Open Source AI — August 27, 2026
Nvidia had its biggest quarter ever — $96 billion in revenue — and, for the first time, gave investors a year-ahead forecast: 70% more growth in fiscal 2028, a number clearly aimed at quieting "AI bubble" and "circular financing" critics.
In Plain English: Nvidia had its biggest quarter ever — $96 billion in revenue — and, for the first time, gave investors a year-ahead forecast: 70% more growth in fiscal 2028, a number clearly aimed at quieting "AI bubble" and "circular financing" critics. But the same report showed memory-chip shortages squeezing Nvidia's own margins and pushing it to raise AI-server prices roughly 15% — a live example of infrastructure costs climbing even as demand booms. In a related twist, Nvidia itself turned out to be the buyer in this week's Hugging Face sale, agreeing to pay $12.9 billion for the site that hosts nearly every open-weight model — putting a hardware giant in charge of open AI's main distribution hub. AWS also locked in 2 million more Nvidia GPUs for 2027–2028, showing the buildout isn't slowing down. The standing Apple argument below lands with extra weight today, given how directly this ties to rising Nvidia hardware costs.
Nvidia Posts a Record $96 Billion Quarter and Guides to 70% Growth Next Year — While Flagging Memory-Driven Price Hikes and Margin Pressure
Nvidia reported fiscal Q2 2027 revenue of $96.22 billion, beating estimates, with data center revenue alone hitting $89.02 billion, up 117% year-over-year. For the first time, the company issued a year-ahead forecast: 70% revenue growth for fiscal 2028, roughly $100 billion above what analysts had modeled — a number CFO Colette Kress explicitly framed as an answer to "circular financing" critics, arguing that financing frontier AI labs is low-risk given how the business has scaled. But gross margins, at 75% this quarter, are guided down to 71–72% by Q4 as memory-chip costs keep climbing, and Nvidia is reportedly passing roughly 15% of that cost increase on to AI-server customers — a concrete, current example of the rising per-unit infrastructure cost this thesis is tracking, even as demand and revenue keep breaking records.
Fortune / CNBC / 24/7 Wall St · Infrastructure & Economics · Aug 26, 2026
Nvidia — Not an Outside Buyer — Turns Out to Be Acquiring Hugging Face, for $12.9 Billion (Follow-Up)
Two days after reports first surfaced that Hugging Face was fielding roughly $13 billion in acquisition interest (covered in this brief on Aug 26), The Information and Reuters confirmed the actual buyer: Nvidia, at $12.9 billion — nearly triple Hugging Face's $4.5 billion valuation from its 2023 funding round. The platform has long stayed neutral, hosting competing open-weight models from Google, Amazon, Microsoft, and others with their own chip ambitions; under Nvidia ownership, that neutrality is now an open question. It's a notable twist for Lucien's thesis: the same week Nvidia is defending its GPU-rental economics against "circular financing" critics, it's also moving to control the main distribution channel for the open-weight models that would otherwise let people route around Nvidia's cloud partners entirely.
Tech Startups / The Information / Reuters · Open-Weight Models · Aug 26, 2026
AWS and Nvidia Commit to 2 Million More GPUs for 2027–2028
AWS and Nvidia expanded their partnership to deploy 2 million additional GPUs — Blackwell Ultra, Rubin, and Rubin Ultra — across AWS's global infrastructure in 2027–2028, on top of a prior 1-million-GPU commitment made earlier this year. It's a concrete sign that, cost pressure and bubble talk aside, the largest cloud provider is still locking in multi-year, massive-scale GPU capacity — the flip side of the profitability question this thesis is watching.
GlobeNewswire (AWS/Nvidia) · Infrastructure & Economics · Aug 26, 2026
The Apple Angle (Standing Perspective)
Today's Nvidia numbers make the standing case here almost write itself: the same earnings report that produced a 70% growth forecast also disclosed that memory shortages are pushing Nvidia to raise AI-server prices by roughly 15%, with gross margins sliding as a result. That's exactly the dynamic some commentators argue favors Apple's approach over Nvidia's. A Mac Studio, at roughly $9,500 in a mid-tier configuration, can be built out to 512GB of unified memory — enough to run a trillion-parameter open-weight model on a single desktop machine. Matching that footprint with Nvidia's RTX Pro 6000 workstation cards takes five or six of them, roughly $60,000–$75,000 combined, at about ten times the power draw — and that gap only widens as Nvidia's own memory costs rise and get passed through to customers.
As frontier-capable open-weight models keep shipping every few weeks, the argument goes, the real competitive edge shifts from who has the best model to who can afford — and physically power — the box to run it locally. Multi-Mac clusters using Apple's MLX framework and Thunderbolt 5's RDMA networking can already pool memory across several Studios for even larger models, and Nvidia's own DGX Spark quietly concedes the point by adopting a unified-memory design of its own. It's a race Apple's hardware line may be winning almost by default, especially as its chief GPU rival raises prices — though the fair caveat still holds: a Mac Studio serves one person running one model locally, while a data-center GPU cluster serves many people concurrently, so this looks like Apple leading a specific, valuable segment rather than replacing data-center AI outright.
Limited Edition Jonathan (Substack) · Perspective
Why This Matters
Today's news sits close to the center of Lucien's thesis from two directions at once. Nvidia's own earnings report — record revenue, an unprecedented 70% growth forecast, and margins compressing under memory-cost inflation that's pushing AI-server prices up roughly 15% — is close to a live case study in the profitability tension this brief tracks: extraordinary demand paired with rising per-unit infrastructure costs and vocal "circular financing" pushback. Layered on top, Nvidia moving to buy Hugging Face outright reframes the open-weight story: it's less a question of whether open models keep closing the capability gap (they are) and more a question of who controls the pipes those models travel through. Set against the standing Apple argument, which gets sharper every time Nvidia hardware gets pricier, today reads as a company doubling down on controlling both the compute and distribution layers of commercial AI, even as the economic case for renting that compute keeps facing scrutiny.
The Daily: Open Source AI — a recurring research brief for Lucien Engelen
Book Lucien for your next event
Keynotes, masterclasses, panels and board-room sessions.