E131: 2024 Fantasy President picks, debt ceiling agreement, Dollar dominance & more

In the 2024 U.S. presidential election, the Republican Party officially nominated former President Donald Trump, a real-estate developer and media figure, not a bank CEO. (en.wikipedia.org) The Democratic Party ultimately nominated Vice President Kamala Harris, a career prosecutor and elected official, also not a bank CEO. (en.wikipedia.org) Meanwhile, Jamie Dimon remained CEO of JPMorgan Chase and repeatedly stated he had no plans to run for office, later explaining he chose not to run for president. (cnbc.com) Since neither major party nominated the sitting CEO of a major U.S. bank, Sacks’s prediction that you would “never get the head of the number one bank in the US to be their candidate” was borne out by events.

The prediction concerns China’s population size in the year 2300. As of today (2025), that date is still 275 years in the future, so the specific claim that China’s population will be “roughly half” of its current size by 2300 cannot yet be verified or falsified.

We can say that the direction of Chamath’s underlying premise (population decline) aligns with current data:

• China’s total population peaked around 2021 and has since begun to fall; official figures show population declines in both 2022 and 2023, driven by low birth rates and an aging population.
• China’s total fertility rate (TFR) is now well below the replacement level of 2.1 children per woman (recent estimates place it around ~1.0 or even lower in some analyses), suggesting continued demographic pressure and likely long‑term shrinkage if trends persist.

However, long-range demographic projections out to 2300 are highly uncertain and depend on policy changes, migration, economic development, technological progress, and cultural shifts over centuries. Current UN and academic projections often extend only to 2100 and already diverge significantly from one another; extrapolating another 200 years beyond that is speculative.

Because the prediction’s target year has not arrived and no model can definitively confirm a specific 2300 population level today, the correct classification is:

• Result: inconclusive (too early to tell).

The prediction is explicitly framed as a multi‑decade process (“beginning of the end… not an immediate collapse”), so evaluating it about 2.5 years after June 2023 is inherently premature.

As of late 2025:

• The U.S. dollar remains by far the leading reserve currency, accounting for roughly 58–59% of disclosed global FX reserves, according to IMF COFER data, with the euro a distant second and others (CNY, JPY, GBP, etc.) far behind.
• The dollar also continues to dominate FX trading and trade invoicing. BIS and academic studies show most trade—even between non‑U.S. partners—is still priced in USD, including a majority of global commodity trade (especially oil), despite some high‑profile bilateral deals in local currencies.
• While China, Russia, some Gulf states, and others have expanded local‑currency or non‑dollar settlement arrangements since 2022–2023, those flows are still a small fraction of global trade and have not yet produced a decisive structural shift away from the dollar.

In other words, some supporting trends exist (more local‑currency trade, de‑risking from USD exposure, etc.), but the core of the claim is about the dollar losing its singular reserve‑currency role over decades. Given how dominant the dollar still is and how little time has passed relative to the forecast horizon, the truth or falsity of that long‑run prediction cannot yet be determined. Therefore, the proper classification is “inconclusive (too early)”.

As of November 30, 2025, the US dollar is still the dominant global reserve currency by a large margin. IMF and central bank data show that roughly 55–60% of disclosed foreign exchange reserves are held in USD, far ahead of the euro, yen, pound, or yuan. No other currency has clearly begun to displace the dollar as the primary global reserve, and there has been no formal or de facto transition to a different main reserve currency.

However, Jason’s prediction is explicitly about the remainder of his and Friedberg’s lifetimes—i.e., “several decades” into the future. That time period has obviously not elapsed by 2025, and the claim concerns the persistence of a trend, not a specific event by a fixed date. Since we are still well within the window of the prediction, and the condition to judge it definitively (end of their lifetimes / several decades passing) has not been met, the correct label today is that it is too early to tell whether the prediction will ultimately be right or wrong.

The prediction is explicitly about long‑term demographic effects on China’s economic strength over “the next 10 or 15 years” from 2023, i.e. roughly 2033–2038, not 2024–2025. As of today (Nov 30, 2025), we are only ~2.5 years into that 10–15 year window.

What we can already see:

• China’s population has begun to shrink, with official data showing its first population decline in decades in 2022 and further decline in 2023, and a rapidly aging population.
• Numerous demographic and economic analyses (UN, World Bank, IMF, various demographers) project continued aging, low fertility, and a shrinking workforce, all of which could weigh on growth over the next few decades.

However, Chamath’s claim is not just that demographics are a headwind (which is already well established), but that they will “diminish them economically” in a strongly exacerbated way over the next 10–15 years. Whether that level of economic diminishment occurs, relative to other countries and relative to plausible counterfactuals, cannot yet be judged in 2025.

Because the forecast period runs out to at least 2033 and up to 2038, and no part of the prediction has been clearly falsified (nor fully confirmed) this early, the correct status is inconclusive (too early).

The prediction is explicitly about a 50–100 year time horizon starting from around 2023 for India to potentially reach US-like scale and predictability as an economic/investment destination. As of today (November 30, 2025), only about 2 years have elapsed since the prediction was made, which is far too short to assess whether India will or will not achieve that outcome over several decades.

Because the claim concerns structural economic development over half a century or more, there is no way to definitively judge it yet. Therefore, the correct evaluation is that it is too early to tell whether the prediction will prove accurate.

By the end of the two‑quarter window (roughly through Q4 2023), the first part of Chamath’s scenario did happen, but the core outcome he was betting on did not.

1. Did AMD, Meta, Google, Microsoft, and Amazon announce substantive AI hardware/infra in that window?
Yes:

• AMD detailed its Instinct MI300 series (MI300X/MI300A) as data‑center AI accelerators for generative‑AI training and inference at its June 13, 2023 "Data Center and AI Technology Premiere," and formally launched them at the Dec. 6, 2023 "Advancing AI" event with major partners like Microsoft, Meta, Oracle, Dell, HPE, and others adopting them. (amd.com)
• Meta publicly described its in‑house Meta Training and Inference Accelerator (MTIA) chip family and a next‑gen, AI‑optimized data‑center design in May 2023, explicitly as custom AI accelerators and infrastructure for its workloads. (about.fb.com)
• Google announced Cloud TPU v5e, a purpose‑built AI accelerator for both training and inference, at Google Cloud Next ’23 on Aug. 29, 2023, describing it as its most cost‑efficient, scalable TPU to date, and followed up with the higher‑end TPU v5p in Dec. 2023. (cloud.google.com)
• Microsoft unveiled its own Azure Maia 100 AI accelerator (and Cobalt 100 CPU) at Microsoft Ignite in mid‑November 2023, explicitly aiming to reduce dependence on third‑party GPUs for Azure AI workloads. (arstechnica.com)
• Amazon/AWS used re:Invent 2023 (late Nov. 2023) to announce Trainium2 (second‑gen AI training chip) and Graviton4, positioning Trainium2 as a high‑performance, lower‑cost alternative for training large foundation models and LLMs, and specifically framing these chips as part of a strategy to reduce reliance on Nvidia GPUs. (infoworld.com)

On the narrow question of announcements, his expectation was accurate: all five players did roll out or materially advance substantive AI hardware/infrastructure efforts within ~two quarters of Q2 2023.

2. Did this create “meaningful vendor diversity” so forecasts were no longer effectively 100% Nvidia?
This is where his bet fails. Despite those announcements, Nvidia’s grip on AI data‑center compute remained overwhelming, and market forecasts continued to treat Nvidia as the overwhelmingly dominant provider:

• TechInsights data (summarized in DCD) shows Nvidia had 98% of data‑center GPU shipments in 2023, underscoring that essentially all deployed accelerator GPUs that year were still Nvidia’s. (datacenterdynamics.com)
• SiliconAnalysts likewise estimates Nvidia had about 98% of data‑center GPU revenue in 2023, and even forward‑looking projections still give Nvidia ~87% of AI accelerator revenues in 2027, i.e., forecasts remain overwhelmingly Nvidia‑centric despite AMD, custom TPUs, and cloud chips. (siliconanalysts.com)
• A mid‑2024 CNBC overview notes Nvidia’s AI accelerators still controlled 70–95% of the AI chip market, even as AMD and custom cloud chips were ramping, indicating only modest erosion of Nvidia’s dominance. (cnbc.com)
• Benzinga/TechInsights reporting in 2024 puts Nvidia at 92% of the data‑center GPU market, with Meta simultaneously planning massive deployments of Nvidia H100‑class GPUs, illustrating that even large hyperscalers that are designing custom chips still rely primarily on Nvidia for training capacity. (benzinga.com)
• Analysts and market research in 2024–2025 consistently describe Nvidia as holding roughly 80–90%+ of the AI accelerator/GPU market for data centers, with AMD and custom cloud ASICs making up only a small single‑digit or low‑teens share. (barrons.com)

In other words, the announcements did not, within the predicted timeframe, translate into AI compute forecasts where Nvidia was just one of several roughly comparable vendors. Instead, those forecasts continued to treat Nvidia as the near‑monopoly, with others as emerging but still marginal alternatives.

Since Chamath’s bet was not just that announcements would occur, but that these would prevent Nvidia from “running away with this” and would force forecasts away from being effectively 100% Nvidia‑centric, and the latter clearly did not happen, the overall prediction must be scored as wrong.

Evidence since mid‑2023 strongly supports Friedberg’s prediction that AI compute and application demand would go nonlinear as models improved and effective costs fell:

• Explosive growth in AI/GenAI spending: Gartner forecasts worldwide GenAI spending to jump from about $365B in 2024 to $644B in 2025, a 76% YoY increase after an estimated 337% surge in 2024, explicitly attributing this to better foundational models and rising demand for AI products. (gartner.com) Gartner also projects overall AI-related spending reaching ~$1.5T in 2025 and surpassing $2T in 2026, indicating a rapidly accelerating curve rather than saturation. (gartner.com)
• Server and data‑center build‑out is described as exponential/nonlinear: Gartner notes that GenAI will "nearly triple server sales from 2023 to 2028" and that data‑center systems are the highest‑growth IT segment due to GenAI, a classic nonlinear infrastructure ramp. (gartner.com) TrendForce reports that capex by the eight major cloud providers in 2025 alone will roughly equal their combined 2023–2024 spending and grow another 24% in 2026, driven specifically by AI GPUs and custom AI ASICs. (trendforce.com)
• Chip and infrastructure vendors see surging, not plateauing, demand: NVIDIA’s revenue more than doubled year‑over‑year in fiscal 2025 (up 114%), with data‑center revenue up 93%, on top of >100% growth the prior year, all driven by AI training and inference workloads. (nvidianews.nvidia.com) Dell and others report multi‑billion‑dollar AI‑server backlogs and rapidly rising AI‑server forecasts. (reuters.com)
• Improved performance and effective cost reductions are explicitly cited as drivers: Market analyses tie rapid AI‑infrastructure growth to new, higher‑performance GPU generations (e.g., NVIDIA’s Blackwell) that deliver much more compute per dollar and thereby catalyze enterprise AI adoption. (prnewswire.com) Even where total frontier‑model training bills rise, research finds this is because organizations are scaling up compute 2.4× per year, not because demand is flattening—another sign of escalating appetite for AI compute. (arxiv.org)

Across multiple independent data points, demand for AI compute and AI applications from 2023–2025 has clearly grown in a rapid, super‑normal (effectively nonlinear) fashion rather than saturating or shrinking, aligning well with Friedberg’s prediction for the "coming years."

Key facts about RFK Jr. and the 2024 Democratic primaries:

• Robert F. Kennedy Jr. launched a 2024 presidential bid as a Democrat in April 2023, then left the Democratic Party and announced an independent run on October 9, 2023, declaring his “independence from the Democratic party.” (en.wikipedia.org)
• While a Democratic candidate, RFK Jr. did attract notable grassroots interest, polling around 19–25% among Democratic primary voters, with Biden still clearly ahead (roughly 57–60%). (pjmedia.com) That’s meaningful support but not a frontrunner “wave” on track to win the nomination.
• RFK Jr. repeatedly accused the DNC of “rigging” the primary via rules changes: no sanctioned primary debates and a reordered calendar that favored Biden (e.g., elevating South Carolina), and he argued that the superdelegate math meant he would need to win an overwhelming share of states to beat Biden. (breitbart.com) He cited these barriers while signaling he might seek “other alternatives,” which he then did by running as an independent.
• Because RFK Jr. exited the Democratic primaries before any voting, there was never a situation where an anti‑establishment Democrat was actually winning primaries and delegates and then blocked at the convention via superdelegates or similar formal mechanisms. The 2024 Democratic primaries themselves were effectively a low‑contest incumbent race that Biden dominated. (en.wikipedia.org)
• After Biden later dropped out in 2024, party leaders changed convention rules so superdelegates could vote on the first ballot in a virtual roll call that nominated Kamala Harris, underscoring the establishment’s willingness to use rule changes and superdelegates to shape outcomes. (en.wikipedia.org) But this move was about replacing a faltering incumbent with his vice president, not about stopping a surging anti‑establishment insurgent.
• Outside the primary context, the Democratic Party and Democratic‑aligned actors did pursue legal and procedural efforts that hindered RFK Jr.’s independent candidacy (e.g., an FEC complaint over alleged illegal coordination with a super PAC, and ballot‑access fights like those in Nevada), but those are not the exact superdelegate/primary‑process scenario the prediction described. (theguardian.com)

Taken together, the condition in Sacks’s prediction—an anti‑establishment Democrat building a sufficiently strong grassroots wave inside the Democratic primaries to plausibly win—never fully materialized. RFK Jr. had meaningful but minority support and left the party before any test of convention‑level intervention, and no similar candidate replaced him.

Because that core scenario never occurred, we cannot definitively say whether the Democratic establishment would have used superdelegates and other primary‑process controls to stop such a candidate. There is suggestive behavior (rule‑rigging accusations, no debates, later use of superdelegates to manage the Harris nomination, aggressive tactics against RFK Jr. as an independent), but no direct, decisive case of an insurgent Democrat being blocked at the convention.

Therefore, the prediction’s truth value is ambiguous: there is circumstantial evidence consistent with Sacks’s expectations about establishment behavior, but the specific “if X, then Y” scenario was never actually tested.

The prediction was that there was a roughly 10–20% chance that the 2024 U.S. presidential general election matchup would be Gavin Newsom vs. Ron DeSantis.

In reality, the 2024 general election was between Donald Trump (Republican) and Kamala Harris (Democrat), with running mates JD Vance and Tim Walz, respectively; neither Newsom nor DeSantis was on either major-party ticket. (en.wikipedia.org)

On the Democratic side, Joe Biden initially ran for re‑election, then withdrew in July 2024; Kamala Harris subsequently became the Democratic nominee. Gavin Newsom never entered the 2024 presidential race and had previously indicated he would not challenge Biden, instead endorsing Biden’s re‑election bid. (en.wikipedia.org)

On the Republican side, Ron DeSantis did run in the primaries but suspended his campaign on January 21, 2024 and endorsed Donald Trump, who then secured the Republican nomination at the July 2024 Republican National Convention. (en.wikipedia.org)

Since the specific scenario “Newsom vs. DeSantis” did not occur as the 2024 general election matchup, the event to which Sacks assigned 10–20% probability failed to happen. Interpreting the question as “Did that predicted matchup materialize?”, the prediction is therefore wrong in outcome terms, even though assigning a low probability to an event that doesn’t occur is not necessarily poor probabilistic calibration.