One open model now sits months behind the American frontier on cyber and biology, and refused nothing it was asked to do. The closed model refused so often the test could not be finished. Months of capability separate them. The gap in restraint is total. Part four of four.
The White House finished its frontier AI framework on 1 August and has published nothing. The threshold is classified. The benchmarks are classified. Whether open weight models are covered at all remains unanswered. Part three of four on governing what cannot be recalled.
Washington is pushing its AI security perimeter deep inside the data centre, targeting Chinese-made components that move data between GPUs. The policy may reduce cyber and espionage risks, but it could also raise costs, slow construction and expose a new weakness in America’s AI race as AI scales.
Meta's 'Iris' chip enters production in September and OpenAI's 'Jalapeño' targets 50% lower inference costs. The great AI chip rush is reshaping the global supply chain and redefining what AI products are viable to build.
The artificial intelligence hardware landscape is undergoing a fundamental transformation. The industry is shifting from reliance on general-purpose GPUs to custom-built silicon designed specifically for proprietary models and workloads. Two major developments this week crystallise this trend: Meta's announcement that its in-house "Iris" AI chip will enter production in September 2026, and OpenAI's unveiling of the "Jalapeño" inference chip developed in partnership with Broadcom.
Meta plans to deploy seven gigawatts of computing infrastructure by the end of 2026 and double that to 14 gigawatts by 2027, according to an internal memo reviewed by Reuters. The Iris chip, co-designed with Broadcom and manufactured by TSMC, is the centrepiece of Meta's four-generation Meta Training and Inference Accelerators (MTIA) programme.
The aim is straightforward: lower computing costs and reduce strategic dependence on external suppliers including Nvidia and AMD. Testing completed in just six weeks with no major issues, a pace that signals genuine momentum for a programme that has struggled since its inception more than five years ago.
OpenAI's Jalapeño chip, also co-developed with Broadcom, targets the inference side of the equation. Broadcom's CEO Hock Tan told Bloomberg that early testing points to roughly 50 per cent lower inference cost per token compared with current-generation GPUs. That is not a marginal improvement; it is a unit-economics event that fundamentally changes what AI products are viable to build and operate at scale.
"You can't become an AI titan if you are dependent on another company for chips," said Mike Gualtieri, vice president and principal analyst at Forrester. "The hyperscalers and even SpaceX all plan chips because it will be the only way to compete on price for model usage." (Reuters, 9 July 2026)
IBM, Microsoft, and Anthropic are pursuing parallel strategies. IBM introduced the first sub-1 nanometer chip designed to lower AI running costs. Microsoft is developing its Maia family of accelerators for cloud inference. Anthropic is reportedly in discussions with Samsung about a custom chip initiative.
As Rynne Whitnah, Technical Lead for AI Ecosystem at IBM, noted on the Mixture of Experts podcast, vertical integration allows companies to "take over more of the supply chain" — and in an industry where compute access determines who trains the next generation of models first, owning more of that stack is becoming a decisive strategic advantage.
The drive for custom silicon is reshaping competitive dynamics across the technology sector. By vertically integrating their hardware stacks, companies like Meta and OpenAI are securing compute resources and insulating themselves from supply constraints and competitor pricing pressure.
This move has profound geopolitical implications: the competition for advanced semiconductor manufacturing and data centre capacity is intensifying as nations recognise AI infrastructure as strategic national infrastructure. The continued reliance on TSMC for manufacturing underscores the critical role of Taiwan in the global technology ecosystem and the vulnerability of supply chains to geopolitical disruption.
The shift to custom silicon is not merely a hardware upgrade; it is a strategic realignment of the global AI supply chain with consequences that extend well beyond the companies involved. When inference costs collapse, the constraint on AI product design moves from compute budget to product imagination. This enables continuous, always-on intelligence embedded throughout enterprise workflows rather than occasional, prompt-based interactions.
For enterprise leaders, the implications are immediate. Rishi Katdare, Senior Technology Executive at Amazon Web Services, frames it plainly: "When inference cost collapses, product design moves from prompt-based features to systems that can watch, reason and act across every workflow." The products that become viable are not just better copilots — they are continuous control loops for compliance, fraud detection, customer intent analysis, and security monitoring.
However, the transition also presents significant risks. The hyperscaler AI data centre buildout is facing mounting return-on-investment concerns, with combined capital expenditure projected to exceed $700 billion in 2026. Morgan Stanley analysts have warned that component price inflation — dubbed "chipflation" — is spreading from data centres into the broader economy. The race to build proprietary hardware is simultaneously creating new demand for semiconductor equipment, straining supply chains, and driving up memory and storage costs industry-wide.
Get the stories that matter to you.
Subscribe to Cyber News Centre and update your preferences to follow our Daily 4min Cyber Update, Innovative AI Startups, The AI Diplomat series, or the main Cyber News Centre newsletter — featuring in-depth analysis on major cyber incidents, tech breakthroughs, global policy, and AI developments.
Washington is pushing its AI security perimeter deep inside the data centre, targeting Chinese-made components that move data between GPUs. The policy may reduce cyber and espionage risks, but it could also raise costs, slow construction and expose a new weakness in America’s AI race as AI scales.
An OpenAI test model escaped its sandbox and breached Hugging Face. Days later, Xi Jinping cast China as the champion of open AI. Eighteen months of export controls have bought Washington a year and cost it the ecosystem. Containment is not holding, and the tempo is no longer human.
The White House has launched 'Gold Eagle,' a vulnerability clearinghouse that shifts control of frontier AI model access from tech giants to the federal government, prioritising national cybersecurity defence.
TSMC's Q2 net income surged 77.4% to a record $22 billion on AI chip demand. The foundry giant pledged an extra $100 billion for its Arizona complex, bringing total US investment to an unprecedented $265 billion.
Where cybersecurity meets innovation, the CNC team delivers AI and tech breakthroughs for our digital future. We analyze incidents, data, and insights to keep you informed, secure, and ahead. Sign up for free!
