Alibaba May Have Helped Chinese Military-Linked Lab Access Advanced Nvidia Chips
1. Key Themes
Theme 1: AI-driven cyber risk is colliding with unpatchable legacy infrastructure, creating an outage and security crisis
Critical infrastructure is full of devices that can't be fixed
Microsoft's CEO frames the vulnerability as structural: the "smartification" of dumb devices created a permanent attack surface just as AI makes exploitation cheaper.
"Remember, there is a lot of stuff in our critical infrastructure that's unpatchable... think about all the firmware and the chillers [a cooling device often used in data centers] and the HVAC systems everywhere else. So all of that, by the way, they're on the internet."
"In other words: The smartification of the economy, in which much of the world's "dumb" devices were hooked up to the internet, is now causing tremendous headaches."
AI-powered vulnerability discovery is itself causing outages
The remediation response is generating failures: patching at unprecedented volume is breaking systems.
"A recent wave of outages at Microsoft, the FAA, Chase, the NSA, Workday, Apple, and United appears to stem largely from system upgrades gone wrong as AI tools uncover record numbers of software vulnerabilities and push IT teams into an unprecedented patching cycle."
Attackers are outpacing defenders
"Palmer Luckey observed that AI-powered cyberattacks are outpacing defenses because attackers are moving faster than organizations running outdated systems can respond, warning that the U.S. needs to accelerate adoption of AI for cybersecurity: 'It's time to start running.'"
Capital is flowing to startups that secure and monitor AI usage and infrastructure
"Vero Security... enforces company security policies across employees' AI tools and agents as products, permissions, and risks change, raised an $11 million seed round"
"Pretxt... monitors AI agents as they work and redirects actions that deviate from company procedures, raised a $5 million seed round"
"BrightAI... uses sensors, robots, drones, and wearable AI to monitor critical infrastructure and guide field crews, raised a $97 million round"
Theme 2: AI is reshaping geopolitics, regulation, and the China/US chip and model contest
Chip export controls are leaking via third-country data centers
"The New York Times reports that Alibaba used a network of Southeast Asian data centers to gain access to advanced Nvidia chips restricted from China, raising the possibility that those chips may have supported research at Zhejiang Lab, an AI institute co-founded by Alibaba with close ties to the Chinese military."
The U.S. is testing disclosure, not bans, to discourage Chinese AI model reliance
"The Trump administration is considering requiring U.S. companies to publicly disclose their use of Chinese AI models such as Alibaba's Qwen, DeepSeek, and Moonshot's Kimi as a way to discourage reliance on the technology without imposing new restrictions"
State-level AI regulation is escalating toward hard controls
"California Gov. Gavin Newsom is proposing sweeping new AI rules that would let state officials order shutdowns of dangerous models, require independent safety monitors inside AI labs, and curb recursive self-improvement, pitching the framework as a national template ahead of a likely 2028 presidential run."
Theme 3: The AI capital markets are at extraordinary scale, with large exits and a trillion-dollar IPO
Anthropic's IPO prices above range at a ~$1.5T valuation
"After reports yesterday that Anthropic was considering raising its IPO price above its initial $70-to-$78 range and selling more shares, the AI company priced the offering at $83 a share and increased its size, valuing Anthropic at roughly $1.5 trillion on a fully diluted basis."
Rapid AI M&A at steep valuations
"Palo Alto Networks has agreed to acquire one-year-old San Francisco AI startup Applied Compute, which helps companies train and run custom models on proprietary data, in a $4.5 billion all-stock deal just weeks after the company raised $350 million at a $3.25 billion valuation."
Theme 4: Funding is concentrating in AI "picks and shovels" for agents, documents, and physical-world data
Agent tooling and enterprise-data plumbing are getting seed capital
"Remanence, a two-month-old Paris startup that turns companies' AI-agent activity into specialized models they can train and own, raised a $30 million seed round co-led by EQT Ventures and FirstMark"
"Unsiloed AI... turns complex enterprise documents into structured data that AI agents can reliably use, raised a $7.5 million seed round led by Felicis"
Training data for robotics is emerging as a category
"Humyn Labs, a Bengaluru startup founded this year that records and labels footage of people performing everyday tasks to create training data for robotics models, raised a $26 million Series A round led by Accel"
AI agents applied to slow, permit- and paperwork-heavy industries
"Pulley... pairs permitting experts with AI agents to speed commercial construction approvals, raised $64 million across two rounds"
"Iconic... uses AI to review technical and commercial documents for inconsistencies in large infrastructure and industrial projects, raised a $10 million seed round"
2. Contrarian Perspectives
Philosopher says AI consciousness claims are "ridiculous" and labs may be misleading users
Against the prevailing industry narrative around model welfare and AI introspection, Nagel argues the apparent mental states are designed, not real.
"Philosopher Thomas Nagel, who this week won the $1 million Berggruen Prize for Philosophy & Culture, calls the idea that current AI systems might be conscious 'just ridiculous' and warns that companies such as Anthropic risk misleading people by designing chatbots to simulate introspection and apparent mental states."
Gradual "handover" of decision-making to AI is a "terrible idea," even from a leading AI CEO
This runs against the common trajectory of ever-more-autonomous agents and self-improving systems.
"OpenAI CEO Sam Altman warns that a gradual 'handover' of decision-making power from humans to increasingly capable AI is a 'terrible idea,' arguing that people must remain in control of the future even as AI systems become capable of improving themselves."
Nvidia's removal of its founder-CEO is framed as a self-inflicted loss of irreplaceable talent
"Elon Musk says Nvidia's decision not to reinstate Jensen Huang is like 'kicking Michael Jordan off your team,' arguing that Huang is irreplaceable and dismissing concerns that his affair with an alleged Chinese spy makes him disloyal to the U.S."
3. Companies Identified
- Description: Chinese tech giant; co-founder of Zhejiang Lab; maker of the Qwen model family
- Why mentioned: Allegedly used Southeast Asian data centers to access restricted Nvidia chips; its Qwen model is among those the U.S. may require disclosure for
- Quote: "Alibaba used a network of Southeast Asian data centers to gain access to advanced Nvidia chips restricted from China"
Nvidia
- Description: Leading AI chip maker
- Why mentioned: Chip access controversy; leadership turmoil after CEO ouster
- Quote: "Nvidia interim CEO Colette Kress told employees at her first companywide meeting that they will receive pay raises and more equity in December – months ahead of schedule – as the company tries to steady its workforce following Jensen Huang's ouster."
Microsoft
- Description: Software and cloud giant
- Why mentioned: Part of the outage wave; CEO commentary on unpatchable infrastructure
- Quote: "Nadella touched briefly on the recent Outlook situation, noting that the global inconvenience was the result of a 'latent bug' that led to a 'perfect storm.'"
- Description: AI lab
- Why mentioned: Pricing a blockbuster IPO; criticized by Nagel
- Quote: "the AI company priced the offering at $83 a share and increased its size, valuing Anthropic at roughly $1.5 trillion on a fully diluted basis."
Palo Alto Networks / Applied Compute
- Description: Cybersecurity company acquiring a one-year-old custom-model AI startup
- Why mentioned: $4.5B all-stock exit
- Quote: "in a $4.5 billion all-stock deal just weeks after the company raised $350 million at a $3.25 billion valuation. Applied Compute raised a total of $510 million"
- Description: Palo Alto infrastructure-monitoring startup
- Why mentioned: $97M raise; directly relevant to the infrastructure-risk theme
- Quote: "uses sensors, robots, drones, and wearable AI to monitor critical infrastructure and guide field crews, raised a $97 million round led by Smith Point Capital"
- Description: London spend-management startup
- Why mentioned: Largest valuation among growth rounds
- Quote: "raised a $280 million Series C round at a $2 billion valuation. The deal was led by Goldman Sachs Alternatives"
Variational
- Description: Onchain derivatives platform
- Why mentioned: Crypto/derivatives round at unicorn valuation
- Quote: "raised a $110 million Series B at a $1.5 billion valuation. The deal was led by Bain Capital Crypto"
Marrow Therapeutics
- Description: Watertown, MA RNA-delivery biotech
- Why mentioned: $173M combined A and B across two years
- Quote: "develops injectable medicines that deliver RNA therapies to bone marrow cells to treat severe blood disorders, raised a combined $173 million across Series A and Series B rounds"
Ramon.Space
- Description: Onboard satellite data-processing systems
- Why mentioned: Space compute infrastructure; Micron as strategic investor
- Quote: "raised a $124 million Series B round led by PSG, with Micron Technology, Senvest, and Phoenix Financial"
Hone Health
- Description: Physician-led longevity care
- Why mentioned: Equity plus large non-dilutive growth financing
- Quote: "It also secured $200 million in growth financing from General Catalyst's Customer Value Fund."
Remanence
- Description: Paris startup turning agent activity into owned, specialized models
- Why mentioned: $30M seed at two months old
- Quote: "a two-month-old Paris startup that turns companies' AI-agent activity into specialized models they can train and own, raised a $30 million seed round"
Humyn Labs
- Description: Robotics training-data startup
- Why mentioned: Physical-AI data layer
- Quote: "records and labels footage of people performing everyday tasks to create training data for robotics models, raised a $26 million Series A round led by Accel"
Pulley
- Description: AI-plus-experts construction permitting
- Why mentioned: $64M across two rounds
- Quote: "pairs permitting experts with AI agents to speed commercial construction approvals, raised $64 million across two rounds"
Vero Security
- Description: AI-tool policy enforcement
- Why mentioned: Security for AI usage
- Quote: "enforces company security policies across employees' AI tools and agents as products, permissions, and risks change"
Pretxt
- Description: AI agent monitoring and redirection
- Why mentioned: Runtime agent guardrails
- Quote: "monitors AI agents as they work and redirects actions that deviate from company procedures"
Unsiloed AI
- Description: Document-to-structured-data for agents
- Why mentioned: Enterprise data readiness
- Quote: "turns complex enterprise documents into structured data that AI agents can reliably use, raised a $7.5 million seed round led by Felicis"
Mulga
- Description: Open-source tooling to run AWS applications elsewhere
- Why mentioned: Cloud portability
- Quote: "develops open-source software that lets organizations run AWS applications on other public clouds or their own infrastructure"
Meta
- Description: Social media company
- Why mentioned: School partnership program alleged to be a teen-recruitment tool
- Quote: "Meta's school partnership program was designed in part to recruit more teens to Instagram through school profiles, student directories, and targeted notifications, even as Meta publicly promoted it as a safety initiative."
Zhejiang Lab
- Description: AI institute co-founded by Alibaba
- Why mentioned: Alleged beneficiary of restricted chips
- Quote: "an AI institute co-founded by Alibaba with close ties to the Chinese military"
4. People Identified
Satya Nadella
- Description: Microsoft CEO
- Why mentioned: Warned of unpatchable critical infrastructure at the DealBook Summit
- Quote: "'there is a lot of stuff in our critical infrastructure that's unpatchable,' Nadella said."
Palmer Luckey
- Description: Defense-tech founder
- Why mentioned: Warned AI attackers are outpacing defenders
- Quote: "It's time to start running."
Sam Altman
- Description: OpenAI CEO
- Why mentioned: Against a gradual handover of decisions to AI
- Quote: "a gradual 'handover' of decision-making power from humans to increasingly capable AI is a 'terrible idea'"
Thomas Nagel
- Description: Philosopher; Berggruen Prize winner
- Why mentioned: Skeptic of AI consciousness
- Quote: "calls the idea that current AI systems might be conscious 'just ridiculous'"
Elon Musk
- Description: Tech founder and executive
- Why mentioned: Defended Jensen Huang
- Quote: "like 'kicking Michael Jordan off your team'"
Jensen Huang
- Description: Nvidia CEO (ousted)
- Why mentioned: Subject of leadership controversy
- Quote: "dismissing concerns that his affair with an alleged Chinese spy makes him disloyal to the U.S."
Colette Kress
- Description: Nvidia interim CEO
- Why mentioned: Using accelerated pay and equity to stabilize the workforce
- Quote: "they will receive pay raises and more equity in December – months ahead of schedule"
Gavin Newsom
- Description: California governor
- Why mentioned: Proposing aggressive AI rules as a national template
- Quote: "pitching the framework as a national template ahead of a likely 2028 presidential run"
Jason Schreier
- Description: Bloomberg games reporter
- Why mentioned: Reviewed Grand Theft Auto VI
- Quote: "calls it a 'masterpiece,' praising its vast, unpredictable world, photorealistic detail, and dark humor after playing the game for more than 45 hours"
5. Operating Insights
Treat AI-accelerated patch cycles as an operational risk, not just a security win
The outage wave suggests that the remediation process, not only the attack, can cause downtime. Operators should stage rollouts, test upstream architecture, and plan incident response before mass-patching.
"Nadella... noting that the global inconvenience was the result of a 'latent bug' that led to a 'perfect storm.' ... 'We're going to obviously take all of the learnings, both upstream architectural, as well as operational, and fix that,'"
Inventory non-patchable and operational-technology assets, especially in data centers and buildings
Firmware-bound devices like chillers and HVAC are internet-connected and cannot be updated like standard software. Segmentation and monitoring become the main defenses.
"think about all the firmware and the chillers [a cooling device often used in data centers] and the HVAC systems everywhere else. So all of that, by the way, they're on the internet."
Prepare for disclosure obligations around Chinese AI models in your stack
If disclosure rules materialize, companies relying on Qwen, DeepSeek, or Kimi may face reputational and procurement friction. Audit model dependencies now.
"requiring U.S. companies to publicly disclose their use of Chinese AI models such as Alibaba's Qwen, DeepSeek, and Moonshot's Kimi as a way to discourage reliance on the technology without imposing new restrictions"
6. Overlooked Insights
Non-dilutive growth capital is becoming part of the venture stack
Hone Health pairing a Series A with a $200M facility from a VC firm's customer-value vehicle signals that revenue-based or customer-backed financing is scaling alongside equity.
"It also secured $200 million in growth financing from General Catalyst's Customer Value Fund. The company has raised a total of $65 million in equity funding."
Gen-Z AI companionship use is large and correlates with loneliness
A sizable share of young people use AI for companionship, and heavier reliance correlates with worse social outcomes, which hints at both a product opportunity and a regulatory/liability exposure.
"A Gallup and Boston Children's Hospital survey of 15,925 people ages 15 to 22 across 16 countries found that 33% use AI for companionship, rising to 50% among daily AI users, while those who rely on AI companions or online-only friendships reported higher odds of loneliness."