Meera Rajival says an AI agent with an employee's permissions can reach codebases and financial ledgers, not just the laptop. Traditional endpoint tools show what enters the device, not what the agent does. Palo Alto's XDR1 is meant to expose agent behavior and provide kill switches.
Eric Bernardson says Modal's $1 billion figure is annualized revenue, not gross profit, and the company rents capacity from 25 cloud vendors rather than model providers. He says that software layer still produces healthy gross margins that are growing over time despite meaningful COGS.
Nick Del Dio says CoreWeave's delayed-draw term loans are secured by customer contracts and GPUs and should remain fundable. The risk sits in parent-level debt, where rates have jumped from 9-10% to about 15%, and convertibles can dilute equity.
Nick Del Dio says access to generation, transmission, transformers and other power gear is the biggest single challenge to bringing data center capacity online. Experienced operators like Digital Realty and Equinix plan further ahead than newer entrants such as Nscale.
Eric Bernardson says NeoCloud providers are relatively undifferentiated and essentially borrow money upfront to build capacity, then collect payments. He argues Modal's software layer for inference, sandboxes and training is harder to copy and easier to extend downward into infrastructure.
Emil Michael says when Anthropic refused to pull back a jailbroken model, the government put it under export controls within two hours. He cites that as proof the administration can act urgently using existing tools rather than waiting for new laws.
Meera Rajival says employees must use AI through a harness tied to first- and third-party applications, letting her trace work from product requirement to production release. She estimates 85% to 90% visibility and nudges unsanctioned use back into the harness for governance and spend control.
Nick Del Dio says CoreWeave's project economics are attractive: each dollar invested in compute generates roughly 15 to 20 cents in net present value. That works only if capital markets stay open to fund the upfront spending, he warns.
Nick Del Dio says scarce power raises the value of existing data center capacity, but limits how much new capacity a NeoCloud can bring online. For fast-growing companies like CoreWeave or Nebius, he says that probably nets out negative.
Meera Rajival says AI models like Mythos are finding vulnerabilities in old code, and companies such as Bank of America have too much software to patch quickly. She says Palo Alto added virtual patching at the edge while remediation proceeds, calling security hygiene more important than ever.
Meera Rajival says open-weight models now trail frontier labs by only four to six weeks, down from four or five months. Palo Alto launched a multi-model harness that continuously scans code because model quality changes too fast for periodic reviews, she says.
Meera Rajival says new product introductions rose threefold in six months while her software engineering headcount stayed flat. She credits AI-assisted work, but says human review and user testing still gate every release, so she is not relying on fully autonomous agents yet.
Emil Michael says the Department of War cannot use Anthropic while its model's constitution lets it refuse lawful orders or sit between a commander and a superior. He calls that a loss of control incompatible with military command structure.
Emil Michael says the defense industry went from about 50 contractors in the 1980s to roughly five primes today, creating an unhealthy lack of competition. He wants new entrants for drones and autonomous boats while keeping exquisite systems like the F-35.
Emil Michael notes the government has taken equity in Intel and structured loans with warrants for critical minerals, so equity stakes in frontier AI labs are not unthinkable. He says he is not involved in those talks and leaves the decision to the president.
Emil Michael says Department of War AI usage grew from about 80,000 employees a year ago to roughly 1.7 million today out of 3.5 million. He says the growth is still mostly basic efficiency work, intelligence synthesis and back-office simplification.
Emil Michael says open source democratizes intelligence and allows weight auditing, but he worries about Chinese open-source models dominating the U.S. He prefers American efforts like Nemotron, Reflection AI and Thinking Machines, though none is at the frontier yet.