The headline number: $950 billion
The center of gravity in AI hardware shifted east this weekend. Samsung and SK Group together announced roughly $950 billion in AI partnerships with U.S. technology companies, according to thelec.net — a figure that would have sounded absurd two years ago and now reads as the going rate for a seat at the table.
The largest single piece is the Nvidia–SK Group pact, valued at more than $500 billion and confirmed by Reuters and Tom's Hardware alike. It is structured around three pillars aimed at next-generation AI infrastructure and memory technology — the unglamorous plumbing that determines whether all those GPUs actually have data to chew on. SK Hynix's high-bandwidth memory has quietly become as much of a bottleneck as the accelerators themselves, and Nvidia is paying to lock in supply rather than compete for it.
Samsung's contribution is a $200 billion partnership with Broadcom, announced Saturday, July 25, and running through 2030. Reuters, Fortune and CNBC all frame it the same way: this is a foundry win. A foundry is the contract-manufacturing arm of a chipmaker — the business of building someone else's designs — and it is the one part of the industry where Samsung has consistently trailed TSMC. A committed, multi-year order book of this size is the strongest hand Samsung has held against its Taiwanese rival in years.
TSMC answers with concrete
TSMC, for its part, is responding the way it usually does: by building. The world's largest advanced chipmaker is committing $265 billion to expanding U.S. operations, with Arizona as the anchor. The stated rationale is blunt — demand for advanced chips is outrunning supply, and has been for a while.
Nvidia is filling in the surrounding infrastructure. A $1.5 billion multi-year agreement with Amkor Technology will expand advanced packaging and testing capacity, also in Arizona. Packaging is where individual dies get assembled into finished, working parts, and it has become a genuine chokepoint; a chip you can't package is a chip you can't ship. Alongside TSMC's fabs, it starts to look less like a collection of announcements and more like a domestic supply chain with an actual back end.
Nvidia's lead, and everyone else's angle
Halfway through 2026, the scoreboard is not especially close. Coverage converging across The Globe and Mail, The Motley Fool and syndicated outlets points to the same read: Nvidia has run away with the AI chip race, while AMD and Intel are chasing different prizes rather than meeting it head-on.
That said, Intel had a good week. Reuters reports the company rebounding on AI optimism — a comeback narrative rather than a completed one, but a welcome change of tone. Sharing the spotlight is quantum-computing upstart IonQ, whose growth multiplied fivefold. Two very different companies, one direction of travel.
The most interesting structural bet may be the one nobody has to win a race to collect. Per Yahoo Finance, ARM is positioned to benefit regardless of which manufacturer ships the fastest processor, because its architecture underpins so much of what gets built in both smartphones and AI systems. Selling the blueprints beats betting on the builder.
Is any of this sustainable?
The question hanging over every one of these numbers is whether the boom breaks. Nvidia CEO Jensen Huang used a Fortune interview to push back hard, denying the chip boom is headed for a bust and making the argument every cycle eventually attracts: this time is different.
He may be right. He is also the single most interested party in that conclusion, and the phrase has a long and unflattering history. What's notable today is that the capital being committed is not speculative equity — it's multi-year supply agreements, foundry orders and physical plants in Arizona, the kind of spending that reflects demand someone has already signed for.
The skeptic's case doesn't rest on whether AI demand is real. It rests on whether $950 billion of it is, and on what happens to Samsung, SK and TSMC if the answer arrives in 2029 rather than 2026. Either way, the bets are placed.