ASUS Takes a $2B Bet That the AI Supercycle Has Years Left to Run
Perplexity Open-Sources Its Agent Sandbox, Bets on Neoclouds for Distribution. Altman Lays Out His Three Endgames for AI Safety.
ASUS Takes a $2B Bet That the AI Supercycle Has Years Left to Run
ASUS just closed its first offshore syndicated loan — and investors didn't just meet the ask, they blew past it. What started as a $1B target got upsized to $2B on strong demand, arranged by HSBC through ASUS's Singapore subsidiary [4]. The money is earmarked across Cloud AI, Edge AI, and Physical AI — basically every layer of the stack ASUS already touches as the world's largest motherboard maker and a growing AI server builder [4].
CFO Nick Wu framed it as international lenders backing ASUS's deepening AI strategy, and analyst Dan Nystedt (@dnystedt) flagged it as another data point in the broader "AI supercycle" financing wave [5]. This isn't a glamorous story, but it's a telling one: hardware suppliers are now raising debt at scale specifically because physical AI — robots, edge inference, on-device compute — is expected to need infrastructure money for years, not quarters.
Perplexity Open-Sources Its Agent Sandbox, Bets on Neoclouds for Distribution
Perplexity gave away something it didn't have to: SPACE, its Firecracker microVM sandbox for running AI agent code securely, is now Apache 2.0 licensed [6]. It's been running in production inside Perplexity's Computer agents since July, survived 108 red-team tests with zero VM escapes, and is now shipping through partnerships with CoreWeave, Akamai, and Crusoe [6][7].
Crusoe (@CrusoeAI) is folding SPACE into its Managed AI portfolio alongside GPU and power offerings — a clear signal that compute providers see "secure agent execution" as a must-have layer, not a nice-to-have. Aravind Srinivas (@AravSrinivas) made the thesis explicit: neoclouds need full agent platforms, not just raw silicon, if they want to compete for enterprise workloads [7].
This matters more than it looks. Every company racing to deploy autonomous agents has been quietly building — or ignoring — the security layer that lets an agent write and run code without torching your infrastructure. Perplexity just made that layer free and auditable. Expect it to become default infrastructure fast.
Altman Lays Out His Three Endgames for AI Safety
In an internal briefing reported this week, Sam Altman sketched three scenarios for how AI governance eventually settles: an IAEA-style international body doing inspections and audits (reportedly his favorite), a CERN-style merger of labs into one multinational entity (closer to Demis Hassabis's stated preference), or one company winning outright and "melting" everyone else's GPUs — which Altman called "terrible" [8][9].
None of this is new territory for AI policy circles, but the fact that OpenAI's CEO is talking endgames at all — rather than just guardrails — tells you leadership inside the labs now assumes concentration or international control is coming, not optional. It's landing alongside renewed pushes for an AI Whistleblower Protection Act and other governance proposals, suggesting the window for self-regulation is narrowing [8].
What This Means For Your Business
Today's stories are really one story told four ways: judgment is becoming the scarce resource, and code — whether it's robot control software or agent sandboxing infrastructure — is becoming a commodity you can rent by the hour or download for free. Dream-1 didn't win by being cleverly engineered for one robot; it won by generalizing, the same way LLMs stopped being "chatbots" and became reasoning engines you point at anything. If your team is still building bespoke automation for specific hardware or specific workflows, you're racing against models that don't care about your specifics.
Perplexity giving away SPACE and Crusoe immediately wrapping it into a paid cloud product is the orchestration economy in miniature: the hard infrastructure gets open-sourced and commoditized, and the money moves to whoever assembles it well, secures it, and operates it reliably at scale. ASUS raising $2B isn't a bet on any single model — it's a bet that someone, somewhere, will need the compute layer underneath all of this for a long time. None of these companies are competing on "who wrote better code." They're competing on who orchestrates faster, safer, and cheaper.
And Altman's three endgames are a reminder that the judgment question isn't just technical — it's structural and political. Whoever controls access, auditing, and standards for frontier AI will shape what every business downstream is allowed to build. If you're making AI infrastructure decisions right now, the smart move isn't picking the best model or the cheapest API — it's building systems flexible enough to swap either out when the ground shifts again, because it will.
Key takeaway: The code — robot control, sandboxing, even the models themselves — is getting faster, cheaper, and more open by the week. What's not getting commoditized is the judgment to orchestrate it well, govern it responsibly, and bet capital on the right layer of the stack.
Sources
- https://dreamlabs.ai/research/dream-1/
- https://www.forbes.com/sites/johnkoetsier/2026/10/19/robot-brains-for-75-centshour-former-nvidia-researchers-raise-90-million/
- https://www.unite.ai/dream-labs-raises-90m-seed-to-bring-robot-ai-across-different-machines/
- https://www.businesstimes.com.sg/companies-markets/asus-secures-us2-billion-hsbc-arranged-loan-support-ai-push
- https://x.com/dnystedt/status/2112329373739372601
- https://siliconangle.com/2026/10/15/perplexity-ai-open-sources-its-secure-ai-agent-sandbox-space/
- https://cellcog.ai/blog/perplexity-space-sandbox/
- https://cellcog.ai/blog/altman-three-ai-endgames/
- https://finance.biggo.com/news/ae7a55fe-369e-43b0-af9f-307da110a0f9
Stay ahead of AI
No spam. Unsubscribe anytime.
Want to go deeper?
Reading the news is one thing. Exploring the frontier is another. See what we're building.