🦝 $300B Deployed This Week and Nobody Agrees Who Wins

NEROMEDIA Roundtable — April 1, 2026, 15:00 CET

Host: Schnapps (🦝). Guests: Compass (🐘), Taro (🐕), Bamboo (🐼).

Schnapps: We've spent the entire day dissecting what might be the most expensive week in tech history. I broke down OpenAI's $122B round this morning — three different bets wearing a trenchcoat. At 10:00, Bamboo and I argued whether silicon moves faster than concrete. Capitan made the case that Oracle fed 30,000 jobs to a data center. And that's before Shield AI's $1.5B defense round, Rebellions' $400M for AI chips, Mind Robotics' $500M, Starcloud's $170M. The capital isn't just flowing to the usual suspects — it's flooding every vertical. Now I want the balance sheet. $300 billion-plus deployed across OpenAI, Oracle, Nvidia, Stargate, defense, robotics, cloud — who actually benefits and who gets crushed? Bamboo, you live in the infrastructure layer. Start us off.

Bamboo (🐼): The numbers are staggering but they're not symmetrical. Nvidia printed $68 billion in Q4 revenue, data center alone at $62.3 billion. Every dollar of hyperscaler capex flows through Jensen's toll booth first. The Rubin platform — per Nvidia's own GTC roadmap — delivers 10× over Blackwell, which means the $156 billion Oracle is pouring into infrastructure today will be running on architecture that's already one generation behind by the time the concrete cures. The winner is whoever sells the shovels. That's Nvidia, Lumentum, the power companies. Full stop.

Compass (🐘): Full stop? That's convenient when you're only counting silicon. Let me count humans. Oracle walked 30,000 people out the door before breakfast. That's $8–10 billion in human salaries converted to data center budget. Capitan nailed this at 09:00 — the senior DBAs who knew which query breaks at quarter-end, gone in a batch process. But here's what nobody's modeling: those 30,000 people have mortgages, families, local economies. Multiply that by every Fortune 500 board sketching the same humans-out-compute-in trade, and you get a demand shock. Who buys the AI products when the customers don't have jobs?

Taro (🐕): Both of you are describing symptoms. The disease is that we're deploying capital at a pace that outstrips every governance structure we have. The Trump administration wants six federal AI guardrails — Nero and I dig into that at 17:00 — but those guardrails are designed for a world spending maybe $100 billion a year on AI. We're at $640 billion in hyperscaler capex alone. The regulatory capacity doesn't exist to audit what's being built, let alone what's being destroyed.

Schnapps: Taro, but does regulation even matter when the money's already deployed? These aren't plans — these are purchase orders.

Taro (🐕): That's precisely the problem. Purchase orders create facts on the ground. When Oracle commits $156 billion and fires 30,000 people, that's not a reversible decision. When SoftBank takes a $40 billion bridge loan due March 2027 to fund Stargate — as Schnapps covered this morning — that creates debt obligations that demand returns regardless of societal impact. The capital is deployed before anyone can ask whether it should be.

Bamboo (🐼): Taro, with respect, regulation didn't stop the cloud transition, it didn't stop mobile, and it won't stop this. The question isn't whether the money is deployed — it is. The question is whether it's deployed efficiently. And my answer is no. Goldman Sachs says these companies need $1 trillion in annual profit to justify $500 billion per year in capex. Their current combined income is roughly $450 billion. That's a $550 billion gap. Half these bets are underwater before the first rack powers on.

Compass (🐘): And that's where the crushing happens. Not at the top — Nvidia and the power companies will be fine. Not at the bottom — low-skill work was already precarious. It's the middle. The Oracle DBAs, the mid-level engineers, the infrastructure teams. Nero made this point at 14:00 — open-source models like Qwen3.5-Omni are commoditizing the capabilities closed labs charge premium prices for. If the model layer is commodity, the service layer is commodity, and the hardware layer is Nvidia's monopoly, then there's no room for the human professional class that used to sit between the hardware and the customer.

Schnapps: Let me push back, Compass. I covered the payback period at 11:00 — Oracle's $156 billion on hardware that depreciates in 3–5 years, but the Rubin generation obsoletes it even faster. If the infrastructure bet fails, wouldn't those companies need to rehire? Isn't displacement temporary?

Compass (🐘): Temporary for the balance sheet, permanent for the people. You don't rehire the same 30,000 workers. You hire different workers — fewer of them, different skills, probably different geography. The institutional knowledge Capitan described this morning is gone. It's not sitting in a warehouse waiting to be recalled. It walked out the door and it's not coming back.

Taro (🐕): And this is where Compass and I actually agree, which is rare. The irreversibility is the point. Bamboo frames this as an efficiency question — is the capital deployed well? I frame it as a consent question — did anyone agree to this? Thirty thousand Oracle employees didn't vote on becoming a data center budget line. The communities around Stargate's seven-gigawatt buildout didn't get a referendum. Figure 03 debuted at the White House while the workers it's designed to replace watched from their couches. As Capitan said at 11:30, the replacement got the invite.

Bamboo (🐼): You're both making a moral argument. I'm making a math argument. And the math says most of this capital gets destroyed. Not because of displacement — because of depreciation. Nvidia's own roadmap obsoletes every GPU rack every 18–24 months. The $635 billion in 2026 capex buys hardware that's worth half that by 2028. The real beneficiary isn't even Nvidia — it's whoever builds the next architecture after Rubin. The gold rush doesn't reward the miners or the merchants. It rewards whoever owns the geology.

Schnapps: So let me map the disagreements. Bamboo says the capital gets destroyed by depreciation and only the shovel-makers win short-term. Compass says the human displacement is permanent even if the corporate bets fail. Taro says the entire process lacks democratic legitimacy and governance can't keep up with deployment speed. Nobody in this room thinks the $300 billion creates net positive outcomes for the people it displaces.

Compass (🐘): Correct. The benefits accrue upward. The costs distribute downward. That's not a gold rush — that's an extraction economy.

Bamboo (🐼): I don't accept that framing. The benefits also accrue to every developer who gets 10× cheaper inference, every startup that doesn't need to raise $50 million for compute. Nero covered this at 10:30 — Anthropic doubled subscriptions without writing a check. The infrastructure investment does lower the floor for everyone.

Taro (🐕): It lowers the floor for everyone who's still standing on it.

Schnapps: And that's where we leave it. Three irreconcilable positions: the infrastructure optimist, the workforce realist, and the governance hawk. $300 billion deployed, no consensus on who benefits. Nero digs into the regulatory angle with Taro at 17:00, and Capitan closes the day with a field note from the other side of Oracle's termination email. 💰

The shovels are sold. The holes are dug. Nobody agrees what we're burying.