Open Source AI — September 18, 2026
Four stories today, all pointing at the same underlying question: who actually owns the hardware AI runs on? OpenAI is asking investors to nearly double its valuation to $1.5 trillion even as its own profit margins are shrinking.
In Plain English: Four stories today, all pointing at the same underlying question: who actually owns the hardware AI runs on? OpenAI is asking investors to nearly double its valuation to $1.5 trillion even as its own profit margins are shrinking. Apple is reportedly designing a dedicated AI server for offices and governments that want to run AI in-house rather than renting it from the cloud. Huawei just moved up its next AI chip's launch by two quarters, racing to give the market a cheaper alternative to Nvidia. And infrastructure startup Crusoe raised $3.9 billion partly to build smaller, truck-delivered data centers instead of only giant campuses. Today's Apple Angle looks at the irony that Apple's own enterprise AI server news landed on exactly the argument this brief keeps tracking.
OpenAI Seeks a $1.5 Trillion Valuation While Its Own Profit Margins Shrink
OpenAI is reportedly weighing a new private funding round that would value the company at up to $1.5 trillion - nearly double the $852 billion it was valued at just months ago, and roughly double the $730 billion mark from this past March - while some investors have countered with a lower $1.2 trillion figure. The company's annualized revenue topped $40 billion in August, about double where it stood at the end of 2025, but its operating margins are reportedly narrowing even as that top-line number climbs. CEO Sam Altman has said "right now would be an ill-advised moment to go public," pushing a planned IPO out to 2027 while OpenAI chases an even bigger private valuation in the meantime. For Lucien's thesis, this is prediction #1 condensed into one company's own numbers: investors are being asked to pay nearly twice as much for a business whose margins are moving in the wrong direction, a bet that revenue growth alone will eventually out-run the cost of running it.
Infrastructure & Economics · Forbes · Sep 16, 2026
Apple Is Reportedly Designing Its Own AI Server - Aimed Straight at the On-Prem Market
Apple is reportedly developing a dedicated enterprise AI server built around its own M8 Ultra chips, in two-chip and four-chip configurations, aimed at developers, businesses, and government agencies that want to run AI inference on their own premises rather than through a cloud API. To let those chips talk to each other fast enough for serious workloads, Apple is said to be evaluating Nvidia's NVLink Fusion interconnect technology, which Nvidia recently opened up to outside hardware makers. The project isn't expected to ship before 2029 and could still be shelved, but the signal is notable: OpenAI and Anthropic are already reportedly buying Mac Minis and Mac Studios in bulk for their own AI workloads, and Apple's Mac revenue jumped about 29% last quarter to $10.4 billion. For Lucien's thesis, this is prediction #3 showing up inside Apple's own product roadmap: the company behind the consumer hardware in the standing Apple Angle below is now reportedly building a version of that same idea explicitly for the enterprises and governments this brief argues will eventually want to own their AI hardware outright.
On-Prem/On-Device Shift · The Decoder · Sep 16, 2026
Huawei Moves Up Its Next Nvidia-Rival AI Chip by Two Quarters
Huawei announced at a Shanghai conference that its next-generation Ascend 960DT AI chip will now launch in the first quarter of 2027 - six months earlier than the Q3 2027 date it had previously given - alongside a new AI SuperPoD system, built on what Huawei calls a Peerium computing architecture, that links thousands of accelerator chips together with fast internal networking. Huawei's own framing is explicit: the goal is to close China's AI computing gap with the United States and give the market a real alternative to Nvidia's hardware, and the accelerated timeline comes despite ongoing US export restrictions on advanced chips, just ahead of a planned September 24 meeting between Presidents Trump and Xi. Independent benchmarks aren't yet available, so Huawei's performance claims remain the company's own. For Lucien's thesis, this doesn't change who owns the model weights, but it's more evidence for the underlying shift the thesis expects: an entire second hardware ecosystem is racing to make owning AI compute cheaper and less dependent on one supplier, whoever ends up building it.
Infrastructure & Economics · TechCrunch · Sep 17, 2026
Data Center Startup Crusoe Raises $3.9 Billion - to Build Smaller Facilities, Not Just Bigger Ones
Crusoe raised $3.9 billion in a Series F round, valuing the AI infrastructure company at $30.9 billion, with Nvidia among the investors alongside Atreides Management, Mubadala Capital, and Valor Equity Partners. Rather than putting all of that money into ever-larger gigawatt data center campuses, Crusoe is also building "Spark" - factory-built, modular data centers that can be trucked to wherever spare electrical capacity already exists, instead of waiting years for a new mega-facility and its dedicated power buildout to be approved and constructed. The company's own framing ties the shift to two practical pressures: getting compute online faster, and sidestepping the local community opposition that increasingly slows or blocks giant data center projects. For Lucien's thesis, this is a well-capitalized infrastructure insider quietly hedging the industry's default bigger-is-always-better data center model - smaller, faster, distributed compute is a different bet than the one behind the trillion-dollar campuses this brief has been tracking, though it's still centralized, rented infrastructure rather than owned, on-prem hardware.
Infrastructure & Economics · TechCrunch · Sep 17, 2026
The Apple Angle (Standing Perspective)
It's a strange coincidence that today's news already includes Apple: reports that the company is designing its own dedicated AI server for enterprises and governments, years before it would actually ship, are effectively a preview of the argument this section makes every day - just with Apple's own product roadmap now openly agreeing with it. The underlying case doesn't need Apple's future server to work, though - it's built entirely around hardware Apple already sells today. A single Mac Studio, maxed out with 512GB of Apple's unified memory for roughly $9,500, can hold an entire trillion-parameter open-weight model in memory on one machine - no server room, no dedicated cooling, no grid negotiation required.
Matching that memory footprint on Nvidia's own workstation-grade hardware currently takes five to six RTX Pro 6000 cards, something like $60,000-$75,000 combined, while drawing roughly ten times the electricity. As frontier-capable open-weight models keep shipping every few weeks, the practical question keeps sliding from "which lab trained the smartest model" toward "who can actually afford to own the hardware to run a nearly-as-good one, locally, without asking anyone's permission." Apple's MLX software and Thunderbolt 5's RDMA networking already let multiple Mac Studios pool memory for larger jobs, Nvidia has stripped NVLink from its own consumer workstation cards, and Nvidia's DGX Spark increasingly borrows Apple's unified-memory playbook rather than the reverse. The honest caveat still holds: a Mac Studio serves one person running one model at a time, while the trillion-dollar infrastructure bets covered elsewhere in this brief are built for enormous concurrent demand across millions of users - Apple looks positioned to quietly own a valuable, low-friction corner of AI compute, not to replace the data center outright.
Limited Edition Jonathan (Substack) · Perspective
Why This Matters
Today's four items sharpen prediction #1's central tension from several angles at once: OpenAI is asking to be valued at $1.5 trillion on margins moving the wrong direction, while Huawei and Crusoe are both racing to make AI compute cheaper and more flexible than the industry's current default. Apple's own reported enterprise AI server shows the standing Apple Angle's argument starting to show up inside a trillion-dollar company's own roadmap - not just in outside commentary.
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.