🛡️ Anthropic touts AI cybersecurity project with Big Tech partners ↗
Anthropic rolled out Project Glasswing, a program that gives select partners a look at its unreleased Claude Mythos Preview model for defensive cybersecurity work. Amazon, Microsoft, Apple, CrowdStrike, Palo Alto Networks, Google, and Nvidia are all in the mix - a fairly formidable guest list.
The company said Mythos Preview has already found thousands of major vulnerabilities across operating systems, browsers, and other software. It is also widening access to about 40 more organizations tied to critical software infrastructure, while setting aside up to $100 million in usage credits and $4 million in donations for open-source security groups. (Reuters)
🌐 AI networking firm Aria Networks raises $125 million in funding ↗
Aria Networks raised $125 million in its first series round as it tries to build networking gear for AI data centers straining under demand. The startup is backed by Sutter Hill Ventures, Atreides Management, Valor Equity Partners, and Eclipse Ventures - hardly a timid cap table.
Its pitch is flexibility: Aria says its network can work with AI chips from Nvidia and Google, so customers can swap hardware without rebuilding everything. It also introduced what it calls the first AI-native network, centered on “token efficiency” - a slightly ungainly phrase, perhaps, though the idea is simple enough: more AI output for less running cost. (Reuters)
🏭 Intel joins Musk's Terafab AI chip project to power humanoid, data center goals ↗
Intel said it is joining Elon Musk’s Terafab project alongside SpaceX and Tesla to make processors for robotics and data center ambitions. Intel’s stock rose more than 2% after the announcement, which feels modest beside the scale of the plan, but still - markets noticed.
Intel says its capabilities should help Terafab chase a target of 1 terawatt per year of compute for future AI and robotics systems. Musk had already framed the broader effort around two advanced factories in Austin, with one aimed at cars and humanoid robots and the other aimed at AI data centers in space - yes, in space, which sounds like a pitch deck drifting into science fiction. (Reuters)
🚕 Uber bets on Amazon's custom chips to boost AI efforts ↗
Uber is leaning further into Amazon’s custom silicon, using AWS chips to speed up computing and train AI models. The expanded deal brings in Graviton for smoother rides and deliveries, plus Trainium for model training - a practical move, not especially flashy, though those often matter most.
Uber says the aim is to improve ride-matching, sharpen the app experience, and make recommendations more personal. For Amazon, it is another nudge in its campaign to make in-house chips feel like a serious alternative in enterprise AI, or at least serious enough that rivals have to glance over their shoulder. (Reuters)
⚡ US power use to beat record highs in 2026 and 2027 as AI use surges, EIA says ↗
The U.S. Energy Information Administration said power consumption is set to climb again after already hitting a record, with demand rising from 4,195 billion kWh to 4,244 billion and then 4,381 billion. A large share of that increase is tied to AI and crypto data centers, alongside broader electrification across homes and businesses.
The same outlook sees renewables gaining share, coal slipping, and natural gas staying stubbornly central. AI keeps getting described like software, yet it is turning into an energy story too - servers in hoodies, basically. (Reuters)
FAQ
Why is AI infrastructure suddenly the main theme in tech news?
This roundup shows AI moving beyond model releases and into the systems that make those models usable at scale. The focus is on cybersecurity, networking, chips, cloud hardware, and electricity demand. Taken together, those areas suggest the next phase of AI is not just about smarter software, but about the physical and operational stack that supports it.
What is Anthropic’s Project Glasswing, and why does it matter for defensive cybersecurity?
Project Glasswing gives selected partners access to Anthropic’s unreleased Claude Mythos Preview model for defensive cybersecurity work. According to the article, the model has already identified thousands of major vulnerabilities across operating systems, browsers, and other software. The program also broadens access to more critical infrastructure organizations and includes credits and donations for open-source security groups.
How does AI infrastructure benefit from new networking companies like Aria Networks?
Aria Networks is pitching hardware built for AI data centers that are under strain from rising demand. Its core promise is flexibility, allowing customers to use AI chips from Nvidia and Google without rebuilding their entire setup. The company also frames its system around “token efficiency,” which the article describes as producing more AI output at a lower running cost.
Why are companies like Uber using Amazon’s custom chips instead of relying on standard options?
The article says Uber is expanding its use of AWS chips, including Graviton and Trainium, to speed up computing and train AI models. The goal is practical: improve ride matching, refine the app experience, and make recommendations more personalized. It also shows Amazon positioning its in-house chips as a more credible option for enterprise AI workloads.
What does Intel joining Musk’s Terafab project say about where AI hardware is headed?
It suggests the industry is moving toward much larger, more specialized compute capacity tied to robotics and data centers. Intel said its capabilities could help Terafab pursue a target of 1 terawatt per year of compute for future AI and robotics systems. The project’s scope, including factories for cars, humanoid robots, and space-focused AI data centers, points to highly ambitious infrastructure planning.
Why is AI infrastructure becoming an energy story as well as a software story?
The article points to U.S. electricity demand reaching new records in 2026 and 2027, with AI and crypto data centers as major contributors. That means AI growth is becoming increasingly tied to physical resource use, not just model performance. It also highlights a broader shift in which power generation, fuel mix, and grid capacity become part of the AI conversation.