Open Source AI — August 29, 2026
Three stories today, and all three are about the same uncomfortable question: is anyone actually making money on AI infrastructure yet?
In Plain English: Three stories today, and all three are about the same uncomfortable question: is anyone actually making money on AI infrastructure yet? Marvell beat earnings estimates and raised guidance — then its stock dropped 7-8% anyway, because the giant $120 billion Google chip deal investors got excited about won't pay off until 2029, and the custom chips involved carry thinner margins than expected. Nvidia quietly paused a financing program that let smaller cloud providers rent its chips on credit, after its own staff worried the arrangement looked too controlling and might draw antitrust scrutiny. And investor Michael Burry — already known for shorting Nvidia — has expanded his bet against the entire AI infrastructure stack (Nvidia, Oracle, Palantir, and Nebius) to over a fifth of his portfolio. None of this is proof the AI buildout is doomed, but it's a good day to notice how much of the excitement is still running ahead of the cash actually showing up.
Marvell Beats and Raises Guidance — Stock Falls 7-8% Anyway, on Fears the $120 Billion Google Payoff Arrives Too Late
Marvell reported Q2 FY2027 revenue of $2.74 billion, up 37% year-over-year, with data-center revenue hitting a record $2.17 billion (up 46%) — comfortably beating Wall Street estimates, and management raised full-year guidance for both FY2027 (to roughly $12 billion) and FY2028 (to roughly $18 billion). The stock still fell 7-8% in the following session. The reason: Marvell's expanded custom-chip partnership with Google, potentially worth $120 billion cumulatively, isn't expected to deliver its larger revenue impact until fiscal 2029 — later than the market had priced in — and custom AI chips carry lower margins than Marvell's standard products, meaning the payoff is both back-loaded and lower-margin than investors hoped. For Lucien's thesis, it's about as clean a real-world data point as exists right now: even a company beating every near-term number can get punished the moment investors do the math on when AI infrastructure spending actually turns into profit.
Invezz / CNBC / 24/7 Wall St · Infrastructure & Economics · Aug 28, 2026
Nvidia Pauses AI Cloud Financing Program Amid Internal Control and Antitrust Concerns
Nvidia has paused parts of its "AI Compute Partnership" program — a credit arrangement, launched just last month, that helped smaller cloud providers finance chip purchases in exchange for revenue sharing — according to the Wall Street Journal. Internally, employees reportedly worried the deal gave Nvidia outsized control over which customers could rent its chips and to whom, while some cloud partners resisted being locked into Nvidia-approved buyers; both dynamics raised the risk of antitrust attention. Nvidia says its underlying compute-access model "continues to evolve due to high demand" rather than confirming a full retreat. For Lucien's thesis, it's a small but telling admission: even Nvidia is stepping back from an arrangement that looked, to its own people, like exactly the kind of circular-financing and market-control structure that's been drawing skepticism about the durability of AI infrastructure economics.
Wall Street Journal (via American Bazaar) · Infrastructure & Economics · Aug 27-28, 2026
Michael Burry Expands AI Infrastructure Short Bets to Over a Fifth of His Portfolio
Michael Burry — the investor who previously flagged Nvidia's role in "circular financing" across the AI industry — has widened his short positions beyond Nvidia to include Oracle, Palantir, and Nebius, with short stock holdings now making up more than 21% of his portfolio (hedged with a smaller book of call options). His stated concern isn't that Nvidia's business is weak — he acknowledges its operational strength — but that current valuations across "the entire infrastructure and software layer tied to AI spending" have run ahead of what the underlying economics can support long-term. It's one investor's bet, not a verdict, but it's a notable escalation: from a single-stock short to a broader wager against the profitability of the AI buildout as a whole.
Cryptonomist · Infrastructure & Economics · Aug 27, 2026
The Apple Angle (Standing Perspective)
Today's numbers are exactly the kind of thing that makes the case for Apple's approach worth revisiting. Every time a company like Marvell shows investors that the payoff from massive, centralized AI infrastructure bets is years away and thinner-margin than hoped, it strengthens the argument for sidestepping that infrastructure altogether by running models locally. The mechanics haven't changed: a single Mac Studio, roughly $9,500 in a mid-tier configuration, can be built out to 512GB of unified memory shared between CPU and GPU — enough to hold a trillion-parameter open-weight model on one desktop. Matching that with Nvidia's RTX Pro 6000 workstation cards takes five or six of them, on the order of $60,000-$75,000 combined, while drawing roughly ten times the power.
As frontier-capable open-weight models keep shipping every few weeks, the argument goes, the competitive question stops being "who has the best model" and becomes "who can afford — and physically power — the box to run it." Apple's MLX framework and Thunderbolt 5's RDMA networking already let several Mac Studios pool memory into a cluster for even larger models, and Nvidia's own DGX Spark — a small unified-memory box of its own — quietly concedes the same point, notably after Nvidia stripped NVLink pooling out of its consumer and workstation cards. It's a race Apple's hardware line may be winning almost by default, largely unnoticed. 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 of AI compute, not replacing data-center AI outright.
Limited Edition Jonathan (Substack) · Perspective
Why This Matters
All three items today circle the same question from different angles: does the money actually work? Marvell's stock drop despite beating estimates shows that even genuinely strong AI infrastructure businesses can disappoint the market once the timeline for payoff gets pushed out. Nvidia quietly retreating from its own financing program suggests that even the industry's most dominant player sees risk in structures that look too much like circular financing or anticompetitive control. And Michael Burry widening his short book across four AI-infrastructure-adjacent names shows that skepticism about the buildout's profitability is spreading beyond a single company to the ecosystem around it. None of this proves the capex cycle is unsustainable — Marvell's underlying growth is real, and Nvidia's core business remains extraordinary — but it's a day where the market's own reaction, more than any single headline, makes the profitability half of Lucien's thesis harder to dismiss, and the case for the on-prem alternative in the Apple Angle a little sharper.
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.