Open Source AI — September 3, 2026
Two stories today, both about the scale and politics of the AI buildout rather than a new model.
In Plain English: Two stories today, both about the scale and politics of the AI buildout rather than a new model. PwC, the consulting firm, projected that global AI infrastructure spending will hit $31.6 trillion by 2050, growing every year instead of leveling off the way past infrastructure booms did — and its report pointedly says nothing about whether any of it pays for itself. Separately, at a G20 gathering of world leaders, Nvidia's Jensen Huang and OpenAI's Sam Altman pushed countries to keep building and adopting AI, but Altman let slip a telling concession: governments are free to choose whether to build their own data centers or simply rent capacity from providers like his. That build-or-rent fork is exactly the question Lucien's thesis is watching. Nothing today addressed a new open-weight model directly, but both stories sharpen the economic and political backdrop the thesis depends on.
PwC Forecasts $31.6 Trillion in Global AI Infrastructure Spending Through 2050 — With No Mention of Payback
PwC's new global outlook projects cumulative AI infrastructure capital spending will reach $31.6 trillion by 2050, climbing from roughly $800 billion this year to $1.8 trillion annually, with equipment replacement — chips wearing out and needing to be swapped every few years — rising from 70% to 93% of that spend, a pattern PwC says never plateaus the way past infrastructure booms did. The report frames affordable, reliable power as the real bottleneck rather than chip supply, and models a scenario where a disrupted chip supply chain would trim total spending to $25.5 trillion. For Lucien's thesis, what stands out is what's missing: a $31.6 trillion, 24-year forecast from one of the industry's biggest consultancies contains no analysis of expected returns or profitability — the case for the spending rests on scale and inevitability, not on whether it pays for itself.
PwC · Infrastructure & Economics · Sep 2, 2026
At G20, Altman Says AI Adoption Is "Not Negotiable" — But Concedes Countries Can Choose to Build or Rent Their Own Compute
Speaking to G20 economy representatives in Chapel Hill, North Carolina, OpenAI's Sam Altman said using AI is mandatory for countries — "it is not negotiable... you have to use it" — but conceded governments remain free to "decide whether to build or rent data centres" to get there. Nvidia's Jensen Huang pushed a harder line, telling ministers every country needs its own AI infrastructure to support "the digital intellectual capacity" of its economy, while officials including UK Science Minister Chris McDonald pushed back for a more measured pace. For Lucien's thesis, Altman's own framing is the tell: even OpenAI's CEO is now treating "build your own vs. rent from us" as a legitimate, open choice for a country to make, rather than assuming everyone keeps renting frontier capacity indefinitely.
Social News XYZ · On-Prem/On-Device Shift · Sep 2, 2026
The Apple Angle (Standing Perspective)
Today's two stories are both about scale and rhetoric rather than hardware, but the standing argument here is really about what happens once the rhetoric meets a balance sheet. PwC can forecast $31.6 trillion in AI infrastructure spending through 2050 without ever addressing whether it pays for itself, and Sam Altman can tell G20 governments that adopting AI is mandatory while quietly admitting the build-or-rent choice is theirs to make — but somewhere in that fork in the road sits a machine that costs $9,500, not $31.6 trillion, and doesn't require a 24-year capital plan. That machine is Apple's Mac Studio: configured with 512GB of memory shared between CPU and GPU, it can hold a trillion-parameter open-weight model entirely in memory on a single desktop. Reaching that same memory footprint with Nvidia's RTX Pro 6000 workstation cards takes five or six of them, roughly $60,000–$75,000 combined, while pulling something like ten times the power.
As frontier-capable open-weight models keep shipping every few weeks, the argument is that the "build or rent" choice Altman handed back to G20 governments today has a third option neither man mentioned: skip both the multi-billion-dollar sovereign supercomputer and the rented API, and just buy a handful of Mac Studios. 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 design logic, notably after Nvidia stripped NVLink pooling out of its consumer and workstation cards. The fair caveat still holds: a Mac Studio serves one user running one model at a time, while the trillions PwC is forecasting mostly fund infrastructure serving many concurrent users at once — so this looks like Apple leading a specific, high-value segment of AI compute (sovereign-scale capability at desktop-scale cost), not replacing the data center outright.
Limited Edition Jonathan (Substack) · Perspective
Why This Matters
Today's two items both sit on the profitability and infrastructure-politics side of Lucien's thesis rather than the model-capability side. PwC's $31.6 trillion forecast normalizes an AI capex trajectory that only accelerates and never plateaus, while explicitly declining to address whether any of it returns a profit; the G20 ministerial showed the industry's most powerful voices telling governments that using AI is mandatory, while conceding, almost as an aside, that how they get the compute — built or rented — is still an open question. Neither story features a new open-weight model closing the capability gap, but together they sharpen the economic backdrop the thesis depends on: the money committed to the rented, centralized model of AI keeps growing without a profitability case attached, even as the people building that infrastructure acknowledge, in their own words, that owning your own compute is a legitimate alternative — exactly the space the standing Apple Angle argues Apple is quietly best positioned to fill.
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.