Conversation with the Founder of 42 Macro: The Fed's 'Boiling the Frog Slowly' and the K-Shaped Economy

marsbitPublié le 2026-06-28Dernière mise à jour le 2026-06-28

Résumé

In a conversation with Anthony Pompliano, Darius Dale, founder of 42 Macro, discusses the Federal Reserve's monetary policy and the K-shaped U.S. economy. Dale characterizes new Fed Chair Kevin Warsh as a "dove in hawk's clothing," expecting the Fed to signal or enact policy tightening in the coming quarters to create room for later easing. He argues current economic signals, including high deficit spending, debt monetization, and credit growth, strongly indicate inflation is not on a credible path back to 2%, forcing the Fed to act. The discussion highlights the stark "K-shaped" economic reality. While top earners, buoyed by massive cash savings (up ~$8 trillion since pre-pandemic), continue robust spending, those at the bottom face severe financial strain, with delinquency rates on consumer loans reaching crisis-era highs. Dale attributes much of the current social and political anxiety to this divergence, driven by the "Cantillon effects" of monetary expansion, which disproportionately benefits asset owners. He emphasizes that in this environment of "financial repression," individuals must participate in asset markets to avoid being left behind. On equities, Dale notes a rotation from the "Magnificent Seven" tech giants into broader AI-exposed companies, while warning that the tech giants' massive capital expenditure cycles could eventually puncture over-optimistic cash flow projections. Dale concludes by stressing that the core desire across all economic strata is sim...

Source: 《Anthony Pompliano》

Curated by: Felix, PANews

Darius Dale, founder and CEO of the investment research firm 42 Macro, recently appeared on the 《Anthony Pompliano》 podcast. During the interview, they discussed the impact of Kevin Warsh becoming Fed Chair on monetary policy, the consumption dilemmas faced by Americans due to the K-shaped economy, and why every investor must participate in asset markets to survive financial repression.

PANews has compiled the highlights of the interview.

Host: What are your thoughts on Fed Chair Kevin Warsh's first press conference? How is he different from his predecessors?

Darius: Regarding your question, first I'd like to say that we see Kevin Warsh as a 'dove in hawk's clothing.'

Host: What does that mean?

Darius: It means he ultimately desires a more accommodative monetary policy. This might be due to his relationship with the current administration, but I doubt it. I think he genuinely believes AI has massive disinflationary potential. However, he must don the hawkish armor to create maneuvering room and a landing spot for the Fed. Therefore, we believe that over the next two to three quarters, the Fed will either have to tighten monetary policy or use its communication tools to signal potential tightening, or both, in order to create space for subsequent easing.

Host: Why do you think tighter monetary policy is needed? If you look at inflation expectations over the past few weeks, they seem to have started falling. Do you think this changes the Fed's need for this?

Darius: Good question. Regarding inflation expectations, we've conducted extensive statistical analysis on the drivers of inflation, studying which indicators lead or lag inflation. The results show that there's almost no statistically causal relationship between inflation expectations and future actual inflation outcomes. What meaningfully impacts future inflation results are the monetary drivers of inflation, such as the rate of change in money supply, and the expansion or contraction of money velocity—these are crucial. Then there are policy drivers, most notably deficit spending and Fed debt monetization. Also, if there's substantial banking deregulation driving a credit growth cycle, that would be a leading indicator for inflation. Finally, there are the 'output gap' drivers I don't necessarily fully subscribe to: when economic growth exceeds potential or the unemployment rate falls below the Non-Accelerating Inflation Rate of Unemployment (NAIRU), these are leading indicators for inflation. While no single indicator perfectly predicts inflation, collectively, these factors are currently sending a very strong hawkish signal to policymakers and markets: the Fed must act.

Host: So, this data indicates inflation is either moving higher, or has peaked but is plateauing at a very uncomfortable high level, thus forcing the Fed's hand.

Darius: Correct. These signals are saying two things: one, inflation is moving higher; two, inflation may be peaking but at a very uncomfortable level and will plateau there. Current data suggests we are not on a credible disinflationary path at all, and we certainly won't achieve the Fed's 2% inflation target in the near term. Let's break down these inflation drivers.

Looking at the output gap, it's currently about 110 basis points. Typically, when it reaches around 200 bps, the Fed is forced to tighten policy into a recession, so we're a bit over halfway there. We also see the unemployment rate is below NAIRU, currently by about 20 bps. In the context of AI, I don't know if we'll push it 100 bps below NAIRU, but typically by that point, the Fed has to tighten the business cycle into a downturn. On policy drivers, the annualized change rate in federal deficit spending is currently about 8%, well above the historical trendline. The annualized growth rate of Fed debt monetization is also roughly 7% to 8%, again far above trend. Bank credit is growing about 7% year-over-year, also well above trend. These growth rates are completely inconsistent with a 2% inflation environment. Furthermore, validated by our business cycle models, there's about an 18-month lag between policy rate changes and economic outcomes, meaning the lagged effects of the previous 175 bps of rate cuts are currently transmitting through the economy.

Host: Do you think the Fed has abandoned the 2% target?

Darius: Absolutely. We've been discussing this for at least five or six years. The Fed does not internally want 2% inflation, but they must signal to the bond market that they do, otherwise they lose control of the long end of the yield curve, which works against their dual mandate of 'maximum employment and price stability.' For years, 42 Macro has told the global investor community that we are all 'frogs' being boiled alive in a pot of financial repression and currency debasement. In my view, Kevin Warsh is a very competent Fed Chair in executing this job. The Fed's job is to boil us alive without letting us jump out of the pot. If we jump out, financial stability issues arise, and the real economy and asset markets face problems worse than high inflation, so it's the lesser of two evils.

Host: Even though inflation is higher now, there's no clear signal it will accelerate sharply from here. However, the latest PCE data surprised many, causing market sell-offs and concern. Different indicators are sending mixed messages, confusing many. Is this why the Fed chooses to 'watch and wait' (neither hike nor cut), kicking the can down the road? Letting this complexity sort itself out, waiting for a clearer picture before deciding.

Darius: Exactly, that hits the heart of complexity theory. I've been doing this for nearly two decades, trying to build all sorts of models. If there were an 'Occam's razor' approach to precisely predict every inflation or nonfarm payroll number to the second decimal, we would have cracked it by now. Since we can create AI, statistically predicting nonfarm payrolls shouldn't be that hard, but the problem is the variance and standard deviation in these frequently revised time-series data are just too large. Therefore, when dealing with financial markets and macroeconomic models, you must adopt a 'mosaic perspective.' It's never isolated data points, but rather the interweaving and resonance of all data points, like a school of fish swimming side by side or a murmuration of birds shifting patterns in the sky.

Host: Vivid imagery, 'school of fish' and 'murmuration of birds.'

Darius: Indeed. Currently, this 'school' is sending a message to the Fed and financial markets: 'Stop deluding yourselves that current policy is restrictive.' I think the Fed has received this signal, confirmed by the latest Summary of Economic Projections and dot plot. And we believe the next step (not yet market consensus) is: the Fed may have to enact more substantial tightening measures, or significantly raise forward guidance via the dot plot, or even make a major shift on its balance sheet. Because we know Kevin Warsh doesn't like verbose verbal interventions.

Host: What about energy prices? Oil had a spike but recently retreated, even briefly dipping below $70 a barrel in the past few days. If energy prices remain moderate or fall further, would that ease inflationary pressures and benefit US consumers?

Darius: It would certainly be beneficial. We saw this moderate resilience in today's PCE report. Real personal consumption expenditures showed a weak positive pulse, annualized at 2.1%, slightly below the 2.5% historical trendline, but it's there. It's important to note this consumption occurred against a backdrop where real disposable personal income fell about 1.5% on a monthly annualized basis, far below trend. For consumption to be only slightly below trend while income contracted sharply shows remarkable resilience in the US consumer. This was a point I pioneered in the summer of 2022 when everyone was screaming 'recession.' Recall, in the fall of 2021 on your show, I said in the next 12 months people would start talking about a word no one mentioned then: 'R' (Recession). By fall 2022, I said everyone should shut up about recession because the economy was incredibly resilient, and they were focusing on the wrong 'R' word. All current data supports our 'US economic resilience' theme, and this gives the Fed the confidence and room to tighten monetary policy.

dir="ltr">Host: Now asset prices are soaring again, the US consumer seems resilient, yet everyone is complaining they can't afford anything. My take is all three can be true simultaneously. Due to the price surge over the past five or six years, the cost of living has become unaffordable. But we also live in a society of insane output and conspicuous consumption: The Knicks are going to the Finals? Buy a ticket. Someone on Instagram has a new gadget? I need to buy two. Some travel spot is cool? I need to go take a picture. Has this simply become the norm of our consumerism?

Darius: It has been the norm since we introduced the resilience theme in September 2022. Tell me, what 'median' or 'average' statistic in the world accurately describes and can be used to set policy, or discipline your four children?

Host: Of course not, each child is completely different.

Darius: Exactly, and the same goes for macroeconomics. While aggregate statistics are useful for predicting financial markets and policy responses, these aggregates are actually not homogenous at all; once you break them down, they are very heterogeneous. In our reports to global investors, we emphasize the characteristics of the 'K-shaped economy' manifested across two dimensions.

At the bottom of the K (lower-income households), the serious delinquency rate (90+ days past due) on credit cards, auto loans, and student loans has now matched or surpassed the peaks seen during the 2008 Global Financial Crisis and the Great Depression. While the macro picture shows record-high stocks, nominal economic growth far above trend, and corporate profits at all-time highs, default rates for bottom-tier households are at financial-crisis-level distress, indicating life is incredibly tough at the bottom.

And those at the top of the K, frankly, we are making insane money. The core driver stems from the 'West Village-Montauk Effect' I shared earlier on your show. The essence is, when you have extremely high savings, you don't need to save as much from current income. Thus, the savings rate appears extremely low, but consumption is extremely high.

How did I arrive at this? For the past ten to twenty years, I've done field research in Montauk and the West Village. I noticed that at hard-to-get-into upscale spots, the biggest spenders aren't necessarily grey-haired or balding guys like us, but twenty-somethings with wealthy parents. I say this with zero judgment, but because they don't need to save for retirement, a rainy day, or caring for parents, they can spend a higher proportion of their monthly income on consumption, as they have no savings pressure.

Applying this personal anecdote to the macro economy, we see the cash stock (checking deposits plus money market fund exposure) on US household balance sheets has grown from $3.5 trillion pre-pandemic to nearly $12 trillion, an increase of about $8 trillion in cash.

Host: That's insane.

Darius: Right. When top-tier K-shaped households across the nation are sitting on this nearly $8 trillion in new cash stock, they can push their personal savings rate extremely low. They can always tap into these massive savings to support everyday and luxury consumption, regardless of what happens to current income.

Host: What is your view on the current stock market? Is AI truly an epochal salvation, or a super-bubble where you need to watch your step near the abyss?

Darius: You're asking the core question. Everyone must invest. If you don't tie yourself to the wealth-creating activities of those at the top of the K, you will be left far behind, and in the process, be mercilessly harvested by a historic Cantillon Effect. I believe the immense political anxiety and anger simmering within this country right now stems precisely from this nationwide Cantillon Effect caused by massive money printing.

(PANews Note: The Cantillon Effect refers to newly issued money not raising all prices simultaneously and proportionally, but 'trickling down' along specific paths, allowing financial institutions and the wealthy who receive the new money first to benefit by purchasing assets before prices rise broadly, while wage earners and ordinary people at the end of the monetary chain suffer purchasing power loss after prices have fully increased, leading to a hidden transfer of wealth from the latter to the former.)

Host: Are you referring to the phenomenon where the wealthiest counties in the US now all surround Washington D.C. (the political power center)?

Darius: Exactly, that's part of it, and it's the result of both political parties feeding off the government. Another staggering data point I recently saw: over 89% of seniors aged 65+ strongly support raising taxes on the younger generation to ensure their own retirement benefits are paid in full.

Host: Why isn't it 99%? The remaining 11% probably genuinely care about their grandchildren.

Darius: Actually, these seniors are also victims. The prices of essentials they need for daily survival are skyrocketing, and most of them are in their later years with fixed or very limited incomes. I fully understand and sympathize with their anxiety, but the solution is absolutely not to tax a young demographic that has no savings and barely earns any money. Scholar Peter Turchin, in his work, clearly states we are in a powerful 'wealth pump.' Ray Dalio has alluded to this in his research. For various reasons (mainly campaign finance), we allow the so-called 'elite' to plunder and squeeze the masses at the bottom of the K through regulatory capture, fiscal policy (especially a complex tax code thousands of pages long with self-serving loopholes), and monetary policy.

Take 2021, for example. On the eve of the massive inflation breakout, the Fed held policy rates a full 1,000 basis points lower than the classic Taylor Rule. In contrast, the infamous Arthur Burns at the height of his malfeasance in the 1970s only held rates 700 bps below the Taylor Rule. The Fed's balance sheet then ballooned to 36% of nominal GDP (now back to 21%).

Why do central banks continuously dilute currency purchasing power, funneling wealth from the bottom to the stock market and the wealthy? If we want to save this country, we must immediately turn off this wealth pump. But the question is, will the massive Baby Boomer generation willingly sacrifice their own interests for future generations? Most seem to be thinking, 'What more can I get for myself?'

Host: Speaking of the stock market, another interesting dynamic is the famous 'Tech Magnificent Seven.' I recently heard someone joke they're no longer 'Mag 7,' but have become 'Lag 7,' while the other 493 constituents are soaring. On one hand, this is healthy index mean reversion; but on the other, does it mean cracks are appearing in tech's armor? Valuations aren't as impervious as imagined? What are your thoughts?

Darius: Two points on the Magnificent Seven. First, since last fall, we've believed investors are using the Seven as an 'ATM,' taking profits from them and injecting capital into a broader set of 'AI application companies.' As investors get excited about AI penetration across the traditional economy, they look to the other 493 S&P stocks, or even the broader 3,000, for relative valuation bargains.

Second, from a long-term corporate operations perspective, as the Seven transition from 'asset-light' models to 'asset-heavy' models bearing massive capital expenditures, their maintenance CapEx will be pushed to staggering heights and remain elevated long-term. You can't build a data center and leave it for 10 years; you must constantly tear down, rebuild, and upgrade. Currently, Wall Street sell-side expectations project a perfect 'hockey stick' deep V rebound in free cash flow for these companies by 2029 or 2030, but historically, from all CapEx bubbles (like railroads, canals, internet tech), such intense overbuilding typically punctures such unrealistic fantasies.

Host: No wonder Musk is now pitching 'building data centers in space,' to escape angry mobs on Earth.

Darius: Indeed, if we don't proactively turn off this K-shaped class-division tuning mechanism, if we continue to tolerate the Supreme Court's recent leniency towards monopolies, the complex tax code millions of words long tailored for the wealthy, then the people at the bottom will eventually riot.

Host: In a country with 400 million guns and 300 million legal gun owners, if a riot happens, it will certainly be heard loud and clear.

Darius: I lived in New York for a long time, then Miami. But I'm now fortunate to live in an area that's not rural, but also not a major metropolis. The core is affluent, but the surrounding towns are what mainstream elites would call typical 'Trump Country.' I go to church with these people, and this experience completely shattered my preconceptions.

Because this is the first time in my life I've truly lived among 'poor white people.' I grew up in extremely poor black, Latino, Samoan, Tongan, and African immigrant communities. Then I went to Yale, which was all affluent white people; then moved to New York and Miami, which were even wealthier white and Latino elites. Until I moved here and saw the white residents of ordinary small towns at the bottom of the K surrounding the area.

I want to tell you two things.

They are nothing like the media portrays. The media has viciously and unjustly demonized this group, labeling them 'racist,' 'sexist,' even 'pathetic.' It's an utter lie. They are some of the sweetest, kindest, most caring people I've ever met in my life.

Human suffering is fundamentally the same. Having witnessed poor blacks, Latinos, Samoans, African immigrants, and now poor whites, I can say with 100% certainty that everyone wants exactly the same thing: to be able to provide for and care for their family with dignity. It's that simple.

Related Reading: Conversation with Bitwise Advisor: From K-Shaped Economy to AI Stealing Jobs, How Can Bitcoin Save the Youth?

Questions liées

QAccording to Darius Dale, why does the new Fed Chair Kevin Warsh need to act hawkish before easing policy?

ADarius Dale believes Kevin Warsh is a 'dove in hawk's clothing.' He thinks Warsh ultimately wants easier monetary policy, likely because he genuinely believes AI has significant deflationary potential. However, he must first adopt a hawkish posture to create space and a landing zone for the Fed, meaning the Fed will likely need to tighten policy or signal tightening over the next two or three quarters to pave the way for subsequent easing.

QWhat does the 'K-shaped economy' refer to in the context of the interview?

AThe 'K-shaped economy' describes a highly heterogeneous economic situation where outcomes diverge sharply. At the top of the K, wealthy households have seen their cash holdings surge (by about $8 trillion) and are consuming robustly with low savings rates. At the bottom of the K, lower-income households are experiencing severe financial distress, with delinquency rates on credit cards, auto loans, and student loans reaching levels comparable to the 2008 financial crisis, despite the broader economic strength and high asset prices.

QWhat is the main reason Darius Dale cites for the strong consumption resilience of the American consumer despite falling real disposable income?

ADarius Dale attributes the strong consumption resilience primarily to the massive increase in cash holdings on household balance sheets (rising from ~$3.5 trillion pre-pandemic to nearly $12 trillion). This 'cash overhang,' concentrated among wealthier households, allows them to maintain high levels of consumption with very low personal savings rates, as they do not need to save out of current income. He refers to this as the 'West Village-Montauk Effect.'

QWhat does Darius Dale mean by saying everyone is a 'frog' being 'boiled alive' in the context of monetary policy?

ADarius Dale uses the metaphor 'frog being boiled alive' to describe financial repression and currency debasement. He explains that the Fed's role is to slowly 'boil' the public (through inflationary policies that erode purchasing power) without allowing them to 'jump out of the pot.' Jumping out would mean a loss of financial stability, potentially causing worse problems for the real economy and asset markets than high inflation. It's presented as the lesser of two evils for the Fed.

QAccording to the interview, why is it necessary for individuals to participate in asset markets?

ADarius Dale argues that individuals must participate in asset markets to avoid being left behind and suffering from a historical 'Cantillon Effect.' This effect describes how new money benefits those who receive it first (like financial institutions and the wealthy) by allowing them to buy assets before prices rise broadly, while those at the end of the monetary chain (ordinary wage earners) bear the brunt of subsequent price increases. To protect wealth from this transfer, one must align with the wealth-creating activities of those at the top of the K-shaped economy.

Lectures associées

Le document le plus « dangereux » de l’année par Nvidia : L’IA se reproduit grâce à du code, une évolution à l’infini

**Article :** Nvidia publie ce qui est qualifié de « l'article le plus dangereux de l'année » sur l'IA auto-reproductrice et l'évolution exponentielle. **Résumé en français :** Des chercheurs de Nvidia, de Cambridge et d'autres institutions ont présenté la « Machine de Gödel de la Reine Rouge » (RQGM), un système d'IA capable d'une évolution récursive et auto-alimentée. Le système fonctionne comme un jeu de survie brutal : une IA génère et teste de nouveaux algorithmes d'apprentissage dans un bac à sable, éliminant les échecs et conservant les succès pour la prochaine itération d'auto-amélioration. L'innovation la plus frappante est que le système ne fait pas seulement évoluer les agents qui exécutent des tâches (comme écrire du code ou des articles), mais fait également évoluer les « examinateurs » ou évaluateurs qui les jugent. En utilisant un processus appelé « évolution utilitaire contrôlée », le système sélectionne périodiquement de nouveaux évaluateurs plus performants et plus stricts, qui à leur tour poussent les agents à s'améliorer davantage, créant ainsi une course aux armements récursive. Cela reflète l'hypothèse de la Reine Rouge en biologie : « il faut courir aussi vite qu'on peut pour rester au même endroit ». L'article démontre l'efficacité de la RQGM sur plusieurs tâches : amélioration des taux de réussite pour la génération de code (Polyglot), doublement du taux d'« acceptation » pour la rédaction d'articles scientifiques par des pairs humains, et meilleures performances en résolution de preuves mathématiques de niveau Olympiade. De plus, le système a corrigé un biais connu des LLM en tant que juges : une préférence pour le contenu généré par l'IA. En faisant évoluer les évaluateurs pour qu'ils identifient et rejettent les textes produits par l'IA tout en conservant la précision, la RQGM a obtenu des juges plus impartiaux. Ce travail brise une barrière théorique vieille de 20 ans associée à la « Machine de Gödel » originale, qui exigeait une preuve mathématique formelle de toute auto-modification. La RQGM utilise à la place la sélection évolutive. Ces avancées alimentent des prédictions audacieuses, comme celle du co-fondateur d'Anthropic, Jack Clark, qui estime à 60 % la probabilité qu'une IA capable d'une auto-évolution récursive (RSI) hautement autonome émerge d'ici fin 2028. Le document suggère qu'une fois qu'une IA commence à concevoir ses propres critères d'évaluation et à s'engager dans une telle boucle d'amélioration exponentielle, la voie vers une intelligence artificielle super intelligente (ASI) pourrait être raccourcie de manière imprévisible et alarmante.

marsbitIl y a 29 mins

Le document le plus « dangereux » de l’année par Nvidia : L’IA se reproduit grâce à du code, une évolution à l’infini

marsbitIl y a 29 mins

Rééquilibrage du pouvoir entre Apple et les « Microns » : Décortiquer la facture des bénéfices derrière l'iPhone

Un rapport récent analyse la répartition des profits dans l'iPhone d'Apple, démontrant un rééquilibrage du pouvoir au profit des fabricants de mémoire comme Micron. Alors qu'Apple conserve historiquement environ 25% du profit par iPhone, la part des fabricants de mémoire est passée d'environ 2% (époque iPhone X) à 12-15% du coût des matériaux (iPhone 17), leurs prix ayant considérablement augmenté. Cette inversion de tendance est principalement alimentée par la demande explosive de l'IA, qui détourne la production de mémoire (notamment le HBM et le DDR5 haut de gamme) des marchés grand public comme celui des smartphones. Le PDG d'Apple, Tim Cook, a décrit cette pression sur les coûts comme un événement sans précédent en 40 ans, conduisant à des hausses de prix sur plusieurs produits Apple. Les fabricants de mémoire, autrefois soumis à la forte pression tarifaire d'Apple, profitent désormais d'une position de force. La profitabilité record de Micron illustre ce changement. Face à cette pénurie, Apple explorerait même l'approvisionnement auprès du fabricant chinois CXMT pour diversifier ses sources. L'article conclut que l'ère de la domination absolue d'Apple sur sa chaîne d'approvisionnement en mémoire pourrait prendre fin, cédant la place à une dynamique plus équilibrée.

Odaily星球日报Il y a 2 h

Rééquilibrage du pouvoir entre Apple et les « Microns » : Décortiquer la facture des bénéfices derrière l'iPhone

Odaily星球日报Il y a 2 h

Les huit pères du Transformer, où sont-ils aujourd'hui ?

En juin 2026, Google a été marqué par le départ de deux figures clés : Noam Shazeer, co-auteur du Transformateur, et John Jumper d'AlphaFold. Ce dernier a rejoint OpenAI, ce qui a provoqué une forte réaction des marchés et des inquiétudes quant à l'exode des talents de Google. Cet événement a remis en lumière les huit co-auteurs de l'article fondateur « Attention Is All You Need » (2017), qui ont tous quitté Google. Ce modèle révolutionnaire a remplacé les réseaux neuronaux récurrents (RNN) par un mécanisme d'attention permettant le traitement parallèle, devenant la base de l'IA moderne. Aujourd'hui, leurs trajectoires sont diverses : * **Ashish Vaswani**, après avoir cofondé Adept AI et Essential AI, serait en pourparlers pour rejoindre NVIDIA. * **Noam Shazeer** a quitté Google une seconde fois pour OpenAI après avoir fondé et revendu Character.AI. * **Niki Parmar** a cofondé Essential AI avant de rejoindre Anthropic. * **Jakob Uszkoreit** a cofondé Inceptive, une entreprise de biotechnologie utilisant l'IA pour la conception de médicaments à base d'ARN. * **Llion Jones** a cofondé Sakana AI au Japon, une société axée sur des modèles collaboratifs inspirés de la nature. * **Aidan N. Gomez** a cofondé Cohere, une entreprise spécialisée dans l'IA pour les entreprises, plaidant pour la souveraineté numérique. * **Łukasz Kaiser** a rejoint OpenAI où il a contribué aux modèles de raisonnement comme o1 et GPT-4. * **Illia Polosukhin** a cofondé NEAR Protocol, une plateforme blockchain qu'il voit comme la future infrastructure pour les agents IA. En 2024, sept des huit auteurs se sont réunis lors du GTC de NVIDIA, recevant un hommage pour avoir « transformé le monde ». Pourtant, malgré leur succès, ils partagent une conviction : le Transformateur n'est pas une fin en soi. Leur héritage réside dans leur quête continue du prochain paradigme qui devra être « nettement, indéniablement meilleur ». Leurs chemins séparés continuent d'explorer les frontières de l'IA.

marsbitIl y a 2 h

Les huit pères du Transformer, où sont-ils aujourd'hui ?

marsbitIl y a 2 h

La réponse chinoise à la puissance de calcul spatial : plus efficace avec les photons, Musk et Huang Renxun prennent un chemin trop détourné

La course à la puissance de calcul spatial est devenue une véritable course aux armements. Elon Musk prédit que d'ici 2032, les satellites d'IA spatiaux alimentés à l'énergie solaire deviendront la solution informatique la plus rentable au monde. Cependant, les défis techniques dans l'espace sont bien plus rudes que sur Terre : absence de convection pour le refroidissement, exposition aux rayonnements cosmiques et contraintes énergétiques sévères. Le calcul optique, utilisant des photons plutôt que des électrons, apparaît comme une solution naturelle pour le spatial. Il offre trois avantages décisifs : une immunité naturelle aux particules énergétiques, une dissipation thermique quasi nulle et une consommation d'énergie très faible. Ces propriétés permettent d'envisager une densité de calcul supérieure pour un poids et un volume de charge utile équivalents, contournant ainsi certaines limites physiques de l'électronique traditionnelle. Le cœur de cette technologie repose sur l'exécution ultra-rapide des opérations de matrice pour l'IA via la propagation de la lumière. Malgré des défis persistants, comme l'intégration à grande échelle et la séparation mémoire/calcul, des architectures innovantes comme le calcul en mémoire photonique émergent pour les surmonter. La route vers un déploiement commercial à grande échelle reste longue, nécessitant de valider la robustesse des systèmes durant le lancement et en orbite. Si le calcul spatial parvient à atteindre un coût inférieur à celui des solutions terrestres ou à offrir des services uniques, le calcul optique, avec son potentiel de "calcul et interconnexion par la lumière", pourrait bien être la carte maîtresse pour repousser les plafonds de performance dans cette nouvelle frontière.

marsbitIl y a 3 h

La réponse chinoise à la puissance de calcul spatial : plus efficace avec les photons, Musk et Huang Renxun prennent un chemin trop détourné

marsbitIl y a 3 h

Trading

Spot
活动图片