Technology
DeepSeek · Nvidia · PHP2Shell · SpaceX
6 min read9 min listen
Listen to the podcast
DHS has prohibited federal civilian agencies from procuring DeepSeek technology and ordered the removal of DeepSeek products and third-party services developed using “DeepSeek implementations,” a term the order does not define. Nvidia separately announced agreements with six suppliers designed to expand aggregate substrate capacity for its AI platforms by 2.5-fold over three years, with major additions staggered through 2028. OpenAI will require approval for third-party model routers and gateways from November 20, while two CVSS 9.2 PHP vulnerabilities demonstrate the growing role of frontier models in security research. SpaceX reported $7.56 billion in quarterly AI infrastructure revenue and $16.44 billion in AI capital expenditure. Together, the developments show that AI competition is increasingly governed by procurement eligibility, distribution permission and control of physical capacity, not only model performance.
DHS immediately prohibited federal civilian agencies from procuring DeepSeek technology and ordered the removal of DeepSeek products and third-party products developed using “DeepSeek implementations.” The published order does not define that term.
Nvidia’s agreements target a 2.5-fold increase in aggregate substrate capacity over three years. The staggered schedule—from current Samsung volume to Unimicron in Q2 2027 and IBIDEN in Q1 2028—indicates that procurement relief will be phased rather than immediate.
Browse a few recent editions, or open the full archive.
The use of frontier models by Calif and Wiz to identify PHP2Shell, a CVSS 9.2 vulnerability affecting PHP 5.4 through 8.5, suggests that AI-assisted discovery can compress the time available to remediate legacy infrastructure.
SpaceX’s AI infrastructure segment generated $7.56 billion, up 830% year over year, and surpassed Starlink’s $5.25 billion. Its $431 million divisional operating profit and $16.44 billion in AI capital expenditure indicate that commercial traction and capital intensity are rising together.
OpenAI’s November 20 approval requirement for model routers, AI gateways and third-party direct-access providers indicates that enterprise inference middleware is becoming subject to upstream vendor permission, not only technical integration.
AI infrastructure buyers face separate but interacting constraints. Supplier financing determines when accelerator capacity becomes available, while procurement and usage rules determine whether a model or access layer can be deployed at all. For federal contractors, incomplete model lineage can now threaten contract eligibility even when a product does not carry DeepSeek branding. For enterprise technology teams, the OpenAI policy and PHP2Shell disclosures add near-term continuity work around inference gateways and exposed web infrastructure.
Unresolved variables that could shift the assessment materially.
Decisions this bears on
Coverage emphasizes headline AI revenue growth, a 2.5-fold substrate target, the DeepSeek exclusion and the severity of PHP2Shell. It underweights the timing and control of access: major substrate additions extend into 2027 and 2028, while DHS and OpenAI are imposing software-lineage and distribution controls now. The operative constraints are therefore who can reserve packaging capacity, prove model provenance and retain approved access to inference middleware.
1 source
2 sources
4 sources
1 source
4 sources
Indicators and developments to monitor in the coming days.
T1 Primary · T2 Analytic · T3 Commentary