Build Out, Not Bailout: What OpenAI’s “$1.4 Trillion,” Jensen Huang’s China Warning, and Zohran Mamdani’s Win in New York Really Tell Us About 2025
Build Out, Not Bailout: What OpenAI’s “$1.4 Trillion,” Jensen Huang’s China Warning, and Zohran Mamdani’s Win in New York Really Tell Us About 2025
I write this as someone who watches technology, markets and politics not just as headlines, but as inputs to capital allocation, risk management and, frankly, national strategy. From Singapore I have a slightly detached and neutral vantage point: I see the U.S. trying to sprint into an AI century while tripping over its own politics, I see China moving in a straight line, and I see global investors trying to work out whether this is a genuine super-cycle or just the 2025 version of dot-com. The All-In episode that sparked this essay packed all of that into one conversation: OpenAI’s “rough week,” Jensen Huang’s deliberately provocative “China is going to win the AI race,” an unmistakable risk-off turn in U.S. markets, and then—because 2025 America can’t help itself—the rise of a self-described democratic socialist, Zohran Mamdani, to be New York City’s mayor. (People.com)
2025’s AI drama was never about OpenAI being “insolvent.” It was about scale, timing, and politics. A perfectly fair question—how do you match US$13–20 billion of fast-growing revenue to US$1.4 trillion of data-centre and power commitments—got clipped into a viral moment and briefly spooked Microsoft, Nvidia, Oracle and the rest. When I strip out the noise, the story is straightforward: OpenAI expects revenue to keep compounding, a large part of the US$1.4 trillion will be borne by partners, and what the company wants from Washington is a smoother, cheaper path to build compute and power, not a handout. That is a national-security and economic-growth agenda, not corporate welfare.
Running underneath this is a deeper strategic worry. Jensen Huang is right to warn that China, with centralised approvals and aggressive nuclear build-out, can deploy AI infrastructure faster than a United States trapped in 50 slightly different state AI laws. If America wants to stay ahead, Congress has to pre-empt the worst of the patchwork and set one innovation-friendly federal framework.
At the same time, AI is colliding with affordability anxiety. Layoffs have hit a 20-year October high, delinquencies are rising, and yet AI-exposed stocks are still up. That dislocation is the backdrop to Zohran Mamdani’s surprise New York win on rent freezes, free transit and taxing the rich. When younger voters are carrying student debt and can’t afford housing, they will back candidates who promise to rebalance the system.
So my takeaway from this messy week is simple: finance the AI super-cycle privately, clear federal obstacles quickly, and deliver visible domestic wins before economic frustration gets channelled into harder-left urban politics.
1. What Actually Happened in OpenAI’s “Rough Week”
The viral moment began with what was, in truth, a very normal question from Brad Gerstner: how do you square a company doing about US$13 billion in revenue with publicly floated capex and partnership commitments totalling US$1.4 trillion? That is not a “gotcha”; it’s how every serious analyst would approach a capital-intensive growth story. I wrote down this here at a previous post, originally from BG2 podcast. https://zionzhao.blogspot.com/2025/11/building-intelligence-at-industrial.html
Sam Altman’s slightly feisty response (“if you want to sell your shares, I’ll find you a buyer”) got clipped, memed and then read as defensiveness. The substance was lost. But Altman and OpenAI quickly clarified through other channels what the run-rate actually is and what they really want from Washington.
Here is the fact pattern once we strip the drama out:
Revenue: multiple outlets reported the “US$13 billion” number, but Altman later said OpenAI will end 2025 on a US$20 billion annualised run rate—that’s ~US$1.66 billion for December alone, which is consistent with very steep cohort curves on ChatGPT and rapid API growth. (Reuters, 2025; Bloomberg, 2025) (Reuters)
The US$1.4 trillion: later reporting made it clear this is an eight-year capital plan covering data centres, AI servers, power and grid components—and, crucially, that not all of it sits on OpenAI’s balance sheet. Partners—Microsoft, Oracle, data-centre financiers—share a very large piece of that bill. The All-In crew guessed “about half,” which is a reasonable working assumption given how hyperscale build-outs are usually syndicated. (Bloomberg, 2025; Reuters, 2025) (Bloomberg)
What the CFO meant: Sarah Friar’s now-famous use of the word “backstop” with the Wall Street Journal set off the “is OpenAI insolvent?” narrative. But 24 hours later both she and Altman clarified: they are not asking for a taxpayer bailout; they are asking the U.S. to lower the cost of capital for national-scale compute by doing things like extending the CHIPS Act tax credit to AI servers and data centres, and by speeding permits so that AI firms can build power “behind the meter.” That is industrial policy, not corporate welfare. (Reuters, 2025; TechCrunch, 2025) (Reuters)
The administration’s line: the Trump administration (we are in November 2025 in this scenario) has been explicit that there will be no federal bailout or blanket guarantee for AI companies—partly because the sector is so competitive (OpenAI, Anthropic, Google Gemini, xAI, Meta, Groq) that letting a single over-levered player fail would not impair national capabilities. (Reuters, 2025; Yahoo Finance, 2025) (Yahoo Finance)
So, does OpenAI need a bailout? On the facts available: no. They need policy enablement so that a privately financed, US$1-plus-trillion AI build-out does not get choked by high rates, slow permits and fragmented state rules. That is the line David Sacks summarised in four words: “build-out, not bailout.”
2. The Math Behind “US$1.4 Trillion” (and Why Markets Still Panicked)
Markets sold off anyway—Microsoft, Nvidia, Oracle, Broadcom, all the key AI exposure names were down 6–20% in the days around these headlines. That overreaction is still rational for three reasons:
We are in a capex-digestion phase. For 18 months investors have modelled AI hyperscale growth as almost linear. When they suddenly see numbers like “US$1.4 trillion” or hear that Meta is at US$600 billion of planned U.S. spend and Oracle has tied US$18 billion in lending to data-centre projects, they go straight to the question: can end-demand actually pay this back? (Reuters, 2025; Bloomberg, 2025) (Reuters)
ROI is hard to model when pricing is collapsing. If Google or Apple makes baseline AI assistants free, that compresses OpenAI’s consumer ARPU—exactly the risk Jason Calacanis raised. Altman is implicitly betting that usage will explode even faster than prices compress, thus preserving revenue. That is plausible in a super-cycle, but it’s not a free lunch.
We just got a macro shock. On 6 Nov 2025, Challenger, Gray & Christmas reported 153,074 announced U.S. job cuts for October—the highest October since 2003, up 175% YoY—and “AI” was specifically cited in more than 31,000 of those cuts. That, plus a 36-day government shutdown, made every large number sound scarier. (Challenger, Gray & Christmas, 2025; Reuters, 2025) (Challenger Gray Christmas)
But even then, the All-In crew’s back-of-the-envelope was directionally right. If OpenAI ends 2025 at a US$20 billion run rate and believes it can get to “hundreds of billions” by 2030, and Nvidia’s Jensen Huang is simultaneously telling markets to expect US$3–4 trillion of AI infrastructure spend by decade-end—ten times the Manhattan Project, but this time mostly private—then OpenAI’s slice (US$1.4 trillion over eight years, say ~US$175 billion per year) no longer looks completely fantastical, especially if you haircut it for partners and scheduling. (Huang, 2025; TechCrunch, 2025) (TechCrunch)
Economically, this is classic “patient capital for general-purpose technology”—the same pattern Mariana Mazzucato (2013) described for the internet and GPS, except this time the private sector is funding first and asking the state to remove frictions, not to write the cheque.
3. Industrial Policy, Not Corporate Welfare
Where I agree most strongly with Sacks and Gerstner is on the shape of the state’s role.
What OpenAI actually asked for in its letter to the White House was to extend the 35% Advanced Manufacturing Investment Credit (AMIC) created under the CHIPS Act to cover AI servers, data centres and even grid components—because at this scale, power is the bottleneck. That is a horizontal incentive open to all AI players, not a bespoke rescue. (Reuters, 2025; Bloomberg Tax, 2025) (Reuters)
What the administration said back is equally sensible: we will help you build faster—through permitting reform, behind-the-meter generation, and maybe through export-linked reinvestment of tariff revenues in small modular nuclear—but we will not socialise your balance sheet risk. (Reuters, 2025; Al Jazeera, 2025) (Yahoo Finance)
This is exactly how modern industrial policy is supposed to work: the state derisks infrastructure and regulation; the private sector takes market risk. Dani Rodrik has argued for years that the challenge is not “picking winners” but “crowding in” private investment in sectors with high social spillovers (Rodrik, 2004). AI fits that description.
So when the internet yelled “bailout,” it was mis-labelling a fairly orthodox public–private partnership.
4. Jensen Huang’s Warning: Why Federal Preemption on AI Matters
The next move in the episode was an elegant segue: Brad Gerstner mentions the federal role, Sacks picks it up, and then they bring in Jensen Huang’s Financial Times line that ‘China is going to win the AI race’ because the U.S. is running with one hand tied behind its back. That sounds dramatic, but in November 2025 that was literally the public line out of Nvidia—then hastily restated to say “China is only nanoseconds behind; America must win by racing ahead.” (IDN Financials)
Here is the core issue:
In 2025 the U.S. still does not have a single, comprehensive federal AI act. Instead it has White House executive orders, NIST frameworks, sectoral rules—and a fast-growing patchwork of state bills covering safety, watermarking, algorithmic discrimination, even reporting obligations for “model incidents.” White & Case’s 2025 tracker called this out explicitly. (White & Case)
Republicans in Congress, notably Senator Ted Cruz, responded with a proposed 10-year moratorium on state and local AI laws—later watered down to five years—precisely to avoid “Californication” of AI. (DLA Piper, 2025; Reuters, 2025) (DLA Piper)
Meanwhile China is centralising approvals, subsidising power, and—per Huang—already has 100 nuclear fission plants under construction, which, if even half correct, means its AI companies won’t have to go begging state commissions for grid hookups. That asymmetry is real. (IDN Financials)
My own view is the same as the one Sacks articulated: AI is clearly interstate commerce; Congress should preempt the worst of the state patchwork and set a single, innovation-forward federal floor. That does not have to mean gutting state authority on consumer protection or child safety—only that we don’t make developers report the same safety incident to 50 agencies on 50 timelines. That is exactly the “nationwide strategy or Californication” dilemma a House Judiciary memo warned about in September 2025. (BGR Group)
Do that, and Huang’s warning stops being prophetic.
5. Markets, Layoffs and the “We Need Domestic Wins” Argument
One of the sharpest turns in the conversation was away from AI and straight into the U.S. macro mood: the consumer is cracking, credit-card and student-loan delinquencies are rising again, and yet the S&P is still up ~14% YTD because seven AI-exposed names pulled the whole index. That decoupling is now visible in data:
The New York Fed’s 2025 Q3 Household Debt and Credit Report shows total debt at a record US$18.59 trillion, with transitions into serious delinquency up to 3.03% from 1.68% a year earlier, and student-loan delinquencies jumping as pandemic-era forbearance finally hit credit files. Young borrowers were specifically flagged as weakening. (New York Fed, 2025) (Federal Reserve Bank of New York)
At the same time October layoffs were the highest for that month since 2003, and both Reuters and Bloomberg noted that companies directly cited “AI” and “cost-cutting” together. That tells me AI is now a public labour-market story, not just a tech-sector one. (Reuters, 2025; Bloomberg, 2025; Challenger, Gray & Christmas, 2025) (Reuters)
This is why you heard the line “we need visible domestic / ‘main street’ wins”: when equity holders, crypto holders and AI startup founders are up 20–30% and 20- to 24-year-olds are at 9% unemployment, you create political space for someone to run on rent freezes, free buses and higher taxes on the rich—and win New York City. That is exactly what happened. (People.com)
Economically, this fits what Acemoglu and Restrepo have called the “task displacement vs. task creation” problem in automation: short-run displacements often hit younger, lower-tenure workers first, long before the new higher-productivity tasks materialise (Acemoglu & Restrepo, 2019). When that happens and you already have a debt overhang from student loans, it is rational—if you are 28 in Queens—to vote for a candidate promising to socialise some of those costs.
6. Zohran Mamdani’s Win and the “Broken Generational Compact”
The People/ABC/Reuters write-ups on Zohran all say the same three things: he is 34; he is the first Muslim and first South-Asian mayor of NYC; and he ran on affordability—rent freezes, free public transit, universal childcare, even city-run groceries. He beat Andrew Cuomo 50.4% to 41.6%, with Curtis Sliwa a distant third. Turnout was the highest in decades because young people registered in droves. (People.com)
Is that “socialism rising”? Partly. But I read it more like Peter Thiel’s 2020 diagnosis, which Chamath paraphrased on the show: if housing is unaffordable and student debt is unpayable, people will eventually turn against capitalism because they have no capital to defend. That is not ideological, it is balance-sheet logic.
Three policy levers could defuse this without burning down the system:
Stop federally underwriting low-ROI degrees and force market pricing of education credit—exactly what Sacks and Chamath said. That is also what many labour economists have argued: federal aid without price discipline leads to tuition inflation and misallocation (Cellini & Goldin, 2014).
Targeted, one-time student-loan relief tied to structural reform. The mistake of the 2020s was trying to forgive trillions without changing the lending machine. Do it the other way round: fix the machine, then clean up the legacy stock. Even Chamath—a self-described capitalist—said 2025 was the first time he became sympathetic to forgiveness because we should never have let the loans be written that way in the first place.
Local housing supply reform. Here, New York and California are guilty as charged; Singapore is actually the better model. You cannot run 4–5% wage inflation, 3% CPI and 0–1% housing supply growth and expect young voters to stay centrist.
Unless the federal government can get out of its own shutdown and actually legislate, Mamdani’s win becomes the template for other blue cities. That is what Sacks was getting at when he warned that if Republicans don’t deliver, “these socialists are going to take over in three years.” That is an outcome analysis, not voter targeting.
7. The Filibuster “Nuclear Option” in a 36-Day Shutdown
None of the above can happen if Congress is closed.
By 5 Nov 2025 the U.S. was on day 36 of the longest shutdown in its history, and President Trump openly called on Senate Republicans to end the filibuster so they could reopen government with 51 votes. Senate GOP leaders, led by John Thune, declined, warning—as always—that Democrats would use the same tool against them later. (Reuters, 2025; AP, 2025; PBS, 2025) (Reuters)
The mechanics here are well-described in Congressional Research Service papers on cloture and the so-called nuclear option: the filibuster is not in the Constitution; it is a Senate rule that can itself be overturned by a simple majority precedent. It was already “nuked” in 2013 for most nominations and in 2017 for Supreme Court nominees. Extending that to legislation would not be technically hard. (CRS, 2024–2025) (Congress.gov)
Why does this matter for AI, housing and student loans? Because a paralysed Congress defaults power to exactly the state-level patchwork that Huang, Sacks and the House Judiciary AI memo all warned about. If California, New York, Colorado and Illinois write the rules for AI safety and algorithmic discrimination, that will be the de facto national standard because that is where the companies are. Federal preemption is the only way to stop that. (BGR Group)
So the filibuster debate in late 2025 is not some procedural nerd fight; it is about whether the U.S. can pass the domestic affordability agenda fast enough to blunt the Mamdani-style appeal in 2026 and 2028.
8. Where This Leaves Investors, Builders and Policymakers
Putting it all together:
AI is a real, privately financed industrial revolution. Nvidia is still guiding to US$3–4 trillion in infra spend by 2030; OpenAI alone has committed to US$1.4 trillion over eight years; Meta is at US$600 billion; Oracle just lined up US$18 billion in DC lending. That is not a bubble chart, it is a capex table. (Reuters, 2025; TechCrunch, 2025; Business Times, 2025) (Reuters)
The U.S. government’s correct role is to smooth the path—permitting, power, tax credits—not to rescue over-ambitious AI firms. That is exactly what the administration said this week. (Reuters, 2025) (Yahoo Finance)
Markets will stay choppy until they can see monetisation that matches the capex. That means watching whether OpenAI’s consumer subs can survive a wave of “free” offers from Apple and Google, and whether enterprises stay with OpenAI’s API or migrate to Anthropic and open-source to avoid being competed with—both headwinds rightly raised on the show.
- Politically, the AI boom is happening at the exact moment voters are looking for affordability. That is why a democratic socialist can win NYC on rent freezes, and it is why Republicans are being told—by their own supporters—to “get back to domestic wins.” AI companies ignore that at their peril.
- Internationally, this is good news. A U.S. that says “build out, not bailout” is signalling to allies—Singapore, Japan, the Gulf—that they can co-invest in AI infrastructure without getting subordinated to U.S. rescue politics.
Singapore and the wider region are entering an AI-accelerated, policy-driven decade.
That is why you need a real estate advisor who reads the same macro signals you do. Every day I invest hours studying global geopolitics, monetary policy, tech capital expenditure and Singapore statutes, then translate that into clear, investable property strategies for international families, China Chinese investors, ASEAN buyers and UHNW or institutional mandates. I do the due diligence so your capital is protected.
Let me help you add Singapore property as the stabiliser in your portfolio – lower volatility than equities or crypto, yet strong capital appreciation potential and rental yields that feel like dividend income. If you are relocating, investing for your children’s education, or building a family office base here, speak with me first. I stay ahead of AI, regulation and the market so your real estate decisions stay ahead too with clarity and confidence today.
9. Disclaimers
This essay is for educational and analytical purposes only.
It does not solicit investment, does not give financial, legal or tax advice, and does not instruct any individual or demographic group how to vote or whom to support.
All facts are taken from publicly available, reputable sources dated up to 10 Nov 2025 and cited below; forward-looking statements are my own interpretation.
References (APA)
Acemoglu, D., & Restrepo, P. (2019). Automation and new tasks: How technology displaces and reinstates labor.Journal of Economic Perspectives, 33(2), 3–30.
ABC News. (2025, November 6). How Zohran Mamdani’s triumph in New York is making waves in U.S. politics. (ABC)
Bloomberg. (2025, November 7). OpenAI asks U.S. to expand Chips Act tax credit to AI data centers. (Bloomberg)
Business Times. (2025, November 7). OpenAI asks US to expand chips tax credit to AI data centres. (The Business Times)
Challenger, Gray & Christmas. (2025, November 6). October Challenger report: 153,074 job cuts on cost-cutting, AI.(Challenger Gray Christmas)
DLA Piper. (2025, May 22). Ten-year moratorium on AI regulation proposed in US. (DLA Piper)
Financial Times. (2025, November 5). Nvidia’s Jensen Huang says China “will win” AI race with US. (Financial Times)
Global Times. (2025, November 7). Nvidia responds to “leaked” remarks on social media. (Global Times)
Mazzucato, M. (2013). The entrepreneurial state: Debunking public vs. private sector myths. London: Anthem Press.
New York Federal Reserve. (2025, November 5). Quarterly report on household debt and credit, 2025 Q3. (Federal Reserve Bank of New York)
People Magazine. (2025, November 5). Zohran Mamdani, 34, defeats Andrew Cuomo to become NYC’s first Muslim mayor. (People.com)
Reuters. (2025, November 6). OpenAI does not want government guarantees for massive AI data-center buildout – CEO.(Reuters)
Reuters. (2025, November 5). Trump calls for end of filibuster as longest-ever shutdown drags on. (Reuters)
Reuters. (2025, November 6). U.S. layoffs for October surge to two-decade high, Challenger data shows. (Reuters)
TechCrunch. (2025, November 8). OpenAI asked Trump administration to expand Chips Act tax credit to cover data centers. (TechCrunch)
White & Case. (2025, September 24). AI watch: Global regulatory tracker – United States. (White & Case)
(Additional classic works on filibuster procedure: Congressional Research Service. (2025). Points of order, rulings and appeals in the Senate.; PBS. (2025). What is the filibuster and why does Trump want to get rid of it? (Congress.gov))
Comments
Post a Comment