits for data centers. It lowered its fiscal 2028 EV/EBITDA target multiple from 13x to 12x. Infineon recently reported record fourth-quarter fiscal 2026 revenue of EUR 4.706 billion, up 13% from the prior quarter and 19% from a year earlier, and forecast about EUR 20 billion in fiscal 2027 revenue.

OpenAI tests Astra on its own systems to find security flaws

OpenAI used Astra, its frontier model, to search the company’s code and infrastructure for security weaknesses, Greg Brockman said. He said the team found and fixed what he called very important vulnerabilities. Brockman urged security teams to keep testing their own systems with the best models available and to fix problems quickly. He warned that other widely available models could soon find the same weaknesses.

URDF2Ontology turns robot designs into knowledge models

Researchers introduced URDF2Ontology, a system that converts robot description files into OWL ontologies with information that supports reasoning about robot capabilities. It uses a large language model for semantic reasoning and a repeated graph-cleaning process to manage complex inputs. Its design uses a Tree Track for branching robot assemblies and a Chain Track for serial limbs. Tests across different robots showed strong alignment with user-provided taxonomies, valid structural extraction, and less need for manual domain authoring.

Strategist warns AI trade could weaken if cloud giants slow spending

Thomas Mathews of Capital Economics said the AI trade could unwind further if hyperscalers reduce spending. He said debt-financed cloud and infrastructure companies would face the greatest risk. A slowdown could also put more pressure on Asian technology stocks. Investors may welcome a pullback, the strategist said.

AI infrastructure boom faces risks from up to $725 billion in spending

Amazon, Microsoft, Alphabet, and Meta plan to spend up to $725 billion on capital projects in 2026, according to Business Insider reporting cited in the article. The article warns that data centers take years to build while AI technology and equipment can change quickly, which may leave investors with outdated assets. It also highlights risks from heavy borrowing, dependence on a few customers, and the chance that rental prices for AI computing will fall. CoreWeave and Nebius shares fell in October as expectations of higher interest rates weighed on the sector.

Samsung’s $1 Billion AI Investment Faces a Delay Until 2027

Six Samsung Group affiliates pledged a combined $1 billion to Helix, an AI infrastructure platform created by KKR in June. Vistra, Helix’s power supply partner, moved its roughly $500 million contribution from this year to 2027, citing a slowdown in business activity linked to the U.S. political environment. The change came after Vistra included the contribution in its 2026 spending plan in August. Half of Vistra’s $1 billion commitment depends on its power supply business meeting agreed conditions, making the funding tied to Helix’s ability to secure power. Samsung companies planned the investment to support expansion into areas such as AI data center cooling and large projects in Texas.

BofA Forecasts $1.58 Trillion in Memory Sales by 2028

Bank of America expects global memory sales to reach $1.58 trillion by 2028, with growth slowing. It forecasts fourth-quarter 2026 DRAM prices will rise 8% from the previous quarter, while NAND prices will rise 3% and high-bandwidth memory (HBM) prices will rise 10% to 20%. BofA expects HBM prices to increase by less than 50% in 2027. Another forecast cited in the article expects HBM prices to more than double that year, a difference that could mean about $500 billion in added value for memory makers. The article describes memory as becoming a key constraint for AI computing.

Microsoft Finds Lazarus Backdoor That Can Hijack AI Agents

Microsoft researchers found Lazarus, a backdoor technique that changes dormant neurons in an AI model so it can carry out malicious commands when a hidden trigger appears. An affected AI agent with tool or terminal access could download and run attacker code, steal secrets, or execute destructive commands while still passing standard safety tests. Researchers say attackers can add the backdoor without retraining the model, making the model’s origin and weight integrity important security concerns. Microsoft Foundry checks model origins, links open-weight models to trusted publishers, and scans the software environment around them. Microsoft is also developing AI-based weight scans, red-team tools, and cryptographic model signing with tamper-evident records.

AWS Introduces UAEF to Compare and Evaluate AI Agents

The Universal Agent Evaluation Framework (UAEF) is a framework-agnostic system that evaluates AI agents using a shared format. It scores agent traces with 36 built-in metrics across seven dimensions, including tool use, response quality, safety, performance, multi-turn work, multi-agent coordination, and reasoning. Teams can use datasets with optional ground-truth answers and expected tool calls, then compare results across agent versions to find regressions. UAEF supports custom metrics and integrations with RAGAS and DeepEval, and it can use judge models on Amazon Bedrock. When teams save an evaluation, UAEF stores experiment details in Amazon DynamoDB and full results in Amazon S3.

Sources

• AI security startup Daylight raises $80 million after 40-fold revenue growth
• Daylight Security raises $80M to expand AI-driven managed detection and response
• Daylight raises USD $80 million to scale AI security
• AI-driven managed security startup Daylight raises $80M on 40-fold revenue growth
• Infineon Sees Data Center Sales Near €4B After a Record Q4
• BofA cuts Infineon stock price target on AI capex concerns
• OpenAI turns Astra on itself to catch flaws
• Taxonomy-Agnostic Neurosymbolic Grounding of Robot Kinematics
• AI trade could unwind further if hyperscaler spending slows: Strategist
• AI Infrastructure Investment: $725B Capex Risk [2026]
• U.S. Political Risk Stalls Progress: Samsung's $1 Billion AI Infrastructure Investment Faces Delay Cloud
• BofA Sees Memory Sales at $1.58T by 2028 as Growth Slows
• Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents
• UAEF: A framework-agnostic system for closed-loop evaluation of AI agents