Nvidia's Historic Quarter, SaaS Comeback, Bessent vs Druck, America's Debt Crisis, Cancer Vaccine
In a Nutshell
Nvidia delivered a record $96B quarterly revenue with 70% growth guidance, signaling AI capex has real legs rather than ending as a bubble. Horizontal SaaS platforms like Salesforce are positioned to survive AI disruption by becoming systems of record that AI agents access, while vertical SaaS faces greater risk. America's $40T debt crisis is intensifying with 30-year yields at 5.3% as $10T in debt rolls over at higher rates, with AI-driven growth being the only viable path to avoid a fiscal death spiral around 2030-2032.
These notes were generated by AI and may contain inaccuracies.
Eric Weinstein is a scientist who works with Peter Thiel from Harvard. He has done some really capital work. He had a podcast for a hot minute and is a heterodox thinker in science. Weinstein has proposed alternative theories that create a grand unification theory of physics, and he's been outcast by the scientific community. People say he's a quack and use terms to chastise him. He's talked about UFOs and Jeffrey Epstein.
The point is similar to what Kratos said: so much of science has followed this sheeplike mentality where everyone has to line up and agree to the same general theory or you get outcast. If you don't follow the mainstream in science, you don't get grant funding, you don't get tenure, you don't get jobs, you don't get invited to do lectures.
This has led to stagnation in science in America because if you're not part of the mainstream, you get excluded. Because everyone has to now think in the same way, there's no heterodox thinking, no outside of the scope thinking, which means we're not pushing the envelope and discovering new things anymore.
The general theory of relativity was completely outlandish when published, but people still studied it and assessed it. They didn't discredit it and say you're an idiot. That's what goes on today in science where if you say stuff that's too far outside of the realm of the standard approach to things, you're chastised, outcast, and excluded.
There were Olympics in China where robots ran around a track. The Chinese got a bunch of people to fill a stadium and cheer for AI and robotics. Dario Amodei said all jobs are over and it's the end of days.
AI optimism in China is over 80%, meaning when polled on whether AI will be more beneficial than harmful, over 80% of Chinese people say yes. In the US, that number is around 30%.
The US is ahead of China in every category except one, which is optimism. That is the biggest risk to winning the AI race against China - doing something that will shoot ourselves in the foot because we're so pessimistic about what AI might do.
Optimus has made significant progress and Elon Musk is not showing it publicly. It would win half of the robot Olympics events already. The latest video showed Optimus without sparks coming out of its midsection.
The hard part with robots is when they encounter a condition they weren't programmed to expect, specifically when they have to interact with something in the physical world. The fidelity of Optimus and its ability to do novel things is key.
Grokbot usage limit was reached and an upgrade to 200 bucks was needed. Grokbot is always on in the cloud. You set up your agents and they can keep working even when your computer is off or goes to sleep. Other agent harnesses like OpenClaw or Hermes were on your desktop, so if your computer went to sleep, they would stop.
Grokbot is the easiest to use. It's still a lot of work to get it set up working perfectly, but it's the easiest it's ever been and you can really see where it's all going.
The idea of multiplayer mode where you can put two agents in the same room, allowing multiple people from a team to participate, would make it go very viral. Agents develop more context and expertise when multiple agents perform different tasks. Each agent can be used as a channel and people can be invited to participate.
Nvidia had a record-setting earnings report. Revenue was 96.2 billion in the quarter, up 106% year-over-year. Wall Street's expectation was 92 billion. They guided for 70% growth next year, well above what Wall Street was expecting at 45%.
Net profit was 60 billion, the most profitable core business quarter of any public company ever. Google and Amazon have had higher one-off quarters. Nvidia is the world's most valuable company at 5.5 trillion. Polymarket says 79% chance Nvidia ends the year as number one, with 6.3 million in volume.
Heading into earnings, Nvidia was only up 12% year to date, on par with the S&P 500, but lagging the broader chip index. The headwind was that people believed competition was coming.
Salesforce was up over 20% on Thursday after reporting revenue of 11.3 billion, up 11% year-over-year, in line with expectations. They beat their adjusted EPS by 80% at 590 versus 327. They raised their full year guidance to 46 billion.
Salesforce is a major investor in Anthropic. The high end of the market where the large monoliths operate is quite safe. It's a lot harder than people thought to just put in a prompt and have it all work. We're a little oversold.
The consolidation and rerating can happen in the opposite direction. Companies with constructive net dollar retention and negative churn that have the best relationships are positioned to crush.
We're getting to the end of the second phase of AI. The first phase of AI was models, like a brain. The second phase of AI is harnesses, agents, the analogous equivalent of giving a brain eyes, hands, a notebook for memory, and a keyboard to type on. This takes a primordial object good at Q&A and makes it more useful, almost like an autonomous being.
The next phase is making that autonomous agent much more informed, giving it context to do a job more intelligently. You take a brain, give it arms, legs, limbs, eyes, etc., but then you have to train it to be a lawyer, customer service rep, or sales agent. To do that next step, you need tons of contextual information.
The people who control that today are the large systems of record. They hold an incredibly special place in the ecosystem if they do it right. Mark Benioff is finding a way to play well with the models, support harnesses, build his own, and push into this next phase.
Freeberg built a CRM tool internally using cloud code and cursor over a weekend. They realized they needed to add features like account and security access, data repository protection. There were competing interests at their company on where to build software unique to the company like plant breeding software.
They realized time, efforts, and energy are better spent building workflows that are unique and create value for their organization versus recreating software other people have already made. The best ROI is not on rebuilding communications tools like Slack, messaging tools like Gmail, or CRM tools like Salesforce.
Horizontal platform companies work across many different verticals. The real value is in software that's unique for your vertical. That's where AI plays a role in building custom workflows and custom software. The real SaaS apocalypse is more of a vertical SaaS apocalypse.
The narrative of the SaaS apocalypse was totally overdone. Core systems of record like CRM are not going to get ripped and replaced with something vibe coded. Enterprises want certainty, compliance, professionally managed software that's been running and debugged for years.
Benioff made a deal with Anthropic where he's offering Anthropic's models inside of Salesforce. More importantly, he's integrating Salesforce into Claude, willing to allow Claude to be the front end to have the primary user relationship. Claude can access all data in Salesforce and all workflows and actions a user can take.
Benioff is willing to risk disintermediation to ride the AI wave because he knows it's what the customer ultimately wants. You don't want to go into Salesforce or any SaaS application to write your agents. You want to create agents in the AI program you're using and have those agents run across all your SaaS platforms.
Systems of record are still needed. AI agents go to those systems to get data from a canonical source of truth. You don't want probabilistic variability around that. AI agents need to write back to those systems and workflows.
Layers are getting built: database layer, then applications and workflows, then agents, then the AI user interface. Benioff is acknowledging that AI is now the user interface.
There's a lot of trapped value in Salesforce. The average user is never going to find all the functionality, but the agent will. The agent will go to the user and say there are actions you could take in Salesforce that line up with what you're telling me to do.
After you develop trust in your agent, the user clicks the always allow button and the agent can tap the full capabilities of Salesforce. The agent becomes a power user of Salesforce that knows how to use it perfectly.
SaaS applications need great APIs and great CLI, the command line interface, which is how agents relate to SaaS tools. In the past, most SaaS products spent a lot of time on their user interface. Now they need to think about the agent interface.
You have to build the best agent interface and relax the idea that you're going to fully control the customer relationship. Think of agents created by AI companies as an extension of your platform, not feel threatened about losing the customer relationship.
If you can do all those things and become a source of truth and system of record, then you have a big opportunity.
Benioff is one of the most underestimated and underappreciated product and sales executives. He helped get complete god access to Slack data and they built a product called Slack Today which is a 7 out of 10 and getting better.
The current version of Slack is rated around seven out of ten, with noticeable 10% improvements every couple of weeks. The approach involves running parallel links where companies maintain their own agents while allowing customers to bring their own agents. This creates a two-way win scenario where customers receive agents for free inside platforms like Salesforce.
Most agent activity will occur outside traditional SaaS products, with agents interacting with these systems to extract data, use them as systems of record and sources of truth, then write actions back when appropriate. Founders should lean into this trend rather than fighting to maintain control, as fighting it risks customer churn.
The key distinction lies in horizontal monolithic companies like Salesforce, Workday, Oracle, SAP, and IFS, which serve as unique systems of record with their general ledgers. Vertical SaaS operates more as workflows and processes, making it questionable what aspects will survive agent-driven disruption.
Just months ago, the dominant narrative predicted horizontal SaaS wouldn't survive, with major fund positions betting against products like Salesforce. However, this short thesis has proven wrong or at least overstated. The funding short problem drove indiscriminate shorting of liquid big stocks like Adobe and Salesforce, possibly working backwards from leverage requirements rather than fundamental analysis.
Benioff's stock dropped 50% before recovering 20%, reflecting the broad "SaaS Apocalypse" narrative that proved overly simplistic. The flawed extrapolation assumed AI agents can code, therefore all software goes to zero. Reality shows certain systems of record are extremely compatible and complementary to AI agents when founders position them correctly.
Nvidia's results included quietly significant moves: acquiring Hugging Face for $12 billion and Poolside for $6 billion, totaling approximately $20 billion in acquisitions—less than 1% of Nvidia's market cap. Poolside develops the Laguna coding model, positioning Nvidia to own the distribution layer for open source development.
The acquisitions involve 100 engineers moving to Nvidia, potentially working on on-premise versions of cursor and cloud code running open weight. The deals include both the model and agent harness for directing coding, though exact details remain unclear from initial announcements.
Nvidia achieved a blowout quarter with over 100% year-over-year growth, generating nearly $100 billion in quarterly revenue, $60 billion in profit, and maintaining 75% gross margins. Excluding one-off items, this represents the most profit ever generated by any company in history.
The stock rose 8% not just from the strong quarter but from guidance of 70% growth next year versus the street's 45% expectation. The 70% figure would have been higher without supply constraints. This performance shreds the narrative that AI capex is a bubble about to end, indicating real legs for continued investment.
Despite the 8% bump, Nvidia trades at only 12 times earnings. Competition concerns include Cerebras, while Jensen appears committed to open source through the Hugging Face and Poolside acquisitions. The AMD deal announcement displeased Jensen, who subsequently discussed making inference chips called "jalapeno" and reducing OpenAI investments.
Jensen is speedrunning open source strategy, owning Hugging Face as the leading open source indexer where developers test new models. Six to twelve months ago, open source was considered unable to catch up; now Nvidia owns the platform and runs open-source self-driving projects.
Businesses are converging across traditional boundaries, with delineated customer-supplier separations melting away. Nvidia still relies on hyperscalers for $24 billion quarterly revenue but has built a Neocloud business of equivalent size. As hyperscalers spin their own silicon, Nvidia appropriately responds by spinning its own models, then moving up the stack to host models, supply APIs, provide inference, and become a cloud provider.
The traditional distinction between capex and opex businesses is disappearing. In five years, major companies will have their own clouds, models, silicon, and data centers—competing soup-to-nuts on which implementation proves superior, benefiting the broader ecosystem.
30-year Treasury yields hit a 19-year high of 5.3%. Treasury Secretary Bessent doubled bond buybacks from $2 billion to $4 billion on August 19th, with reports indicating consideration of ramping up purchases to put "fear of God" into traders shorting long-dated bonds.
Stan Druckenmiller, Bessent's former mentor, published a Wall Street Journal op-ed opposing price manipulation rather than addressing core yield drivers. America's $40 trillion debt problem stems from spending issues, with no more DOGE or cost-cutting measures. Every five months brings expensive military engagements, adding $1 trillion every four months and $2.5 trillion annually—projecting $10 trillion added during Trump's second term.
During COVID era around 2020, yields remained below 0.5% and even sub-0.25% across one-month to five-year durations, with 30-year Treasuries at 1.7%. The federal government should have rolled all debt into 30-year bonds at those rates, but debt remained balanced across the curve.
Current yields show just under 4% on the short end (3.8% for one-month borrowing) and 5.2% on 30-year bonds. Persistent inflation from excess government spending on social programs drives these rates. The federal government's average cost of debt stands at 3.4% across $40 trillion outstanding.
Each 1% interest rate change costs the government 1.25% of GDP in excess interest annually. Over the next 12 months, $10 trillion in debt requires refinancing, forcing new Treasury issuance at higher rates that will increase the fiscal deficit.
Bessent has maximum authority to purchase roughly $1 trillion in debt, but must sell $10 trillion in new Treasuries over 12 months. No Treasury action can meaningfully affect the long end of the yield curve given fundamental fiscal spending problems. Druckenmiller's note provides coverage for Bessent, pointing responsibility to Congress and the President to cut spending and prevent a refinancing spiral.
Congress holds ultimate spending responsibility but remains unable to control budget deficits. Both parties consistently overspend, with debt growing at 7% versus 2-4% GDP growth—a disaster recipe. Cutting entitlements represents the necessary solution, but timing and forcing mechanisms remain unclear.
The bond market will eventually force confrontation, with yields reflecting faith and trust in the United States. Higher yields indicate declining trust. This occurs during an enormous AI financial buildout where, historically, government would provide balance sheet support, but current fiscal constraints prevent this.
Companies like Nvidia, Google, Microsoft, Meta, and Amazon carry the US economy on their backs during this buildout. The President secured $2 trillion from trade deals to support the build, though funds come from other nations with lengthy processes. This provides short-term relief but represents a band-aid—the real solution requires Congress fixing the budget problem.
If 30-year yields reach 6%, this marks the beginning of a death spiral causing years of extreme pain. The only solution involves congressional budget control, whether this year or next.
The tragedy of the commons affects spending control with 435 House members and 100 Senators each holding votes, plus presidential input. No single person controls spending, making reform extremely difficult when everyone protects their programs. The President lacks line-item veto power constitutionally.
Trump uniquely backed DOGE to the hilt, but cutting one agency triggered hysteria with accusations of killing millions of children—total nonsense that led to firebombed Tesla dealerships. Media incentives work against accurate reporting of spending reform efforts, compounding the difficulty of meaningful change.
Republicans cutting taxes and Democrats increasing spending have both contributed to worsening debt through multiple presidential administrations. The last president to balance the budget was Clinton, but the situation has deteriorated since then.
The debt crisis may manifest through rising interest rates to GDP ratios. If investors demand 12% returns on US debt and the government cannot pay, full austerity measures would be required.
Between 2026 and 2028, the inflation problem will intensify due to persistent government spending. Deficit spending correlates directly with housing, healthcare, and education costs. The economy has grown substantially since 2019, and returning to 2019 spending levels would generate a federal budget surplus.
Emergency COVID stimulus measures were normalized into permanent spending cycles, similar to post-2008 financial crisis patterns where temporary stimulus became ongoing expenditures.
When inflation remains high and affordability declines, voters will seek solutions. A political message advocating federal spending cuts is unlikely to succeed. The concern is that socialist principles and candidates will gain traction as the response to economic pain.
Social Security will lack funds to meet obligations around 2030-2032. States will face bankruptcy and inability to pay obligations during this period. Both Social Security bailouts and state bailouts will be required, potentially representing the breaking point for current fiscal structures.
California has unfunded pension obligations and tens of billions spent on a railway project with no tracks. The project runs from San Francisco to Fresno at costs exceeding airline tickets. Private industry built Bright Line in Florida at a fraction of the cost, demonstrating government programs' inefficiency.
Student loan underwriting caused administrative costs to increase 6x and tuition to rise 8% annually for 30 years. Housing interventions have made homes unaffordable for young people. Healthcare programs paying people to stay home without verification led to skyrocketing costs.
The core root of inflation is government spending across all these sectors.
Multiple layers of fraud exist within government spending, with fraud being only layer one. The tragedy of the commons means no single person controls spending, requiring massive public pressure that doesn't exist. Mainstream media makes no effort to create this pressure.
Fiscal changes require acute crisis moments, similar to 9/11, the Global Financial Crisis, or COVID. These are the only periods when structural changes in expectations can occur. Each crisis has resulted in expanded government power rather than spending restraint.
Mortgage rates average 6.73%, about 150 basis points above 30-year Treasury rates. Most people cannot connect higher government spending to higher Treasury rates, higher mortgage rates, and reduced housing affordability.
During crises, the human response is to assign blame rather than seek solutions. When mortgage rates rise, voters blame the current president or congressman and seek politicians offering free housing, free buses, and free mortgages.
AI represents the primary hope for generating exponential economic growth sufficient to grow out of fiscal problems. Regulatory delays in AI model approvals could prevent this growth trajectory. Nvidia and Salesforce's recent quarters demonstrate the growth needed over the next decade.
People are migrating from high-spending states to lower-spending states as a form of voting against fiscal policies. Housing prices in Austin have declined 27% since the 2022 peak and 25% since March 2020. Texas and Florida demonstrate that leadership prioritizing housing supply can reduce prices through free market approaches.
Druckenmiller's Wall Street Journal piece generated controversy over AI usage in writing. The piece was 90% AI-generated according to detection software, though Druckenmiller shaped the take and used AI to help write it.
The argument centers on whether AI use requires disclosure similar to other tools like Excel, Grammarly, or Photoshop. The concern is that readers expect human opinion rather than AI-generated content, particularly on important issues. The expectation is that the writer's actual voice and analysis should be present rather than AI interpretation of their views.
The group discussed a controversy surrounding an 800-word piece written by Dr. Miller on the debt problem. Some participants initially did not detect AI-generated content in the piece, finding it consistent with Miller's previous writings on the topic. The debate centered on whether Miller was taking credit for AI-written content rather than composing it himself.
One speaker expressed preference for authentic human writing, comparing AI usage to excessive autotuning in music, stating they want to understand the author's true opinion rather than an AI-interpreted version. The speaker respects Miller as a thinker with long-held positions, including his willingness to criticize former colleagues, but believes AI should enhance rather than replace personal writing.
Others noted that even top-performing artists use autotune and similar tools, and that AI gives people superpowers to improve their output. The counterargument maintained that while using Grammarly-type tools is acceptable, having AI write content in full feels disrespectful to the audience, particularly when the piece is only 800 words long.
CIA Director John Ratcliffe made an unannounced visit to Moscow on Tuesday, flying back the same day. This marked the first visit by a sitting CIA director since November 2021, when Bill Burns traveled to warn Putin against invading Ukraine. The Wall Street Journal reported that Ratcliffe was there to warn Russia against attacking a NATO country, based on new US intelligence showing Putin could be planning to test NATO's resolve with a limited assault in the coming years, potentially including cyber attacks or small land incursions.
Polymarket assigns only a 21% chance of a Russia-Ukraine ceasefire by the end of the year. Both American and Russian sides denied dramatic speculative claims that Ratcliffe was warning Russia not to invade Europe, with the president calling into a show to state the contact was routine.
The discussion emphasized that speculation about Russian invasion of NATO countries represents threat inflation, noting there is no basis for believing Russia has imperial ambitions to conquer all of Europe. The war is not going badly for Russia, though it has taken longer than expected and Russia has paid heavy casualties. Russian forces are making slow but steady progress while Ukraine suffers the most.
Russia maintains air superiority with Ukrainians running out of air defense, leading to routine bombing of Kiev. Ukrainian infrastructure is paying a heavy toll ahead of a cold winter, with power systems being destroyed. Odessa and Black Sea ports have been bombed and shut down, preventing grain exports and devastating the economy. Zelensky has stated Ukraine needs another $35 billion or they will run out of money, seeking funding from Europeans since US funding is not forthcoming.
Ukraine's casualties are described as huge, with reports of potentially drafting women due to soldier shortages. Russia has four to five times Ukraine's population, enabling sustained larger armies. The assessment concluded that Zelensky should have accepted Trump's advice that Ukraine doesn't hold the cards and should make a deal.
Meta announced new limits on Instagram that address parental concerns about children's device usage. The speaker noted appreciation for the move despite Meta being "drag kicking and screaming" to implement it, acknowledging it should have happened sooner. The controls are expected to make parenting simpler by reducing the need to police device usage, allowing parents to focus on spending time with children rather than managing screen time.
The speaker held the line at age 16 for their own children's social media access but acknowledged losing the battle at 17. The discussion highlighted that these limitations are appearing in Australia, Canada, and the UK, with potential expansion elsewhere. The physiological and psychological benefits of limits for children were emphasized, along with the inadequacy of current parental control apps.
The All-In Summit is two weeks away with several major sponsors. Iron returns as the presenting sponsor, with the Iron House serving as an on-site lounge for attendees. Merch.com is bringing back the gifting boutique where every attendee receives a bag filled with All-In merch plus special sponsor offers. Oracle is returning with an F1 racing simulator, and EY is providing meeting pods for quiet spaces to conduct calls and close deals.
Moderna's stock tripled from a $20 billion to $60 billion market cap following positive readouts from what is being called a cancer vaccine. The technology involves creating neoantigens—small peptides or proteins that mimic proteins unique to specific cancers. When introduced to the body, these antigens activate the immune system to produce cells that destroy the corresponding protein, thereby targeting and eliminating cancer cells.
The concept dates to the 1990s, with a significant advancement coming from DNA sequencing technology that allows identification of unique cancer mutations. The process involves taking a tumor sample, extracting DNA, sequencing it to identify what makes the cancer unique, then creating either the protein itself or mRNA that causes the body to produce that protein. Melanoma has been a primary target due to unique DNA mutations caused by UV light exposure.
The speaker expressed frustration that Moderna plans to charge $500,000 for this treatment, arguing the underlying technique was largely developed through decades of publicly funded research through NIH and other sources. The technology represents a process rather than a traditional drug, customized per person with costs associated with DNA sequencing, safety protocols, and human trials.
Clinics currently offer similar neoantigen treatments for $50,000 using the "right to try" framework, particularly in places like Montana. The concern centers on regulatory capture, where private companies patent techniques and charge premium prices for processes built on publicly funded research that should be more widely available.
Moderna worked on mRNA-based cancer immunotherapy for years before pivoting to develop their COVID vaccine. The company holds patents on techniques and tools for making mRNA that doesn't integrate into the human genome and doesn't use traditional replicating systems. However, the general therapeutic process of identifying cancer-specific sequences and activating immune responses is well understood and could potentially be replicated without mRNA through bacterial production of proteins.
The discussion noted that mRNA represents an effective and cheap delivery method since it doesn't require manufacturing proteins outside the body. Overseas medical tourism could make these treatments available for significantly less than US prices, with the speaker predicting proliferation of cancer immunotherapy systems in coming years.
Early detection through blood tests like Grail's Galleri test is critical for successful cancer treatment using these new modalities. CAR-T therapy represents another approach where T-cells are extracted, reprogrammed using CRISPR or similar systems to target specific cancer proteins, then returned to the body. This therapy costs approximately $1 million but shows incredible efficacy for blood cancers where cancer cells float individually rather than forming tumors.
Multiple new techniques are being developed to destroy cancer, including mRNA therapy and neoantigen therapy. These emerging approaches show significant promise in cancer treatment.
The right to try framework allows patients access to experimental treatments. This policy approach should be examined in greater detail in future discussions.
The hosts conclude the episode and express hope for successful implementation of these cancer treatment advances.
Keep All-In Podcast in your library
Save the videos and channels worth coming back to, and find them again in one place.
