E156: Ivy League antisemitism, macro, SaaS recovery, Gemini, Figma deal delay + big Friedberg update

The prediction was that starting in late 2023, American Jews would begin a measurable political pilgrimage toward the right, with more identifying with or voting Republican than in the past.

1. Party identification data so far
Recent large‑N surveys still show Jews as one of the most strongly Democratic religious groups:

• Pew’s 2023 data (published April 2024) finds about 69% of Jewish voters aligning with or leaning Democratic and 29% Republican, and notes that the Democratic share is up 8 points compared with 2020, i.e., if anything Jews became more—not less—Democratic over that period. (pewresearch.org)
• A 2024 Manhattan Institute survey of the Jewish electorate similarly reports 60% of Jewish voters identifying as Democrats and 23% as Republicans. (manhattan.institute)
• An April 2024 Jewish Electorate Institute (JEI) poll finds 57% self‑described Democrats, 27% independents, and only 14% Republicans. (jewishelectorateinstitute.org)
• A Jewish Virtual Library synthesis shows Jews by party ID at roughly 60% Democrat and 23% Republican in 2024, broadly consistent with these other sources. (jewishvirtuallibrary.org)
  For context, Gallup’s 2014 data already had about 61% of Jews identifying as Democrats and 29% as Republicans. (news.gallup.com) Overall, the aggregate Republican share in 2023–24 looks similar to, or slightly lower than, the last 10–15 years, not the start of a dramatic new rightward realignment.

2. 2024 presidential vote among Jews
The first post–Oct. 7 election cycle we can measure is the 2024 presidential race. Here too, data show continuity more than a break:

• Multiple exit‑poll summaries (CNN/NBC as reported by Anadolu Agency; Edison‑based coverage; JTA) indicate that Kamala Harris won roughly 78–79% of the Jewish vote nationwide, versus about 21–22% for Donald Trump—very close to the classic 70–30 type splits seen in recent decades. (aa.com.tr)
• A post‑election survey by the Jewish Electorate Institute finds 71% of Jewish voters backed Harris vs 26% for Trump—still overwhelmingly Democratic, and described as within the recent historical range. (jewishelectorateinstitute.org)
• A quantitative analysis by Split Ticket, averaging 2024 exit polls, estimates about a 6‑point swing toward Republicans relative to 2020 (they infer a true rightward shift on the order of 5–10 points), but emphasize that Jewish voters remain overwhelmingly Democratic. (split-ticket.org)

In other words, there is some evidence of a modest rightward movement compared with the unusually Democratic 2020 election, but GOP vote shares in 2024 are still broadly within the long‑standing 20–30% range for Republicans among Jewish voters.

3. Subgroup and local shifts
There are clearer rightward shifts in particular segments:

• Orthodox Jews have become heavily Republican over the last decade (a trend that predates 2023), while non‑Orthodox Jews remain overwhelmingly Democratic. (jewishvirtuallibrary.org)
• In some geographies, especially New York, Trump’s Jewish support rose notably in 2024—one exit‑poll‑based report suggests he climbed from about 30% of New York’s Jewish vote in 2020 to roughly 45% in 2024. (nypost.com)
• A Combat Antisemitism Movement poll ahead of 2024 found a non‑trivial minority of typically Democratic Jewish voters saying antisemitism might push them to vote Republican, but the net effect in that survey was only a few percentage points. (israelhayom.com)
  Complementing this, a 2025 analysis by the Jewish Voters Resource Center concludes that non‑Orthodox Jews supported Democrats in 2024 at higher levels than in 2012 and 2016, while 2020 remains a special high‑water mark driven by Trump’s unpopularity. (jewishvoters.org)

So there is some movement to the right in specific sub‑groups and locales, but at the national aggregate level Jews are still heavily clustered on the left.

4. Why the verdict is “inconclusive”
Sacks’ normalized prediction wasn’t just about a one‑off bump; it foresaw a broader, durable realignment beginning in late 2023 and playing out over multiple election cycles. As of late 2025, we have only one national cycle (2024) plus a year or so of polling to examine:

• Current evidence mostly contradicts a large‑scale “pilgrimage” to the right: party‑ID data from Pew, JEI, and the Manhattan Institute show Jews still strongly Democratic, with no clear uptick in Republican identification relative to the 2010s; in Pew’s series, the Democratic share actually increased since 2020. (pewresearch.org)
• Voting behavior in 2024 stayed within the historical band of about 70–80% Democratic vs 20–30% Republican, with several sources explicitly saying patterns were largely unchanged. (aa.com.tr)
• At the same time, there are hints of incremental rightward movement (e.g., Split Ticket’s estimated R+5–10 swing vs 2020 and state‑level gains like New York), which could, in principle, be the start of the trend Sacks described—but they are not yet outside prior historical variation. (split-ticket.org)

Because Sacks framed this as a multi‑cycle realignment and we have observed only the first cycle—with mixed but mostly continuity‑oriented evidence—it is too early to say definitively that the prediction has either clearly succeeded or definitively failed. On the balance of current data, the strong version of his claim (a marked, obvious pilgrimage to the right) has not materialized yet, but the long‑term trajectory he speculated about cannot be ruled out. Hence the verdict: inconclusive at this point in time.

Evidence shows that, following the December 5, 2023 congressional hearing on campus antisemitism where the presidents of Harvard (Claudine Gay), the University of Pennsylvania (Liz Magill), and MIT (Sally Kornbluth) testified, at least one of those presidents was indeed forced out within the ensuing months in the context of donor and political backlash.

1. Liz Magill (University of Pennsylvania)

   • Magill resigned as Penn’s president on December 9, 2023, just four days after the hearing. Multiple reports explicitly tie her resignation to the uproar over her testimony and related donor reaction, including the withdrawal of a $100 million gift by major donor Ross Stevens and intense criticism from politicians and Penn’s own advisory boards. (time.com)
   • TIME and other outlets describe her resignation as a direct consequence of the backlash to her congressional remarks about whether calls for genocide of Jews violated university policy. (time.com)

2. Claudine Gay (Harvard)

   • Gay resigned as Harvard president on January 2, 2024, roughly four weeks after Magill. Coverage consistently links her departure to a combination of plagiarism allegations and backlash from the same Dec. 5 antisemitism hearing, noting that she became the second Ivy president to step down in the wake of those testimonies. (theguardian.com)
   • Reporting also notes pressure from donors and major political figures, as well as paused donations pending Harvard’s response to antisemitism concerns. (washingtonpost.com)

3. Hearing context and chain of events

   • At the Dec. 5, 2023 House hearing, Gay, Magill, and Kornbluth were all criticized for legalistic answers about whether calls for genocide of Jews violate campus conduct codes. This testimony triggered widespread public, political, and donor backlash; more than 70 members of Congress called for the presidents to be removed. (washingtonpost.com)
   • MIT’s Sally Kornbluth, by contrast, received a statement of full support from her board and remained in office. (washingtonpost.com)

Because at least one of the presidents who testified (Magill, and later Gay as well) resigned within weeks to a month of the hearing under intense donor and political pressure, Jason’s prediction that at least one would be removed/forced to resign in the ensuing months due to donor backlash is substantively correct.

Multiple university presidents who testified in the December 5, 2023 House hearing on campus antisemitism (including leaders of Harvard and the University of Pennsylvania) resigned within weeks to a month of the hearing, under intense political and donor pressure.

• Liz Magill (University of Pennsylvania), one of the presidents at the hearing, resigned on December 9, 2023—four days after the testimony—amid heavy backlash from donors, lawmakers, and others over her comments at the antisemitism hearing. Reports describe her departure as coming "amid pressure from donors and criticism over testimony at a congressional hearing" on antisemitism. (cnbc.com)
• Claudine Gay (Harvard), another president who testified, resigned on January 2, 2024, less than a month after the hearing. Coverage notes that she stepped down following mounting criticism of her congressional testimony on antisemitism and additional plagiarism allegations, explicitly identifying her as the second Ivy League president to resign after the hearing. (theguardian.com)

Both resignations occurred within the "near term" (weeks to a month) after the December 2023 hearing, and both were clearly linked to the fallout from that hearing and the surrounding antisemitism controversy. Since Friedberg’s prediction was that at least one of these leaders would be fired or effectively forced out, and in fact two resigned under pressure, the prediction was right.

The prediction hinged on a rate cut occurring in Q1 2024 and that cut beginning an easing cycle that would help President Biden’s 2024 reelection prospects.

1. No rate cut in Q1 2024:

   • At its January 2024 meeting, the FOMC maintained the federal funds target range at 5.25–5.50%. (econbrowser.com)
   • At the March 20, 2024 meeting (still Q1), the FOMC again left the target range unchanged at 5.25–5.50% and explicitly stated it did not yet expect it would be appropriate to reduce rates. (bankingjournal.aba.com)
   • There were no inter‑meeting cuts, so no rate cut occurred between January 1 and March 31, 2024.

2. Easing cycle started later, not in Q1:

   • The Fed’s first rate cut of this cycle came at the September 2024 FOMC meeting, when it lowered the target range by 50 bps to 4.75–5.00%, its first easing move in four years. (jpmorgan.com)
   • Subsequent projections and moves in late 2024 confirm that the easing cycle began in the fall of 2024, not Q1. (federalreserve.gov)

3. Biden’s reelection prospects did not materialize:

   • Joe Biden withdrew from the 2024 race on July 21, 2024; Kamala Harris became the Democratic nominee. (en.wikipedia.org)
   • Donald Trump defeated Harris in the November 5, 2024 presidential election; Biden was not even on the ballot, so there was no Biden reelection for any easing cycle to assist. (en.wikipedia.org)

Because (a) no Fed rate cut occurred in Q1 2024 and (b) the eventual easing cycle began much later and did not lead to a Biden reelection bid or victory, Sacks’s prediction is incorrect on both its timing and political implications.

The prediction was explicitly conditional: “If you see a quarter point rate cut in Q1…”. In reality, the Federal Reserve did not cut rates in Q1 2024. The federal funds target range remained at 5.25–5.50% after the January 31, 2024 and March 20, 2024 FOMC meetings (no 25 bps cut occurred in that quarter, as confirmed by the Fed’s published meeting decisions and historical rate data).

Because the stated condition (a 25 bps cut in Q1 2024) never happened, the claim about $1T of the $5.7T in money market funds flowing into risk assets was never actually tested. We therefore cannot say it was right or wrong; it’s a counterfactual scenario.

So the correct evaluation is that the prediction is conditional and untriggered, hence inconclusive rather than right or wrong.

At the time of the December 2023 podcast, the Federal Reserve’s target range for the federal funds rate was 5.25%–5.50%, and the 10‑year Treasury yield was around 4.23% on December 8, 2023.(focus-economics.com)(etfdb.com) Chamath predicted that within roughly two years, benchmark U.S. interest rates would be “around 2.5%,” implying a drop of about 160 basis points from then-current levels. In reality, after a series of cuts in late 2024, the Fed had only reduced the target range to 4.25%–4.50% and then further to about 3.75%–4.00% by late 2025, with the effective federal funds rate running near 3.88% in November 2025—well above 2.5%.(federalreserve.gov)(federalreserve.gov) Over the same period, the 10‑year Treasury yield has hovered around 4.0% in late November 2025 and has not approached 2.5% at any point, with its 2025 lows just below 4%.(ycharts.com)(gurufocus.com) Because neither the policy rate nor the 10‑year yield has come close to 2.5% at any time within this roughly two‑year window, the prediction that rates would be around 2.5% by then is wrong, even though the general direction of lower rates was correct.

Chamath’s claim was that after the mid‑2022 to mid‑2023 software/SaaS slowdown, software revenues would start to rebound from late 2023 onward.

Sector-wide data shows there was a software/SaaS spending slump in 2022–2023, sometimes described as a software spend recession, with slower growth and weaker net new ARR for many public SaaS names. A Q4 2024 review of an index of 100+ public SaaS companies notes that this software-spend downturn in 2022–2023 is now behind the industry and that SaaS spend across all company sizes increased about 9% in 2024, based on High Alpha’s 2024 SaaS Benchmarks Report. (practicalvc.com)

Jamin Ball’s State of SaaS analysis (widely used by SaaS investors) shows that quarterly net new ARR had been slowing or shrinking for several years but began to improve in the most recent quarters before 2024, and he explicitly interprets those recent quarters as evidence that growth is starting to rebound and that 2024 is likely to be a year of reacceleration. (saastr.com) ChartMogul’s 2024 data likewise reports that SaaS companies experienced record‑low growth rates in 2023 but were back to steady positive ARR growth around 23–24% by late 2023 and into the first half of 2024, with some customer and ARR-size segments showing clear acceleration from those lows. (chartmogul.com)

On the broader software/IT side, Gartner’s figures (summarized by Investopedia and TechRadar) show worldwide software spending rising strongly again: software outlays in 2024 are expected to be up roughly 12.6% from 2023 to nearly $1.1 trillion, with a further ~10.5% increase forecast for 2025—growth that clearly contrasts with the prior slowdown. (investopedia.com) Market-size estimates for SaaS itself show global SaaS revenue growing from about $273.5 billion in 2023 to roughly $317.5 billion in 2024 (mid‑teens percentage growth) with high‑teens CAGRs projected going forward. (reddit.com)

Finally, large public cloud and SaaS vendors—Salesforce, ServiceNow, SAP’s cloud segment, Google Cloud, AWS and Azure—have all been reporting double‑digit year‑over‑year revenue growth through 2023–2024, with some (especially Google Cloud and the hyperscalers) showing renewed acceleration tied to AI workloads, indicating that aggregate software/cloud revenues are again rising meaningfully rather than stagnating. (technologymagazine.com)

Taken together, these sources show that after a clear SaaS/software slowdown in 2022–2023, sector‑wide software and SaaS revenues returned to solid, positive growth from late 2023 into 2024, with net new ARR and customer spend improving versus the trough period. That matches the substance of Chamath’s prediction that software revenues were going to rebound.

Evidence from sector-wide SaaS and cloud data supports Sacks’s call that the 2022–mid‑2023 “software recession” ended around Q3 2023 and did not re‑start afterward.

• In the episode itself, Sacks cites Jamin Ball’s Clouded Judgement chart showing that after roughly four quarters of negative net‑new ARR growth, Q3 2023 flipped to slightly positive growth (~+2%), which he interprets as the end of the software recession and the start of a rebound in software revenues. (podscripts.co)

• Subsequent Clouded Judgement data (as summarized by HC Andersen Capital) show that aggregate net‑new ARR for a large basket of global SaaS companies remained positive through 2023–2024, though at much lower levels than 2021–2022 (for example, Q2 2024 net‑new ARR was about $615m vs. $860m in Q2 2023 and $969m in Q2 2022—still growth, just weaker). (hcandersencapital.dk) This is consistent with the recession ending but the recovery being modest and choppy.

• A Q4 2024 analysis of the PVC SaaS Index explicitly concludes that “the software spend recession of 2022 and 2023 seems to be fully behind us now,” and notes that overall SaaS spend rebounded by about 9% across all company sizes in 2024, based on High Alpha / OpenView benchmark data. (practicalvc.com) That independent characterization closely matches Sacks’s thesis that the “software recession” period had ended.

• Bessemer’s State of the Cloud 2024 and Cloud 100 work show strong growth in leading cloud/SaaS names—Cloud 100 average growth accelerating from ~55% in 2023 to ~70% in 2024, and broad AI‑driven expansion—again indicating a sector that is growing, not in contraction. (bvp.com)

• By early 2025 the environment had weakened again, but in slowdown terms, not a return to outright contraction. SaaStr’s summary of Jamin Ball’s Q1 2025 data reports that aggregate net‑new ARR for public cloud software was about $1.65B, down ~30% year‑over‑year from $2.33B in Q1 2024—described as the worst quarterly performance in years, but still positive net‑new ARR, i.e., the ARR base continued to grow. (saastr.com) Commentary that growth rates have been oscillating and that “re‑acceleration” hasn’t materialized underscores a sluggish recovery, not a renewed software recession. (hcandersencapital.dk)

Taken together: since Q3 2023, sector‑wide ARR has continued to grow, and multiple independent analyses explicitly say the 2022–2023 “software spend recession” is behind us, even though growth has remained volatile and often disappointing. That matches the core of Sacks’s prediction (the end of the software recession and a return to positive net‑new ARR), so the call is best judged as right, albeit overly optimistic about the strength and steadiness of the rebound.

The prediction concerns the ultimate performance of the 2023 early‑stage startup funding vintage relative to all other venture vintages over the hosts’ entire investing lifetimes. As of November 30, 2025, we are only about two years past the 2023 vintage. Venture capital vintage performance is typically evaluated over long horizons (often 8–12+ years) because meaningful exits, IPOs, and large secondary sales usually take many years to materialize. There is no credible industry data or consensus yet that would allow anyone to definitively rank the 2023 vintage against all prior and future vintages on a lifetime basis. Therefore, it is too early to determine whether this prediction is right or wrong.

Friedberg’s core claim was that Google’s Gemini launch marked the beginning of Google being a major competitive player and a leading force in AI "for the foreseeable future (at least several years)."

What has happened so far (Dec 2023–Nov 2025):

• Google has aggressively iterated on Gemini: launching Gemini 1.5 with very large context windows and multimodal capabilities, and rolling it out broadly via Google AI Studio and Vertex AI, aimed squarely at high‑end frontier‑model use cases. (blog.google)
• It then introduced Gemini 2.0 (Flash and later Pro, Flash‑Lite), positioned as its cutting‑edge family of models, integrated into the Gemini app and exposed via APIs for developers and enterprises—exactly the behavior of a sustained top‑tier AI competitor. (blog.google)
• Market and ecosystem signals show Google as one of the central AI players:
  • A Reuters piece in November 2025 ties Alphabet’s near–$4T valuation surge explicitly to AI advances, including its Gemini 3 model and AI‑driven cloud growth, framing this as a comeback in AI leadership. (reuters.com)
  • Enterprise model‑usage data cited around Anthropic’s Claude Opus 4.5 launch reports Anthropic with ~32% enterprise AI share, OpenAI 25%, Google 20%, Meta 9%—placing Google clearly in the top tier of AI providers, not a marginal player. (businessinsider.com)
  • Reports that Meta is negotiating multi‑billion‑dollar deals to use Google’s TPUs and Google Cloud for AI workloads show Google is also a major infrastructure and chip player, competing against Nvidia on the hardware/compute side of the AI stack. (nypost.com)

Why the verdict is “inconclusive”:

• As of November 30, 2025, we are not yet two years past the December 8, 2023 prediction date—short of the "at least several years" horizon implied in the normalized prediction.
• The evidence strongly supports that so far Google has remained a leading, highly competitive AI player and that Gemini did indeed mark a durable strategic push rather than a one‑off launch.
• However, because the prediction explicitly concerns performance over several years into the future, the full time window has not elapsed. We can say the prediction is on track and directionally correct to date, but we cannot definitively confirm its multi‑year component.

Therefore, the appropriate classification is "inconclusive" (too early to fully judge), with the important nuance that all current evidence aligns with Friedberg’s forecast rather than contradicting it.

Sacks’ claim has two parts: (1) as AI answers replace traditional search, Google will not be able to maintain its historically dominant search share, and (2) Google’s AI franchise will not achieve the same kind of dominant market share that Google Search has.

On (1), as of late 2025 Google still controls about 90% of the global search-engine market, with Bing at ~4% and others far behind—very similar to its pre‑ChatGPT dominance, despite some brief dips just below 90%. (gs.statcounter.com) AI chatbots and answer engines are growing, but they have not yet replaced a large enough fraction of search queries to clearly show Google losing its “same dominant share”; its lead remains overwhelming.

On (2), in the AI‑chatbot/answer market, OpenAI’s ChatGPT currently has ~80%+ global share, while Google Gemini is only around 2–3%, well behind ChatGPT, Perplexity, and Microsoft’s Copilot in most measurements of referral or usage share. (gs.statcounter.com) That suggests Google’s AI franchise is far from Search‑like dominance, but the claim that it will never reach comparable relative share is inherently long‑term and can’t be settled in 2025.

Because the prediction is explicitly about what happens “as more and more searches get replaced with AI” over the coming years, and both Google’s search dominance and its AI position could still change substantially after 2025, there isn’t enough elapsed time to determine the final outcome. The current data are directionally consistent with the second part (Gemini is not dominant) but do not yet confirm the irreversible loss of Google’s search dominance or permanently cap its AI share. Therefore the prediction must be rated as inconclusive (too early to tell).

As of November 30, 2025, this prediction cannot yet be definitively evaluated.

Reasons:

1. Timeframe not reached
   Jason’s claim was explicitly multi‑year: “over the next several years.” From December 8, 2023 to November 30, 2025 is under two years, so we are still inside the forecast window rather than at its end. The fact that AI assistants are growing quickly does not tell us what the steady‑state multiplier versus web search will be in “several” (likely 3–5) years.

2. Lack of reliable, comparable metrics
   The prediction compares:

• “questions asked” to AI systems like Gemini/Bard (and similar assistants), vs.
• total volume of traditional web searches at the time of prediction.

But:

• Major search engines (Google, Bing) do not publish up‑to‑date, precise query counts; public estimates for Google vary widely and are often extrapolated from older disclosures or third‑party traffic tools.
• AI assistant providers (OpenAI, Google Gemini, Anthropic, Microsoft Copilot, etc.) generally report growth qualitatively (e.g., "hundreds of millions of users") or via revenue rather than disclosing total daily/annual query counts that could be cleanly summed and compared to web search volume.
• Third‑party analytics (Similarweb, Sensor Tower, etc.) give relative traffic estimates but not a unified, audited measure of all “AI questions asked” across products.

Because of this, there is no authoritative, quantitative series that would allow us to say with confidence that AI queries are now 10x–100x the baseline volume of web searches in late 2023, or even of comparable magnitude.

3. Current evidence is directionally supportive but not conclusive
   What we can say from public data and reporting is:

• Generative AI traffic and usage (e.g., ChatGPT and other major models) grew extremely rapidly through 2023–2025, indicating that “questions asked” to AI systems have indeed increased a lot.
• However, there is no credible public claim that total AI assistant queries now exceed traditional web search queries by an order of magnitude, and conventional search still appears to handle an enormous share of global information‑seeking behavior.

Given the incomplete timeframe and the absence of robust, directly comparable usage statistics, the only defensible verdict is that the prediction’s correctness is inconclusive (too early to tell) rather than clearly right, wrong, or permanently ambiguous.

Available evidence indicates that Jason’s prediction has largely played out by late 2025.

• Google did integrate ads directly into AI answer experiences. In 2024 Google began testing and then rolling out ads inside “AI Overviews” (its Gemini-powered, AI-generated answers in Search), with clearly labeled sponsored units embedded in or alongside the AI responses, explicitly framed as a way to extend its core ad business into generative AI search. (reuters.com)

• Monetization of AI answers is performing about as well as classic search ads. On Alphabet’s Q4 2024 and subsequent earnings calls, Chief Business Officer Philipp Schindler said that AI Overviews are monetizing at “approximately the same rate” as regular search results, even with a limited baseline of ads within and below the AI response. This indicates that click behavior and targeting are integrating effectively into the AI format rather than cannibalizing it. (medianama.com)

• These AI answer experiences have scaled to massive usage and are driving more (and more commercial) queries. By early–mid 2025, Google reported AI Overviews usage in the billions of users and described AI Mode (a more conversational, Gemini-heavy search experience) as driving longer and more complex queries, with noted increases in commercial query volume—i.e., the kinds of searches that are most valuable for advertisers. Management has repeatedly said these AI features are expanding the universe of queries and creating more monetization opportunities. (blog.google)

• Search ad revenue growth has materially accelerated in the period when Gemini/AI Overviews rolled out. Before these generative AI features, Google Search & Other grew at about 5% year-over-year in Q2 2023. (dazeinfo.com) After AI Overviews and related AI features were introduced and then broadened, Search revenue growth jumped into the low‑ to mid‑teens: Google Search revenue grew 14% YoY in Q2 2024, and Search & Other ad revenues rose 13% in Q4 2024 to $54B for the quarter. (searchenginejournal.com) For full-year 2024, Google Search + Other reached about $198.1B in revenue, up 13% year-over-year, and management and analysts explicitly tie a significant part of this acceleration to AI-driven search experiences such as AI Overviews and AI Mode. (visualcapitalist.com)

• Executives characterize AI as a broad revenue and growth driver, not a drag. Sundar Pichai has stated that AI is “positively impacting every part of the business” and specifically calls out AI Overviews and AI Mode as contributors to higher engagement and monetization in Search, supporting the view that these AI answer experiences are now a major, lucrative part of Google’s advertising engine rather than an experimental side feature. (nasdaq.com)

Putting this together: Google has successfully embedded ads into Gemini-powered search answers, achieved monetization rates comparable to classic search results at enormous scale, and seen a clear step-up in Search ad revenue growth during the rollout period. That aligns well with Jason’s forecast that AI-driven answer experiences would integrate ad clicks/targeting effectively and become a highly lucrative driver of new ad revenue growth over the ensuing years.

Chamath’s claim had two parts: (1) many foundational models would proliferate; (2) the marginal cost of using them would trend toward (effectively) zero, commoditizing access, over the next several years.

1. Proliferation of foundational models – clearly happening
Since late 2023 there has been an explosion of base/model families with released weights:

• Meta’s Llama series (2, 3, 3.x, 4 variants), Google’s Gemma (1/2/3 plus PaliGemma, MedGemma), Alibaba’s Qwen (multiple generations, dense and MoE, including Qwen3 with many parameter sizes), DeepSeek, Mistral, DBRX, and others, many under permissive licenses or open-weight terms.
• Alibaba alone reports 100+ open‑weight Qwen models with over 40 million downloads.(en.wikipedia.org)
• Google DeepMind’s Gemma line shows repeated open‑weight releases (Gemma 1–3 and variants) specifically intended as widely usable foundation models.(en.wikipedia.org)
• A 2025 study of Hugging Face’s ecosystem notes >1.8 million models hosted by June 2025, indicating massive proliferation of base and derivative LLMs.(arxiv.org)
• Domain‑specific foundational models like ORANSight‑2.0 build on 18 open LLMs, illustrating how many base models are now available to specialize.(arxiv.org)

On this dimension, reality strongly matches the prediction: there is a proliferation of foundational models from many vendors and communities.

2. Cost trending toward (but not reaching) “zero”
Multiple independent indicators show a dramatic fall in per‑token AI usage costs:

• Analyses of LLM pricing find that, for a given quality level, inference cost has fallen by roughly 10× per year, amounting to a 1,000× drop over about three years (e.g., from ~$60 to ~$0.06 per million tokens for models with similar benchmark scores).(thestack.technology)
• Sam Altman has publicly described a ~10× annual decline in the cost of using AI, citing a 150× reduction in token cost from GPT‑4 (early 2023) to GPT‑4o (mid‑2024).(businessinsider.com)
• Model‑as‑a‑service providers like Together.ai now sell competent open‑weight models such as Llama 3.2 3B at about $0.06 per million tokens, which is effectively near‑free for many applications.(together.ai)

However, costs have not literally gone to zero, and access to top frontier models (OpenAI, Anthropic, Google, etc.) still commands a clear price premium. Infrastructure scarcity and capacity crunches (e.g., AWS Bedrock’s 2025 GPU shortages) also show that the underlying compute is still economically scarce, which supports ongoing positive margins rather than pure commodity pricing.(businessinsider.com)

3. Why the verdict is “inconclusive”
The prediction explicitly referred to a horizon of “the next several years.” As of 30 November 2025, only about two years have passed since December 2023:

• The direction of change strongly supports Chamath’s view: there is a clear proliferation of foundational models and a steep trend toward very low marginal costs.
• But the end state—foundational model access being effectively commoditized, with costs approaching zero—has not yet fully materialized. Frontier models remain differentiated products with meaningful pricing power, and it is not yet clear whether the market will settle into full commoditization or an oligopoly of high‑end providers sitting atop a commoditized open‑model layer.

Because the forecast window (“several years”) has not fully elapsed and the ultimate market structure is still unsettled, the fairest grading as of late 2025 is “inconclusive”—the evidence so far leans in favor of his thesis but does not definitively confirm its final outcome.

Chamath’s claim is explicitly long‑dated (“over the next several years”) and about where most of the economic surplus in AI will ultimately accrue (infra/cloud and apps vs. model providers). As of late 2025, both the time horizon and the competitive dynamics are still in flux.

Evidence that hyperscaler/cloud infrastructure is capturing a lot of value:

• Microsoft reports very strong growth and profits in Intelligent Cloud: Azure revenue grew more than 30% year‑over‑year and helped push Microsoft Cloud revenue to tens of billions per quarter, with management explicitly attributing this to AI workloads.(news.microsoft.com)
• Amazon Web Services (AWS) is growing ~20% year‑over‑year on a >$130B annualized run rate, with AI and generative AI cited as key demand drivers; AWS contributes a disproportionate share of Amazon’s operating income and is backed by an AI‑driven backlog around $200B.(ir.aboutamazon.com)
• Alphabet’s Google Cloud and broader AI offerings have helped drive a ~70% stock rally in 2025 and pushed Alphabet’s market cap toward $4T, with analysts calling it a major AI beneficiary as cloud and AI tools become core profit drivers.(reuters.com)

Evidence that model providers themselves are also capturing enormous (though often unprofitable) value:

• OpenAI has raised tens of billions of dollars at valuations around the low‑hundreds of billions, with revenue already in the low‑teens of billions annually but very large projected operating losses. HSBC estimates OpenAI has signed cloud rental agreements worth hundreds of billions of dollars with Microsoft and Amazon and will need to raise over $200B by 2030 to fund compute and operating losses, underscoring both its scale and its capital intensity.(roic.ai)
• Anthropic has raised more than $13B, reaching a valuation around $183B with estimated 2025 annualized revenue around $5B and aggressive growth projections toward tens of billions in the late 2020s. It still expects to break even only around 2028, meaning much of the value is in anticipated future cash flows rather than current profits.(cnbc.com)

Evidence that hardware and models are not yet commoditized, contrary to the premise of the prediction:

• Nvidia has become the single biggest financial winner of the AI boom so far, surpassing a $5T market cap with data‑center revenue up roughly 9x in two years, and is widely described as the backbone of AI infrastructure with a strong software moat (CUDA). This is the opposite of a commoditized hardware supplier.(nasdaq.com)
• Hyperscalers themselves are pouring enormous capex into proprietary AI chips and data centers (e.g., AWS’s multi‑gigawatt expansion and custom Trainium chips, Google’s TPUs), which suggests a continuing struggle for differentiated, not yet commoditized, infrastructure.(earningsiq.co)

Netting this out as of November 30, 2025:

• Part of the prediction looks directionally plausible: cloud/infrastructure providers (Azure, AWS, Google Cloud) are clearly making substantial, real money from AI, with strong revenue and profit contributions.
• Model providers are also clearly capturing very large economic value via enormous valuations and rapidly growing revenue, even if many are still deeply loss‑making and heavily dependent on cloud partners. Whether, over the full multi‑year horizon, their investors will capture more or less value than the hyperscalers and application companies is still unresolved.
• The key assumption that models and specialized hardware will be commoditized is not yet borne out; Nvidia and a handful of top labs remain highly differentiated and richly valued.

Because (1) we are only about two years into a “several‑year” horizon, and (2) the ultimate distribution of profits between infra/cloud, applications, and model providers is still evolving with strong arguments on both sides, there is not yet enough evidence to say the prediction is clearly right or clearly wrong. It is therefore too early to call, so the outcome is best classified as inconclusive at this time.

Evidence since late 2023 shows EU and U.S. regulators taking similarly aggressive, coordinated positions on large tech acquisitions alongside the UK CMA, matching the substance of Chamath’s prediction.

1. Adobe–Figma: EU follows CMA, U.S. DOJ aligned

   • On December 18, 2023 (10 days after the podcast), Adobe and Figma terminated their $20B merger, explicitly citing “no clear path to receive necessary regulatory approvals from the European Commission and the UK Competition and Markets Authority”.(en.wikipedia.org)
   • EU and UK regulators argued the deal would eliminate a key competitor and reduce innovation in interactive product‑design and creative‑software markets—essentially the same nascent‑competitor / innovation theory of harm the CMA had advanced.(theguardian.com)
   • In parallel, the U.S. DOJ had issued a second request and was reported to be preparing a lawsuit to block the deal on similar grounds, indicating substantive alignment even though the case never reached a U.S. court because Adobe abandoned the transaction.(goodwinlaw.com)
   • Later commentary on Figma’s IPO explicitly frames the outcome as a trustbuster win after intensive scrutiny in the U.S., UK, and EU.(reuters.com)

2. Amazon–iRobot: explicit EU–U.S. coordination against a Big Tech deal

   • Amazon’s $1.4B acquisition of iRobot was dropped in January 2024 after the European Commission’s in‑depth probe concluded the deal would allow Amazon to foreclose rival robot‑vacuum makers by degrading their access to Amazon’s marketplace, and Amazon said the deal had “no path” to EU approval.(en.wikipedia.org)
   • Reuters and related reporting note that the FTC was poised to reject the deal as well, and a U.S. lawmaker opened an investigation into the FTC’s “work with the European Commission” in the collapse of the merger—direct evidence of cross‑Atlantic coordination against a large tech acquisition.(reuters.com)

3. Broader pattern: large tech and digital M&A increasingly blocked or reshaped by EU/UK/US in concert

   • A 2024 global merger‑control review finds that tech deals account for a disproportionate share of blocked or frustrated transactions, with antitrust intervention in the tech sector rising and the number of frustrated tech deals tripling year‑on‑year. Examples include: EC’s prohibition of Booking/eTraveli; CMA’s block-and‑restructure approach to Microsoft/Activision; Adobe–Figma abandoned due to EU/UK concerns; and Amazon–iRobot terminated after the EC looked poised to block it.(aoshearman.com)
   • These analyses explicitly describe authorities “ramping up” enforcement on digital/tech M&A and frustrated multi‑jurisdiction deals where firms withdraw rather than face likely prohibition—exactly the sort of “regulatory resistance” Chamath described.(aoshearman.com)

4. Demonstrated multi‑agency coordination on tech and adjacent deals

   • In 2025, the FTC’s Synopsys/Ansys decision states that FTC staff “cooperated closely” with competition agencies in the EU, UK, Japan, and South Korea on the analysis and remedies for that $35B semiconductor‑software merger, and contemporaneous commentary highlights multi‑authority coordination on timing and global remedies.(ftc.gov)
   • Earlier and continuing cases (e.g., Nvidia/Arm) show the FTC working “closely” with EU and UK authorities on tech mergers, reinforcing that such cross‑border coordination is now standard practice, especially for large tech or chip‑related transactions.(ftc.gov)

5. U.S. regulators adopting similar theories of harm, even when courts push back

   • In Microsoft–Activision, the FTC pursued theories about foreclosure and harms to cloud‑gaming and multi‑platform access that substantially overlapped with UK and EU concerns, even though it ultimately lost in U.S. courts and dropped the case in 2025.(en.wikipedia.org)
   • The fact that U.S. courts sometimes reject FTC challenges doesn’t negate the prediction, which focused on regulators’ positions and coordinated resistance, not guaranteed courtroom victories.

Taken together, post‑December‑2023 developments show:

• The EU clearly moving in tandem with the UK CMA on major tech deals such as Adobe–Figma and Amazon–iRobot.(apnews.com)
• The FTC/DOJ adopting similar nascent‑competition and foreclosure theories and actively collaborating with EU/UK peers on large technology and adjacent deals.(goodwinlaw.com)
• A broader, well‑documented pattern of coordinated, increasingly skeptical review of large tech acquisitions over the subsequent years, which has led to multiple high‑profile deals being blocked, restructured, or abandoned.

That combination matches the essence of Chamath’s forecast that, following the CMA’s aggressive stance, the EU and U.S. agencies would align in reasoning and act in a coordinated way to resist large tech acquisitions in the near term.

The prediction is explicitly about “over the long run” / “the coming decade” (roughly 2023–2033). As of today (2025-11-30), less than two years have passed since the podcast, so it is too early to judge decade‑scale effects on UK productivity and startup activity.

Current evidence on the UK tech/startup ecosystem is mixed but not decisively negative in a way that can be clearly attributed to the CMA’s merger process:

• The UK remains Europe’s leading tech ecosystem and raised about €17.5B in 2024, still the top destination for European tech investment, though down ~30% from 2023 in line with a broader global funding cooldown. (tech.eu)
• UK startup funding in H1 2025 fell by nearly a quarter year‑on‑year and deal count declined slightly, but the UK is still the best‑funded ecosystem in Europe; these trends are widely linked to global VC cycle shifts and AI‑focused reallocation rather than specifically to CMA merger rules. (sifted.eu)
• The CMA’s own 2024–2025 impact assessment and annual report emphasize large estimated consumer benefits from its interventions and a policy stance that frames competition enforcement as supporting innovation and growth, not undermining it, though these are self‑assessments and do not resolve the causal question either way. (gov.uk)

Because (a) the time horizon of the prediction (a decade) has not elapsed, and (b) existing data can be explained by broader macro and global VC conditions rather than clearly by CMA behaviour alone, the claim that the CMA’s current merger regime will over the long run measurably depress UK productivity and startup activity cannot yet be confirmed or falsified.

The prediction is framed over “the next several years” starting from late 2023, while as of November 2025 barely ~2 years have passed and the key legal changes (the Digital Markets, Competition and Consumers Act) only began to take effect in 2024. That is too short to reliably observe a structural trend in global tech-location decisions.

Available evidence focuses on regulatory design and concerns rather than on documented behaviour of firms avoiding the UK. Government reforms explicitly added a UK‑nexus criterion and safe‑harbour thresholds to ensure only mergers with a meaningful UK link are caught and to reduce burdens on businesses, suggesting policymakers are trying to limit over‑reach rather than entrench it. (gov.uk) The CMA continues to run a voluntary merger regime where most deals are never reviewed at all, which further complicates attributing any changes in office‑location decisions to merger control. (gov.uk)

Commentary in policy debates and think‑tank pieces speculates that harsh merger control could deter entrepreneurs from the UK, but other experts argue that this is overstated because the CMA can already intervene in foreign‑to‑foreign mergers with limited UK sales, as shown by the Meta/Giphy case, meaning simply avoiding a UK office would not reliably avoid scrutiny. (globalcompetitionreview.com) Meanwhile, CMA leadership publicly emphasises a more pro‑growth, proportionate approach and the need to avoid a “chilling effect” on investment, indicating political pressure to reduce the very deterrent Sacks is predicting. (theguardian.com)

Crucially, there is no systematic data or clear series of public examples showing that an increasing number of tech firms have chosen not to establish a UK nexus specifically to avoid future CMA review. Without that behavioural evidence and with the time horizon still incomplete, the prediction cannot yet be judged as clearly right or wrong, so it remains inconclusive (too early to tell).

Evidence since late 2023 shows some headwinds for tech/startup M&A and exits, but they are driven by several overlapping forces (rates, valuation reset, weak IPO window, and broader regulatory pressure), and it’s not possible to cleanly attribute a distinct, regulator‑caused “chilling effect” of the kind Sacks describes, nor to isolate its impact on risk capital.

Key points:

1. Tech and broader M&A volumes fell, but values rebounded and activity is now rising again.

   • PwC reports that in 2024 global TMT (tech/media/telecom) deal volumes were 27% below 2023, and technology deal volume specifically declined 29% year‑on‑year, even as technology deal values rose 33%, driven by fewer but larger software deals. Explanations emphasize macro factors (rates, valuation gaps) rather than regulation as the primary driver. (pwc.com)
   • Across all sectors, 2024 global M&A deal values increased about 5–10%, while deal counts fell roughly 14–17% versus 2023, again framed mainly as a consequence of higher interest rates, valuation mismatches, and PE constraints rather than a primarily regulatory story. (pwc.com)
   • By late 2024 and into 2025, several analyses (Bain, PwC, Barron’s) describe M&A as rebounding: overall deal value in 2024 is up mid‑teens percent vs 2023, deal volume up modestly, and 2025 year‑to‑date M&A value is surging, with large contributions from technology and AI deals. This is not consistent with a sustained, system‑wide freeze in M&A. (bain.com)

2. Startup exits are weak, but they were already depressed and are mainly tied to macro and valuation resets.

   • Carta’s data show startup M&A transactions in its universe fell 11% in 2023 vs 2022; IPO activity remained at historically low levels after the 2021 boom. The commentary frames the “IPO chill” and exit slowdown primarily in terms of the post‑2021 bust and market conditions, not competition policy. (carta.com)
   • PitchBook’s 2024/2025 analyses describe the startup exit market as “pretty stuck,” with LP liquidity back to global‑financial‑crisis‑era levels and many bridge/insider rounds, again pointing to valuation overhang and macro uncertainty as central issues. (techcrunch.com)
   • These trends began well before December 2023, and continue through 2024–25; they are clearly consistent with “fewer good exits,” but they are not uniquely or even primarily attributed to agencies like the CMA.

3. Regulators have increased scrutiny and some practitioners explicitly talk about a chilling effect—but aggregate data say deals are still getting done.

   • In the U.S., antitrust lawyers and deal advisors describe the FTC/DOJ’s novel merger guidelines and aggressive posture as having a “chilling effect” and raising the “merger tax” (greater time, cost, and uncertainty). Sellers sometimes accept lower‑priced bids with cleaner antitrust profiles, indicating real friction from enforcement. (dailyjournal.com)
   • Policy and advocacy pieces (e.g., NYU Law Review commentary summarized by PULSE) argue that expanded merger review requirements and a perception that acquisitions by incumbents are risky reduce the appeal of exiting by acquisition, and may therefore reduce capital flowing into the startup sector—a mechanism very close to what Sacks describes. However, these are largely theoretical or qualitative arguments, not clean causal measurements. (pulseforinnovation.org)
   • At the same time, hard data from the U.S. HSR (merger notification) report show that while second‑request investigations and enforcement intensity have risen, overall M&A filing levels in 2024–2025 are close to or above 2023, and Reuters notes that stricter review “has not significantly deterred dealmaking,” even if it increases the chance of extended investigations. (reuters.com)

4. CMA‑specific behavior is mixed, and recently has shifted away from the posture Sacks worried about.

   • The UK CMA’s tough stance did help sink high‑profile deals like Adobe–Figma, which was abandoned in December 2023 after UK and EU antitrust concerns about future competition—an example of exactly the kind of subjective, future‑competition theory Sacks criticized. (theguardian.com)
   • But by 2024–2025, the CMA is under strong political pressure to support economic growth. Its leadership has publicly committed to a more “pro‑growth” approach, stated it will scrutinize fewer global deals, and has streamlined consultation timelines. The authority also dropped a formal probe into Microsoft’s partnership with OpenAI, a major AI/tech relationship. (ft.com)
   • This evolution undercuts the specific conditional in Sacks’s prediction (that regulators continue in the same slow, expansive direction); in practice, at least in the UK, the political system has pushed the CMA to moderate rather than double down.

5. The link to “less risk capital” is debated and not empirically nailed down.

   • VC activity is clearly below 2021’s peak, and down‑rounds and lower IPO valuations are widespread in 2024–2025, but most analyses cite the interest‑rate regime and valuation reset as the dominant explanations. (fortune.com)
   • Academic work on antitrust and VC is mixed. For example, one 2023 paper finds that reductions in local DOJ antitrust enforcement actually reduced VC investment and successful exits, implying that at least some enforcement can support venture ecosystems by constraining incumbents’ anticompetitive conduct—essentially the opposite of Sacks’s argument that tougher enforcement necessarily chokes off risk capital. (arxiv.org)

Why this is “ambiguous” rather than clearly right or wrong:

• Parts of Sacks’s causal story are superficially consistent with what we see: regulators did block or derail some marquee tech deals (e.g., Adobe–Figma, earlier phases of Microsoft–Activision), practitioners and commentators talk about a “chilling effect” and higher merger‑process friction, startup exits and VC‑backed liquidity are indeed weak, and some policy analyses explicitly warn that making M&A exits more uncertain can depress startup investment. (theguardian.com)
• However, macro factors and the post‑2021 valuation bust are clearly the primary, measurable drivers of the downturn in exits and funding. Global M&A and tech M&A have rebounded in value and are growing again by 2024–2025, even under heightened scrutiny, suggesting no decisive, sustained collapse of M&A activity. (pwc.com)
• CMA‑style regulators have not unambiguously “continued” along the exact path Sacks criticized; in the UK especially, the trend since 2024 is toward somewhat more pro‑growth, selective enforcement rather than ever‑expanding, slow, subjective review of tech M&A. (ft.com)
• Finally, the empirical literature on antitrust and VC is not one‑sided; there is no robust consensus that stricter merger enforcement in tech has, in net, reduced risk capital.

Given this mix, the available evidence does not allow a clean judgment that Sacks’s prediction has either clearly come true or clearly failed. The environment is more nuanced: regulatory scrutiny has introduced friction and some chilling at the margin, but broader data show M&A and capital flows recovering under multiple influences. Hence the classification as "ambiguous."