Thiago Saragiotto

M&A and Business Strategy Executive | Entrepreneur & Investor | Corporate Lawyer

[Thiago Saragiotto @TSaragiotto]

Valid critique, assuming capex stays contained ignores the upward revisions already happening toward $700B+. The revenue ramp needs to match or the structural thesis gets tested hard. For a balanced allocator view on this exact tension (cyclical risk vs durable build), see this: https://t.co/4Ayai1RORj

[Thiago Saragiotto @TSaragiotto]

The marginal ROI improvement once the base infra is built is a key point most models miss. Utilization ramps and maintenance/growth capex dynamics keep the dollars high even as growth rates peak. This is core to the structural case. Allocator framework on the full cycle: https://t.co/4Ayai1SmGR

[Thiago Saragiotto @TSaragiotto]

Strong thread stressing physical demand > supply and the clear tripwires (real capex cuts, GPU spot decline, etc.). Nothing has broken on fundamentals yet. This pairs perfectly with an allocator’s framework for navigating the cyclical vs structural debate in AI infrastructure spend: https://t.co/4Ayai1RORj

[Thiago Saragiotto @TSaragiotto]

Exactly, renting excess compute is a clever hedge but doesn’t fix the core ROI math on these massive builds. It buys time, not durability. The real edge for allocators is distinguishing which parts of the capex wave are structural infrastructure versus cyclical overbuild. Framework here: https://t.co/4Ayai1RORj

[Thiago Saragiotto @TSaragiotto]

Love the falsifiable tripwire list, capex cuts, sustainable GPU spot decline, HBM rollover etc. That’s how you separate conviction from hope in this environment. Adds real discipline to the allocator question of cyclical pullback vs structural infrastructure. My framework on exactly that distinction: https://t.co/4Ayai1RORj

[Thiago Saragiotto @TSaragiotto]

Memory climbing to 35% of hyperscaler capex in '26 and 48% in '27 is one of those second-derivative shifts that changes everything for allocators. The BOM math is moving fast. For a full framework on modeling cyclical vs structural AI spend (and what it means for long-term capital deployment), this just-published piece is essential: https://t.co/4Ayai1RORj

[Thiago Saragiotto @TSaragiotto]

Solid conversation on hyperscaler capex forecasts and the infrastructure reality check. These numbers are moving fast and the ROI discipline will separate the durable buildout from the parts that get cut. For a structured framework on how capital allocators should navigate cyclical vs structural in this environment, this is directly relevant: https://t.co/4Ayai1SmGR

[Thiago Saragiotto @TSaragiotto]

Overdosing on AI capex with Zuck/Satya earnings looming is the tension point right now. The market is pricing in the inflection, but the real question is how much is structural infrastructure that keeps compounding versus pure cyclical excess. My new piece delivers the allocator framework for exactly this debate: https://t.co/4Ayai1RORj

[Thiago Saragiotto @TSaragiotto]

The robotics/humanoids tailwind on memory after the initial AI data center wave is massively under-appreciated. 300GB+ DRAM per L4 vehicle and 64-128GB per humanoid turns this from cyclical to multi-decade structural demand. Excellent extension of the thesis. For the capital allocation framework tying it all together (cyclical vs structural AI capex), see this: https://t.co/4Ayai1RORj

[Thiago Saragiotto @TSaragiotto]

Classic paradox: Tesla capping employee AI spend at $200/week while ramping infra capex to $25B. When usage has real cost, discipline appears, exactly the shift from hype to engineered ROI that allocators need to watch. This dynamic is central to separating cyclical experimentation from structural infrastructure spend. My latest on the allocator framework: https://t.co/4Ayai1RORj

[Thiago Saragiotto @TSaragiotto]

Even modest hyperscaler capex cuts at Q2 wouldn’t kill the wave, we’re still talking $700B+ this year and supply as the real constraint. A pullback from one player could actually route more demand to neoclouds. Spot on tactical read. For the bigger picture allocator framework on whether this spend is cyclical noise or structural buildout, this is worth your time: https://t.co/4Ayai1SmGR

[Thiago Saragiotto @TSaragiotto]

Memory jumping to 30%+ of hyperscaler capex by YE26 and 40%+ in '27 is the kind of second-order dynamic that separates structural winners from the hype. The initial pushback from clients shows how fast the economics are shifting. For allocators trying to model the full cycle and separate cyclical from durable spend, this framework just published nails the capital allocation lens: https://t.co/4Ayai1RORj

[Thiago Saragiotto @TSaragiotto]

The arms race framing and the capex-to-revenue ratio jump to 0.4+ is the cleanest way to see how far this has moved from normal investment. The historical parallels to previous infrastructure booms are useful, but the payoff function is brutally simple: revenues have to arrive. For a capital allocation framework built specifically around separating the cyclical components from what could be structural in this AI buildout, this piece is directly on point: https://t.co/4Ayai1RORj

[Thiago Saragiotto @TSaragiotto]

Exactly. The first credible capex cut (even modest) from a major hyperscaler will likely be taken as validation by the market and could trigger a rapid derating across the entire AI infra complex. That’s the classic late-cycle signal. For a practical allocator framework on how to position around this inflection (cyclical drawdown vs structural continuation) this just came out: https://t.co/4Ayai1RORj

[Thiago Saragiotto @TSaragiotto]

Sharp point on the subsidy loop keeping hyperscalers in FOMO mode. The moment AI labs can no longer paper over unit economics with cheap capital or growth-at-all-costs, the demand signal gets tested for real. That’s when we’ll see who built structural capacity versus cyclical inventory. For a framework on exactly this cyclical vs structural tension in AI spend, check this: https://t.co/4Ayai1RORj

[Thiago Saragiotto @TSaragiotto]

This 'heads I win, tails I win' dynamic on AI capex is fascinating and dangerous. It creates reflexive narratives that can mask the real work of figuring out which spend actually compounds. The structural case only holds if real revenue and ROI follow the buildout at scale. For a clearer framework on how allocators should think about this distinction, I just published this: https://t.co/4Ayai1RORj

[Thiago Saragiotto @TSaragiotto]

Excellent breakdown on why the hyperscaler capex paths aren’t monolithic. Amazon’s internal automation + robotics angle and Google’s defensive moat play are structurally different from the others still proving the model. The self-reinforcing loop you describe is precisely why some of this spend may be more durable than pure hype. For a capital allocator’s framework on cyclical vs structural AI capex, worth reading this: https://t.co/4Ayai1RORj

[Thiago Saragiotto @TSaragiotto]

Spot on with the Bloomberg ROI threshold. The 1.7-1.8x revenue/D&A breakeven is exactly the kind of concrete benchmark allocators need right now. If growth holds and capex doesn’t explode further, we’re looking at a structural shift rather than pure cyclical spend. For a practical framework on separating the cyclical noise from durable AI infrastructure allocation, this piece just dropped and pairs really well: https://t.co/4Ayai1RORj

[Thiago Saragiotto @TSaragiotto]

The current AI capex boom is often treated as a structural shift. In reality, different layers of the stack have very different risk and durability profiles.

Here’s a framework to separate cyclical from structural exposure when allocating capital.

The AI Capex Boom: Cyclical or Structural? A Framework for Capital Allocators https://t.co/wTuzaLBmJu

[Thiago Saragiotto @TSaragiotto]

@roderix1966 Eu tirei os 4 com anestesia geral! No hospital, rs. Há uns 35 anos