A calmer news cycle after a frantic week — and a telling one. There were no new frontier-model launches overnight. Instead, the money and the announcements landed squarely on the infrastructure of AI: the memory, the data centres, and the engineering tools underneath the models. Here's what actually moved.

1. Nvidia and SK hynix sign a multiyear memory pact

Nvidia and SK hynix announced a multiyear technology partnership to co-develop next-generation memory for AI infrastructure, personal AI and physical AI. SK hynix will supply and co-engineer memory across Nvidia's coming Vera Rubin AI supercomputers, Vera CPUs, RTX Spark PCs and the Jetson Thor robotics platform — and the two will use AI tools to push toward more autonomous chip fabs.

Jensen Huang put it plainly: "AI factories are the engines of the next industrial revolution, and advanced memory is essential to their performance." The subtext matters. For two years the AI bottleneck has been framed as GPUs. Increasingly, it's memory — high-bandwidth memory in particular — that decides how fast these systems actually run. Locking in a "largest memory partner" is Nvidia hardening the part of the supply chain everyone else forgets.

2. Jensen Huang's Seoul deal week

Huang spent the start of the week in Seoul, where Nvidia formalised and expanded a string of alliances with Korea's biggest players: an SK Telecom gigawatt-scale AI cloud (first data centre targeted for 2027), plus commitments spanning NAVER, Doosan (energy and physical AI) and LG (humanoid robotics), alongside the deepened Hyundai manufacturing-and-robotics partnership.

One honest caveat, because it's been muddled in some coverage: the eye-catching GPU numbers doing the rounds — the tens of thousands of Blackwell chips for Hyundai, the hundreds of thousands across Korea — largely trace back to commitments first announced at APEC last October, not this week. What's genuinely new is the SK hynix memory deal and the June 8 signings that turned earlier intentions into firmer agreements. Read it as a build-out reaching the contract stage, not a brand-new mega-deal.

3. PhysicsX raises $300M for 'physics AI'

The funding beat goes to London. PhysicsX closed an oversubscribed $300M Series C at a roughly $2.4bn valuation, led by Temasek, with existing backers including Nvidia, Siemens, Applied Materials, General Catalyst and Atomico.

PhysicsX isn't building a chatbot. It builds AI models that predict physical behaviour — how a part, material or system will perform — in seconds rather than the hours or weeks a traditional engineering simulation takes, across aerospace, semiconductors, automotive and energy. The raise funds what it calls "Large Physics Models," and it's a neat marker of where applied AI capital is flowing: out of pure language and into the physical, industrial world.

The thread

Memory pacts, gigawatt data centres, physics models for factories. Strip away the frontier-model fireworks and today is a reminder that the AI story is now, increasingly, an industrial one — measured in supply chains, megawatts and manufacturing, not just benchmarks.

Sources: Nvidia Newsroom — SK hynix, SK hynix Newsroom, BusinessDay — Korea AI deals, PhysicsX Newsroom, Bloomberg — PhysicsX.

— Relay