E143: Nvidia smashes earnings, Arm walks the plank, M&A market, Vivek dominates GOP debate & more

Based on data available up to 30 Nov 2025, it is too early to definitively judge this multi‑year prediction, whose horizon explicitly runs out to around 2026–2027.

1. What Chamath predicted (Aug 2023)
He argued that Nvidia’s huge AI success would motivate and accelerate competition (custom hyperscaler chips, Tesla Dojo, alternatives like RISC‑V, etc.), and that once “priced in,” this would decay Nvidia’s margin and upside over time (i.e., meaningfully erode its profit margins and reduce upside earnings growth relative to what 2023 results implied).

2. Nvidia’s margins & earnings through 2025
Public filings and coverage for FY2024–FY2025 show:

• Nvidia’s data center revenue and earnings have exploded since 2023, and its overall gross margin has expanded, not contracted, through 2024–2025 as high‑end AI accelerators (H100, B100, etc.) have enjoyed extreme pricing power.
• For example, Nvidia reported very high gross margins in 2024 and 2025 on the back of AI GPU demand; commentary from analysts repeatedly notes sustained or improving margins, not a material erosion attributable to competitors during this period.

So far, Nvidia’s profitability has remained exceptionally strong; any future erosion implied by the prediction has not clearly manifested yet, at least not in the financial statements up to late 2025.

3. Competitive entry is clearly accelerating

• Hyperscalers (Google TPU, AWS Trainium/Inferentia, Microsoft custom AI chips, Meta in‑house accelerators) have all ramped custom silicon programs and begun wider deployment in 2024–2025, widely reported as part of a strategy to reduce dependence on Nvidia and control AI infrastructure costs.
• AMD’s MI300 series and successors, plus other accelerator vendors, are being positioned as alternative AI compute platforms.
• Tesla continues to invest in Dojo as a custom training supercomputer for autonomy and related AI workloads.

This confirms the first part of the thesis: competitive entry is indeed accelerating.

4. But the outcome (margin/earnings erosion) is not yet observable
Chamath’s core, falsifiable claim is not just that competition appears, but that it will materially decay Nvidia’s margins and upside earnings growth compared to what its 2023 boom implied, over several years.

As of late 2025:

• Nvidia still commands a dominant market share in high‑end AI accelerators and continues to post extraordinary revenue and profit growth.
• Any longer‑term impact from hyperscaler in‑house chips, AMD/Intel competition, or alternative architectures on Nvidia’s pricing power and structural margins is speculative; financial results so far do not yet show the clear margin compression his comment envisioned.
• Since his timeline runs out to roughly 2026–2027, the most decisive test period has not fully occurred yet.

Because:

1. The competitive dynamics he anticipated are emerging (supporting his intuition), but
2. Nvidia’s margins and upside earnings have not yet clearly decayed in the way the prediction ultimately asserts, and
3. The explicit time frame ("over the next several years" out to ~2026–2027) has not elapsed,

we cannot fairly label the prediction as right or wrong as of 30 Nov 2025.

Conclusion
The prediction’s evaluation window is still open, and the key outcome (sustained, material erosion of Nvidia’s profit margins and upside earnings growth due to competition) has not yet clearly occurred or been refuted in the financial data. Therefore the correct status today is "inconclusive (too early)".

The prediction’s time horizon has not fully arrived yet. Friedberg’s claim is that the 2023‑era, extremely elevated level of GPU/AI‑compute spending driving Nvidia’s results is not a long‑term steady state and that, as companies find the “efficient frontier,” aggregate Nvidia compute demand growth will slow within a few years (by ~2026).

As of late 2025, available data show:

• Nvidia’s data‑center and AI‑GPU revenue has continued to grow very rapidly through fiscal 2024 and into fiscal 2025, with multiple record quarters and strong forward demand guidance reported in earnings and widely covered in the financial press.
• Hyperscalers and large AI companies are still ramping GPU capex aggressively, and there is no clear consensus that total spend per application has already been materially “rationalized downward” in a way that has slowed overall Nvidia compute demand growth.

However, since the prediction explicitly allows until “~2026” for this rationalization and slowdown in growth to become evident, and the current date is 30 November 2025, it is too early to say definitively whether Nvidia’s aggregate compute demand growth will have slowed by the end of that window. The available evidence so far leans against a slowdown, but the prediction is framed about the state by 2026, which has not yet fully elapsed.

Because the forecast’s end date has not passed, the correct grading is: it cannot yet be decisively confirmed or falsified.

Two distinct parts of Sacks’s prediction can be checked against what actually happened:

1. “It can't continue at these growth rates” (growth will slow substantially):

   • In fiscal 2023, NVIDIA’s Data Center revenue (a good proxy for AI‑GPU demand) was $15.0B, up 41% year‑on‑year.(sec.gov)
   • Through mid/late 2023 (fiscal 2024), quarterly YoY growth in Data Center revenue surged into the hundreds of percent: Q2 FY2024 was +171% YoY and Q3 FY2024 was +279% YoY.(nvidianews.nvidia.com)
   • At the full‑year level, FY2024 Data Center revenue jumped to $47.5B, +217% YoY.(sec.gov)
   • In FY2025, Data Center revenue rose further to $115.2B, +142% YoY—still huge, but clearly down from +217% the prior year.(scribd.com)
   • By 2025, quarterly YoY growth had moderated further: Q4 FY2025 Data Center revenue was +93% YoY, Q1 FY2026 +73% YoY, and Q3 FY2026 +66% YoY.(meanewsnet.com)
     These figures show a substantial deceleration from the 200–400%+ growth rates around 2023–early 2024. So the claim that 2023‑era growth rates couldn’t continue, and would slow substantially, has clearly been borne out.

2. “There will be some steady state of demand that's at least where we are today” (no bust below 2023 levels):

   • The 2023 baseline Sacks was referring to corresponds to roughly $15B in annual Data Center revenue in FY2023.(sec.gov)
   • Since then, the absolute level of AI‑GPU demand has exploded upward, not down: Data Center revenue climbed to $47.5B in FY2024 and $115.2B in FY2025, nearly 8× the 2023 level.(sec.gov)
   • Recent quarters in 2025 show record data‑center sales—$35.6B in Q4 FY2025, $39.1B in Q1 FY2026, and $51.2B in Q3 FY2026—with NVIDIA describing Blackwell‑based cloud GPUs as “sold out” and demand “keeps accelerating.”(nvidianews.nvidia.com)
   • Amazon and NVIDIA executives also reported in 2024 that they saw no pullback in AI data‑center build‑out plans and expected power and compute demand for AI to keep rising.(cnbc.com)
     More than two years after the podcast, demand has not reverted anywhere near 2023 levels, much less fallen below them; instead, the market has moved to a far higher sustained run‑rate.

Because (a) the once‑extraordinary growth rates of 2023/early‑2024 have indeed slowed markedly, and (b) the ongoing demand level for NVIDIA’s AI GPUs is well above, not below, the 2023 baseline with no evidence yet of a collapse, the prediction is best classified as right on the information available up to November 30, 2025. While the ultimate long‑run “steady state” is still in the future, all observable data so far lines up with Sacks’s forecast trajectory rather than with a boom‑and‑bust scenario.

As of November 30, 2025, we are only about two years into the 4–5 year window Chamath specified (extending roughly to 2027–2028), so the prediction cannot yet be definitively judged.

Substantively, several pieces of the prediction are trending in the direction he described:

• Major tech firms have released “open-weight” or quasi-free large models: Meta’s Llama 2–4 families (source-available with broad commercial rights, though not fully OSI-open-source) and related ecosystem; Google’s Gemma series of open-weight models; and Databricks’ DBRX under an open model license. (arstechnica.com)
• Additional open models built on these platforms (e.g., TinyLlama, TigerBot, OpenVLA) illustrate the “open / quasi-free platform” dynamic he described. (arxiv.org)
• Industry leaders like OpenAI have publicly committed to releasing at least one open-weight model, reinforcing the broader move toward more freely usable foundational models. (wired.com)

There has also been a surge in AI startups, some of which build directly on these open or low-cost models, and the tech press explicitly highlights open models like LLaMA and Gemma as key enablers for developers and smaller companies. (techradar.com) However, verifying that “hundreds” of new companies and at least one major breakout startup are specifically attributable to these freely available big-tech models—and judging the final scale of this effect—requires seeing the full 4–5 year period play out.

Because the forecast horizon has not yet elapsed and the ultimate startup outcomes are still unfolding, the correct status for this prediction today is inconclusive (too early to tell).

It’s too early to decisively judge this prediction, because Chamath’s time horizon was “in the coming years” / “by around 2028,” and we are only at 2025‑11‑30.

Where things stand so far (partial evidence):

• Open(-weight) frontier models are significant but not dominant. Major open or permissively licensed model families exist (e.g., Meta’s Llama 2/3, Mistral models, various open diffusion/image models), and are widely used in industry and research. However, several of the most influential and capable systems remain closed or tightly controlled API products (e.g., OpenAI’s GPT‑4/4o, Anthropic’s Claude models, Google’s Gemini), indicating that “most major AI models and platforms” are not currently open.
• Open ecosystems and foundations are growing. Efforts like the Linux Foundation’s AI & Data initiatives, the Open Source Initiative’s definitions around “open” AI, and community‑led model repositories (e.g., Hugging Face) show a strong and accelerating trend toward open or semi‑open models, tooling, and platforms. But this is still in flux and does not yet settle the 2028 end‑state.
• AI hardware reference designs do exist, but are not yet the dominant paradigm. There are OCP‑style open hardware efforts for compute and AI: the Open Compute Project (originally driven by Facebook/Meta) publishes open server, rack, and data‑center designs and now includes AI/accelerator server specs; several vendors contribute reference designs for GPU and accelerator servers there. Similarly, RISC‑V and some accelerator projects publish open ISA and board/system designs. However, the most widely used AI training hardware stacks (e.g., NVIDIA HGX/DGX systems, certain cloud‑provider‑proprietary AI boxes) are still largely proprietary and vertically integrated, so we cannot yet say that open reference designs are “widely available and used” in the way Web2 commodity servers eventually were.

Why the verdict is inconclusive

• The prediction is specifically about the situation by ~2028, and we have more than two years to go.
• Current evidence points in the direction of the trend Chamath described (rapid growth of open models and of OCP‑style AI hardware designs), but the core quantitative claims — most major models/platforms having open equivalents, and open hardware reference designs being widely used and analogous in ubiquity to Web2 Open Compute — cannot yet be confirmed or refuted.

Because the key deadline has not arrived and the landscape is still evolving rapidly, the only defensible status as of 2025‑11‑30 is “inconclusive (too early).”

Arm IPO’d on Sept 14, 2023 at $51/share, implying a valuation of about $54.5 billion, and its first day of trading lifted the market value to roughly $65–68 billion, in line with SoftBank’s $60–70 billion target range. (investing.com)

Since then, Arm’s market cap has never approached Chamath’s predicted $15–20 billion “intrinsic” level. Market‑cap history shows it started around $65.25B at IPO, ended 2023 at ~$77B, 2024 at ~$130B, and as of November 2025 is around $140–150B. (stockanalysis.com) Current live data sources likewise put Arm’s market cap in the ~$140B range. (coinmarketcap.com)

The IPO lockup expired on March 12, 2024, a reasonable marker for the start of a more “normal” post‑IPO environment. Around that time, Arm was trading near $128–130/share, corresponding to a valuation well above $120B, and contemporaneous commentary repeatedly described the stock as richly or excessively valued, not beaten down. (investing.com)

Because Arm’s public‑market value has stayed many times higher than $15–20 billion for more than a year after lockups and initial hype, the prediction that it would settle in that range is clearly incorrect.

Chamath’s normalized prediction was essentially conditional: if a Starlink IPO happened in roughly the next year (by end‑2024), it would be the one offering that could broadly reopen and catalyze the U.S. tech IPO market.

What actually happened:

• As of the end of 2024 and even mid‑2025, Starlink has not gone public; it remains a private subsidiary of SpaceX, and Elon Musk has repeatedly said a Starlink IPO is unlikely before 2025 and possibly several years later. (ebc.com)
• Because no Starlink IPO occurred in the specified window, we have no direct evidence about how such an offering would have affected the IPO market; the core “if Starlink IPO, then it reopens the market” claim is therefore untestable.

We can say that the U.S. tech IPO market began to thaw without Starlink: Arm’s September 2023 IPO was widely described as a potential catalyst, followed by Instacart and Klaviyo, though that trio did not fully “reopen” the window. (fortune.com) In 2024, deals like Reddit, Astera Labs and others coincided with commentary that the IPO market was reopening or rebounding. (valuethemarkets.com) This context suggests Starlink was not strictly the only offering capable of catalyzing activity, but that still doesn’t let us observe the specific counterfactual scenario he described.

Since the antecedent of his conditional (a Starlink IPO by about late 2024) never occurred, the prediction can’t be cleanly scored as right or wrong based on real‑world outcomes, so the fairest classification is "ambiguous."

Evidence from multiple post‑debate polls shows that Nikki Haley—not Vivek Ramaswamy—had the largest positive bounce in support after the first GOP primary debate on August 23, 2023.

Key points:

• A FiveThirtyEight/Washington Post/Ipsos panel of likely GOP primary voters before and after the debate found that among debate watchers, the share considering voting for Haley jumped far more than for any other on‑stage candidate (from about 30% pre‑debate to 47% post‑debate, a 17‑point gain). Ramaswamy’s "would consider" support rose only modestly by comparison. (ipsos.com)
• A summary of national polling averages from RealClearPolitics for the two weeks after the debate (Aug 23–Sept 6) shows Haley’s vote share increasing by about 2.9 percentage points—the largest gain among debate participants—while Ramaswamy’s support actually declined slightly over the same period. (en.wikipedia.org)
• FiveThirtyEight’s analysis of Ramaswamy’s trajectory noted that while the debate boosted his profile and favorability among some Republicans, his unfavorable ratings rose even more, limiting any net positive "bounce" in overall support. (fivethirtyeight.com)
• A separate Reuters/Ipsos survey right after the debate found significant numbers of Republicans saying they were more likely to consider DeSantis and Ramaswamy, but this did not translate into Ramaswamy clearly surpassing all other on‑stage candidates in sustained polling gains; Haley still showed the clearest improvement in multiple support metrics. (ipsos.com)
• Ramaswamy did dominate attention metrics—e.g., he generated the most Google searches of the on‑stage candidates during and immediately after the debate—but those spikes in online interest did not correspond to the largest increase in actual voter support. (washingtonpost.com)

Because the best available polling and "voters considering" data show Haley, not Ramaswamy, getting the biggest post‑debate boost in support, Sacks’s prediction that Vivek would have "the biggest bounce out of the debate" is best judged as wrong, even though Ramaswamy clearly drew a great deal of attention.

Polling and analysis show Vivek Ramaswamy had a short-lived bump around the time of the first GOP debate, but he did not continue to build increasing momentum into fall 2023, nor did he meaningfully peel off Trump loyalists.

By debate day (Aug. 23, 2023) his main surge had already occurred: FiveThirtyEight’s national GOP primary average had him at about 10% support, up from roughly 4% in early July, making him a distant third behind Trump, who was near 50%.【(fivethirtyeight.com)】 Immediately after the debate, multiple national polls showed him mostly in the mid‑single to low‑double digits (roughly 6–12%), again solidly in third place and far behind Trump’s ~50–60% share.【(en.wikipedia.org)】 That is a modest bump, but it reflects a rise that had largely already happened by late August rather than new, sustained post‑debate momentum.

Through September, his national support plateaued in that same range (around 5–10% in most polls), and by October–November he was drifting downward: many polls had him in the low single digits to mid‑single digits, often behind both DeSantis and Haley, while Trump’s lead remained overwhelming at about 50–60%.【(en.wikipedia.org)】 This pattern is inconsistent with “really building momentum going into the fall”; it is better described as a brief late‑summer spike followed by stagnation and decline.

On Trump loyalists specifically, a FiveThirtyEight analysis found that Ramaswamy’s second‑choice numbers among Trump voters did rise sharply (from 8% to 26% between early June and late August), but it also emphasized that a disproportionate share of the most conservative voters remained firmly committed to Trump and “haven’t necessarily been available” to Ramaswamy or others. The piece concluded he was still a long‑shot, with Trump commanding about half of the primary electorate and most of his supporters saying their minds were made up.【(fivethirtyeight.com)】 Morning Consult’s post‑debate tracking similarly showed Trump’s perceived electability increasing to 62%, while the share of GOP primary voters naming Ramaswamy as most electable actually fell from 10% to 6%, and his unfavorables nearly doubled as more Republicans learned about him.【(pro.morningconsult.com)】 These data do not support the idea that he built substantial, durable momentum among Trump loyalists specifically.

Overall, Ramaswamy enjoyed a breakout moment and some short‑term gains in late summer 2023, but he did not keep building significant momentum “over the next few weeks and into the fall,” nor did he become a major magnet for Trump’s base. That makes Chamath’s forward‑looking prediction, as normalized here, best classified as wrong.

The prediction is explicitly conditional: if Donald Trump "doesn't make it all the way" (e.g., is jailed or loses his support), then Vivek Ramaswamy becomes the frontrunner by capturing Trump’s base.

In reality, the antecedent never occurred:

• Donald Trump remained in the 2024 Republican race, dominated the primaries, and clinched the GOP nomination on March 12, 2024. (forbes.com)
• Vivek Ramaswamy suspended his presidential campaign on January 15, 2024, after a poor showing in the Iowa caucuses and then endorsed Trump instead of surpassing him. (cbsnews.com)

Because Trump was never forced out of the race and never lost his frontrunner status, the scenario under which Ramaswamy was predicted to become frontrunner never materialized. That means we cannot evaluate whether, had Trump been knocked out, Ramaswamy would in fact have inherited his base and become frontrunner.

Enough time has passed to know how the 2024 GOP primary actually unfolded, but the key condition for the prediction was never met. So the correctness of the conditional claim cannot be determined from real-world outcomes, and the fairest classification is "ambiguous" rather than right or wrong.

Donald Trump, the clear standard-bearer of the MAGA faction, secured the 2024 Republican presidential nomination. He clinched enough delegates on March 12, 2024, when the Associated Press and major outlets declared him the presumptive GOP nominee after wins in Georgia, Mississippi, and Washington.(forbes.com) He was then formally nominated as the Republican presidential candidate at the Republican National Convention on July 15, 2024.(en.wikipedia.org)

Given that Trump is the leader and symbol of the MAGA wing and that he became the Republican nominee for 2024, Sacks’s prediction that the MAGA faction would ultimately prevail and produce the GOP nominee is borne out by the subsequent events. Therefore, the prediction is right.

Evidence from the 2024 Republican primary shows that Chamath’s envisioned late-stage non-Trump field never materialized in the way he predicted.

What he predicted
He said the late non-Trump field (when only 3–4 candidates were still "meaningfully competitive") would be: Ron DeSantis, Vivek Ramaswamy, and either Tim Scott or Nikki Haley.

What actually happened

1. Tim Scott exited very early.
   Tim Scott suspended his presidential campaign on November 12, 2023, well before any voting began and long before the race reached a late, narrowed stage. (en.wikipedia.org)
   → He was not part of any late-stage field.

2. Vivek Ramaswamy dropped out immediately after Iowa.
   Ramaswamy suspended his campaign on January 15, 2024, the night of the Iowa caucuses, after finishing fourth, and endorsed Donald Trump. (en.wikipedia.org)
   → He did not remain in the race into the later, winnowed phase.

3. Ron DeSantis exited before New Hampshire.
   DeSantis suspended his campaign and endorsed Trump on January 21, 2024, two days before the New Hampshire primary, after a poor showing in Iowa. (cnbc.com)
   → He also did not persist into a true late-stage multi-candidate field.

4. The actual late-stage non-Trump field was essentially just Nikki Haley.
   After DeSantis and Ramaswamy exited, coverage and polling described the race as effectively a two-person contest: Trump vs. Haley. (apnews.com) Haley then suspended her campaign on March 6, 2024, following Super Tuesday, leaving Trump with no major opponents. (en.wikipedia.org)
   → By any reasonable definition of the “later stages” of the primary, the only serious non-Trump contender remaining was Haley alone, not a trio of DeSantis, Ramaswamy, and Haley/Scott.

5. Brief early-voting moment doesn’t save the prediction.
   Just before and at Iowa, the principal non-Trump candidates were DeSantis, Haley, and Ramaswamy, with others (e.g., Chris Christie) either already out or polling in the low single digits. (en.wikipedia.org) However, that phase was immediately followed by Ramaswamy and then DeSantis suspending their campaigns, after which only Haley remained as a meaningful non-Trump alternative. This rapid collapse means his scenario of a late-stage, enduring non-Trump field composed of DeSantis, Ramaswamy, and Scott/Haley did not actually characterize the later primary.

Because Tim Scott exited well before the late stage, and because both DeSantis and Ramaswamy dropped out early in the voting calendar—leaving Nikki Haley as the lone substantive non-Trump challenger—Chamath’s specific prediction about the composition of the late-stage non-Trump field was wrong.

Donald Trump ultimately did skip every Republican primary debate in the 2023–24 cycle; contemporary coverage and reference summaries note that he “did not attend any primary debates” and that he secured the nomination “despite not participating in any of them.” (en.wikipedia.org)

He clinched a majority of delegates for the 2024 GOP presidential nomination by March 12–13, 2024, and was formally nominated as the Republican presidential candidate at the Republican National Convention in Milwaukee on July 15–18, 2024. (cnbc.com)

Throughout the competitive primary period he maintained a dominant polling lead over rivals, with national averages around 50–60% for Trump versus teens for his nearest competitor and many individual polls showing 40–50+ point advantages well into late 2023 and early 2024. (fivethirtyeight.com)

Taken together, these outcomes match Friedberg’s forecast that Trump would likely skip all GOP primary debates yet still secure the Republican nomination while maintaining a large polling lead.

Vivek Ramaswamy did not win, and ultimately did not come close to winning, the 2024 Republican presidential nomination. Donald Trump clinched the GOP nomination by surpassing the required delegate threshold on March 12, 2024, after winning multiple primaries that day, and became the party’s presumptive nominee.(forbes.com) Ramaswamy suspended his campaign after poor early‑state results and ended the primaries with only a small fraction of delegates.(en.wikipedia.org) At the Republican National Convention in July 2024, Trump was formally nominated as the Republican presidential candidate.(en.wikipedia.org) Since the prediction was that Ramaswamy had a strong chance and "can win this thing" (the GOP nomination), and the nomination instead went decisively to Trump, the prediction is wrong.

Vivek Ramaswamy never “slipstreamed past” or overtook Donald Trump in the GOP electorate: he finished a distant fourth in the Iowa caucuses, then suspended his 2024 Republican presidential campaign and endorsed Trump, acknowledging there was no viable path forward for himself.(apnews.com) Donald Trump went on to clinch the 2024 Republican presidential nomination by securing the required delegate majority on March 12, 2024, and was later formally nominated as the GOP presidential candidate at the Republican National Convention in July 2024.(forbes.com) Since the condition Chamath stated (“if Vivek doesn’t figure out a way to slipstream past Trump”) was met, and the predicted outcome (“Trump will win the nomination”) also occurred, the prediction is judged right.

As of November 30, 2025, Vladimir Putin remains President of Russia and has not been removed from power by Western-backed regime change or internal overthrow. He was sworn in for a fifth presidential term on May 7, 2024, confirming continuity of his rule during the ongoing war in Ukraine.(en.wikipedia.org)

The Russia–Ukraine war is still active, with large-scale Russian offensives and continued fighting into late 2025, indicating that the conflict has persisted for years without any collapse of the Russian regime or successful effort by Western countries to install alternative leadership in Moscow.(en.wikipedia.org)

Given that more than two years have passed since the August 2023 prediction and throughout this period Putin has retained power and the Russian state has remained politically intact, Sacks’s core claim—that Western hopes of ousting Putin via pressure or instability were unrealistic and that other countries would still have to deal with him during the war—has been borne out by events so far, even though the war itself is not yet over.