Trump-Xi Summit, Benioff: "Not My First SaaSpocalypse," OpenAI vs Apple, Multi-Sensory AI, El Niño
In a Nutshell
All-In Podcast #273 features Salesforce CEO Mark Benioff on the Trump-Xi summit, emphasizing economic entanglement via trade deals (soybeans, Boeing jets, chips) to avoid Thucydides Trap conflict, with CEOs like Musk and NVIDIA poised to penetrate China's market amid Taiwan tensions and chip proliferation debates. Benioff dismisses the "SaaSpocalypse," highlighting Salesforce's AI strategies (AgentForce, Informatica acquisition, $300M Anthropic spend for coding agents) and resilience with $46B revenue, predicting market re-rating as high-end SaaS thrives. Segments cover OpenAI-Apple disputes, multi-sensory AI hardware shifts, record El Niño food crisis risks, and critiques of Anthropic's dark SPVs, ending with Benioff's 1-1-1 philanthropy model and tribute to Susan Wojcicki.
These notes were generated by AI and may contain inaccuracies.
All-In Podcast episode 273. Sax is out, Mark Benioff, CEO of Salesforce, is the guest. No software CEOs right to China. Benioff got the invite. Discussion on his status with the administration. Benioff clarifies he's not a Democrat or on the left, supports the country, allegiance to America, independent American. References to invites: Windsor Castle, Prince Charles visit, Saudi dinner.
Trump-Xi summit begun after two-month delay due to war in Iran. First visit to China since 2017. Seventh face-to-face meeting for Trump and President Xi. China 12-15 hours ahead. Thursday was day one. China agreed Strait of Hormuz should remain open, no military commitment, Iran should not have nuclear weapon. On Taiwan, Xi warned if handled poorly, countries will collide or clash, putting U.S.-China relationship in dangerous situation. Polymarket: Taiwan safe for 2026, 6% chance China invades this year on $23 million volume; 17% chance by end of 2027. Debate on timing: after Trump out of office or during.
On trade, China committed to buy more U.S. soybeans, oil, LNG, 200 Boeing jets. Talks laid out vision for constructive, strategic, stable ties. Trump effusive in praise for friendship with Xi, with disclaimer people don't like it when he praises.
Xi referenced avoiding Thucydides Trap in opening remarks: rising power meets declining power, often leads to conflict. Opportunity in resource expansive world with increasing productivity via AI, automation, biotech for abundance, multipolar world, share growing pie over next 30 years, avoid conflict assuming static pie. Political win for Trump: trade deals increasing job security, prosperity, income, investment in American jobs.
Strategic goal: chart economic cooperative path avoiding conflict, sharing bigger pie. Trump bringing CEOs to penetrate China's closed consumer economy with financial firepower: planes, cars, chips, hard equipment. Chinese top-down Confucian vs. American bottoms-up individualist, room for cooperation via economic entanglement, bi-directional after years of one-way cheap goods.
Prominent Democrat called media wrong on CEOs; simplest path to no-conflict detente is economic entanglement. Behind closed doors, dividing pie: globe, economy. China needs energy, critical tech from U.S.; quid pro quo de-escalation in Central/South America, Middle East, Asia-Pacific, carving bipolar world.
Xi wants wider door for business. Best salespeople there: President Trump, Elon Musk, Jensen (chips), Kelly Ortberg (200 Boeing planes, wants 500), Brian Sikes (Cargill soybeans, world's largest private company in Minneapolis), financial services. CEOs return with orders.
Business as greatest platform for change. Trump and Xi like each other, sales-oriented, not first meeting. Salesforce in China: no offices/employees, data residency laws, exclusive partnership with Alibaba, divests acquired presences. Significant revenue, key customers like Louis Vuitton, Loro Piana use Salesforce globally and via Alibaba in China; data kept in China, accessible by CCP. Elon exceptional: Tesla in China, no partnership, phenomenal Xi relationship, AI cars with cameras in Chinese factories.
Cynical view: China studied Tesla, now biggest EV competitor with BYD (cool cars like Gullwing, potentially $20-30k in U.S., hollow out U.S. auto industry). Tim Cook stores data in China, controversial for private info/dissidents.
Inflation over 3%, Iran war seen as betrayal, China not material to well-being. Trump on six-month midterm timeline, two years left in office; Xi on 100-year playbook. MAGA evolving to America First/Only. Second/third-order effects key: job/income security, cost of living. Deflationary trade deals lower goods prices 30-40%, boost exports (soybeans, oil, tech/software) for productivity/jobs.
Schizophrenic: want cheap Chinese goods but tariffs/decoupling. Midterms swing more on gerrymandering (Supreme Court, Virginia Supreme Court, state legislatures), money organized (Andreessen Horowitz largest donor per NYT, building to trillion AUM like Blackstone, politics permanent for geopolitical investments).
Incentive to divide world: China stranglehold on critical resources, U.S. on energy/AI. Trade: China out of Chile/Venezuela/Panama, U.S. less on Strait of Malacca for rare earths/oil/abundance. China issues: population, real estate, GDP decline; Xi focus avoiding Tiananmen, sustaining 500 million from poverty to middle class (largest globally, luxury goods), factories operating against undercutting.
Trump wants board of trade beyond peace. Xi wants wider door. Visa/Mastercard CEOs there to open commerce dominated by Alibaba super apps. NVIDIA, Qualcomm CEO (Cristiano) for chips. More economic collaboration = more peace.
Hard questions: Sell latest chips? Benioff: irrelevant, Chinese models competitive without highest-end chips (fast follow 6 months behind), ego gratification; sell whatever for cooperation, Taiwan nonsense story.
Inevitable China gets chips. Taiwan less relevant as U.S./China mainland fabs (TSMC Arizona running, China with Huawei multi-deca-billion investment in fabs/semiconductor equipment vs. ASML blocks). By 2040 cultural moment for China, U.S. may not care economically/security-wise.
Global productivity index climb benefits all; West-only growth forces conflict, diffusion proliferates tech, commoditizes, boosts living standards/economies, reduces conflict. Arms trade: ask China not sell to Iran; nuanced if trade for no U.S. arms to Taiwan.
So you say make the trade.
Chamath predicts 18 months until Taiwan is less critical due to advancing to 1-2 nanometer chips and scaling chip fabs. Orthogonal technologies like Neuralink's nanometer-scale machine for automated brain implantation provide mechanical dexterity. Current Taiwan importance is economic; removing it changes attitudes. Sell chips to China so Nvidia wins over Huawei. Models are catching up at same pace. Need reasonable KYC for model usage to prevent bioweapon creation in basements. Proliferate chips, let Jensen win.
Mark disagrees on defending Taiwan if China blockades; calls it nonsense, predicts China-Taiwan reconciliation. Schedules Chamath for Neuralink Tuesday at 5 PM jokingly to add empathy.
Elon says Neuralink doesn't work on Chamath. Jokes about Wizard of Oz: "If I only had a heart," "brain," "Neuralink," courage for Friedberg. Chamath took kids to Wizard of Oz at the Sphere; enriching experience. Chamath avoids large crowds due to panic disorder, bad thoughts about exits or terrorist attacks.
Tony Robbins at Mark's conference. Mark attributes Salesforce success to Tony: without him, no Salesforce. Key insight: quality of questions determines quality of life. Questions: What do you want? What's important? How to get it? What's preventing it? How know you have it? Wrote it down for clarity and motivation. Ran across coals multiple times.
All In Summit September 13-15 at allin.com/events. Mark returns in 2027.
PSA: Dog Scooter in Baldwin Park at LA Animal Care and Control needs home; shelter overwhelmed, dogs euthanized. Link on YouTube. $10,000 keeps dogs alive one week. Chamath has shelter cat named Cloud.
Salesforce down 37%, $90B loss. ServiceNow 42%, Workday 45%, $180B total market cap loss. Fear AI eliminates need for Slack, Salesforce, HubSpot; users ask AI via glass interface.
Mark: Not first SaaSpocalypse. Drinking Salesforce Espressos, pet Sasquatch, more sassy. Market re-rated; cycles every few years (27 years Salesforce, 40 enterprise software). Top 10 enterprise software great quarters, trading at 2x sales (e.g., HubSpot). AI hypnosis, no numbers yet proving impact; enterprise software still selling.
Past SaaSpocalypses: 2020 COVID collapse, 2016.
Strategy: Ignore stock price, anchor on customer success. $46B revenue, $16B cash flow, 83,000 employees. Focus on long-term success. Public markets rationalize; private valuations fantasy.
Chamath: Low-end SaaS finished, high-end safe (large monoliths like Salesforce). OpenAI's $4B deployment company deal ($17.5B valuation, 1.5% preferred return) competes with EY, Accenture, Deloitte, PwC, Cognizant; doesn't make sense vs. their multiples. High-end harder than prompts. Oversold; consolidation re-rating reverses. Opposite trade: constructive net dollar retention, negative churn, trusted C-suite relationships (20 years). Markets ask ROI on $3T AI spend; go to Mark et al. to sell tokens. Multiples on revenue/assets correct down, SaaS up.
Salesforce $50B stock buyback, one of largest.
Mark: Coding agents awesome; $300M Anthropic usage for coding. Cheaper, efficient, faster implementation/sales, humans+agents+headless platforms. Unprecedented efficiency in service/support/distribution/marketing. Acquired Informatica ($8-9B) for data harmonization, semantic layer, single source of truth.
AgentForce: Call 1-800-KNOW-SOFTWARE, authenticates, escalates to human with full context. Trinity: phone, web, human.
Slack headless via APIs (XML, SOAP, REST, CLI, MCP, AXL); apps embedded, stream to LLMs/devices.
Context in Slack/email for world models. Slack bot reads DMs/channels, reveals business insights (top deals, employee issues, priorities). Connects to Salesforce/Google. Prompt: every 2 hours, decisions made, who, handicap as chief of staff/CEO/board member personas. Upgrade to higher plan ($6-18K for small company) for more context to open Claude, Perplexity, etc. OpenAI/Anthropic standardize on Slack, Salesforce clouds.
Advice to pre-AI private software CEOs: Focus revenue, customers, cashflow, profitability, innovation, customer value. Public markets rationalize.
Friedberg: Dropping vertical software, doubling down horizontal platforms for custom apps/workflows. Arbitrage founder-led vs. non, vertical vs. horizontal.
Salesforce: 15,000 salespeople; called back 20-30M leads via agents (acquired Qualified). Qualify BDR/SDR, outbound.
OpenAI considers suing Apple over ChatGPT Siri/iOS/MacOS integration. Deal poor: requires "Okay, let's go to ChatGPT"; no promotion; users prefer standalone app; no billions in subscriptions. Apple: privacy concerns, OpenAI hardware competing (with Jony Ive), recruited design guru.
Mark Benioff, you know the players.
Do you trust Sam Altman at OpenAI? LLMs are coming out, every company chose a different path. Elon went with Grok, building companions and sex bots as a huge focus. OpenAI doing Sora video and ad networks. Gemini has nano banana. Anthropic rejected sex bots and nano banana, focused on coding agents. Anthropic was right, their rocket ship took off. Now everyone is doing coding agents too, resetting: kill Sora, kill sex bots, cursor on.
This is a dynamic moment in industry, exciting because companies must pivot, refocus, constantly ask what users want now. A year ago everyone was in a different place. With Anthropic 4.6, everyone can code in their companies. Before, it was minor productivity improvement. Now unbelievable. Working on technology inside Slack to make coding easier—breaking news coming.
Chamath has a whole company around coding agents. Before coding was humans only—speaker started at 15 in Burlingame High School, hands-on keyboard writing video games in BASIC 68000, 6502.
How has it changed org? Product manager, developer, UX designer all blended into one.
Most engineers just speak. One guy has foot-button-operated thing with Whisper Flow plus pedal—all talk, no hands-on keyboard. Dolphin flow at Salesforce, dolphins on keyboards.
Now that Sam shut down sex bots, how filling that void?
Take on strained OpenAI relationship with world, specifically Apple. Not many long-term partnerships with OpenAI. What should Apple do—go back to Google where they made $100 billion on search default partnership?
Google opportunity to integrate Gemini assistant into personal info: Gmail, Calendar, Google Drive, Google Photos. Point for calendaring, asking questions about personal life like "when did I email this person." In enterprise, G Suite up for grabs. Google can own assistant interface. Salesforce maybe too.
Apple has embedded install base, billions using Apple services for mail, calendaring, storing info.
World may silo into assistants with access to all your info, enterprise or personal. Apple needs to build own or white label with partner like Anthropic.
Google Assistant will be kick-ass once flywheel spins.
Apple clearest path to top 2-3 AI player by buying Perplexity, Mistral, or AI lab, using hardware footprint. New MacBook with 48GB RAM M5 unbelievable, new studio models with 1TB RAM. Clear path to performance, privacy with local models. Index images without sending data to OpenAI (nobody trusts) or Gemini.
Picked current CEO because hardware engineer. Future is local models on desktop hardware. Employees with this hardware 10x more valuable.
Not huge believer in local models—need persistence across multiple devices, personas, web browser vs home computer. In 2026, no persistence is breaking feature. Won't lug 5-pound MacBook Pro everywhere.
Both right—use iCloud for this. Form factors may change, iPhone moment surprise with new device.
Apple incredible from 30-40 years process optimization—what if form factor totally different? How pivot organization?
Mira Murati released Thinking Machines new real-time—impressive. Apple patented cameras in AirPods. Persistent hardware like watch, mirror pieces know entire reality, monitor desktop.
Leads to 1000x token use vs current business users. Mira's model watches desktop, listens to voices, webcam simultaneously, uploads every 200ms to two models: one deep thinking on 32s clip, one real-time. Not turn-based prompting, always querying LLM—requires hardware upgrade to average desktop.
One small step for man, one giant leap towards AGI.
We've thought LLMs are be-all-end-all to AGI, but LLMs are just language, word-by-word prediction. Won't get to Minority Report or sci-fi.
Multi-sensory models next big wave. Humans are multi-sensory biological computers with eyes, ears, mouth, brain, heart. Still not AGI. Demos awesome, everyone should see. Model companies pivoting again, pick poison, can't do everything.
Spending $300M on Anthropic tokens this year for coding. Vast majority don't need Anthropic—need intermediary maestro layer to route to most affordable model.
Early moment, edge and cloud come together, distributed intelligence. More efficient, not always expensive. Hot new company will sit between you and Anthropic/OpenAI, use their tokens only when needed—like needing some water, not all the water.
Forecast for sea surface temperature anomalies—ocean heated so much, temperatures to exceed anything in recent history, 4 degrees above normal.
Top image: sea surface temperature anomaly vs 1877 biggest El Niño year.
Oceans are battery of weather—absorb heat, then release into atmosphere driving global weather, predictable based on current ocean temps.
11 million terawatt hours excess energy stored (Earth uses 25,000 TWh/year)—500 years human energy. Releases over next months into atmosphere—99% confidence upcoming year hottest on record.
Changes trade winds, atmospheric moisture. Highlights:
- Major atmospheric river events: Southwest California, Gulf Coast.
- Low snowfall, high heat waves Northern US to Canada (remember Canada fires).
- Interior US (Phoenix) already 106°F in May, super El Niño extends heat domes.
- Record heat Southern Argentina, Chile, Brazil—crops fail.
Brazil, India, Australia critical for local consumption and ag exports. Australia's wheat to Indonesia/Philippines (hundreds of millions depend). Brazil world's largest ag exporter.
Monsoons fail high probability in India—150M farmers, 1.5B people depend on that food.
Effects: energy/electricity prices spike, grid fails Southwest; commodity prices spike; unrest in India, Philippines, Vietnam from food shortages.
India double whammy: monsoon dependence + nitrogen fertilizer shortage from Iran/Strait of Hormuz crisis—calorie deficit, economic crisis South Asia.
Data confirms record-breaking El Niño next months.
Fucking scary. Commodity markets wild.
Last El Niño had food shortages from Australia/Brazil—regular. This year food shortage index off charts, beyond anything, correlated with severity.
US net ag exporter—issues: fire season, decreases Atlantic hurricanes (fewer South), major California storms, heat waves, electricity prices. More stable than world.
Brazil ag exports key to economy, has debt problem—could become economic issue.
Commodity markets trading El Niño actively.
Super seeds longer-term. Crop genetics adapting to changing climate critical next decades.
Macro: places heating up become viable farmland e.g. Canada, northern US. Soybeans in Canada not possible two decades ago—plant breeding + warmer temps/shorter winters expand farmlands.
Couple years ago Australia heatwaves/droughts/huge fires—wheat to Indonesia/Vietnam (hundreds of millions depend), they scramble.
Global give/take, but this year with so much atmospheric energy, may buckle.
Early warning. Food crisis warnings before—Ukraine 4 years ago led to negotiation allowing wheat flow, carve-out in war vs famine. Fertilizer exports negotiated at war start.
Anthropic unrolling dark SPVs sold to dentists for 10% load-in fees—Wild West, calling out platforms, negating them.
Tempest in teapot or long overdue? Punishment for bad actors. All companies should go public for evaluation, focus higher order bit. Ticky-tacky mechanisms to stay private longer need bullet—layered SPVs on SPVs disgrace.
Once SpaceX, Anthropic, OpenAI go public, litany of lawsuits back and
Once SpaceX goes public, once Anthropic goes public, once OpenAI goes public, there will be a litany of lawsuits back and forth between the purveyors of these SPVs. They should not be allowed and should be all negated if possible. You can't unbreak the egg, so call it what it is. Every company should adopt this, resulting in companies going public sooner and rationalizing their equity structure faster. It's the right thing to do.
Anthropic just saying. The actors are disgracing the odd. A lot of people will be very upset once the SpaceX thing comes up because it happened in Uber—somebody gets screwed in a layered SPV somewhere that they didn't know. They'll say, what happened? Read the fine print. You're getting double carry, paid 10% on the load-in fee, and paid a price 10% over what the last round was. SPVs are a recipe for disaster. The fine print matters.
Great job. Thanks so much for coming on the program. Come back any time.
Fan question: When you do your charity work or innovation as a founder, at this stage in your career, are you more Steve Jobs or more Laurene Powell? The PR department knows the answer.
The best decision when starting Salesforce: put 1% of equity, 1% of profit, and 1% of all employees' time into a foundation. Today, more than 10 million hours of volunteerism, given away more than a billion dollars in grants, and run over 50,000 nonprofits for free on the platform. That's just Salesforce. Every company should do this. Pledge1percent.org—everybody should jump on the 1-1-1 platform. Best decision 27 years ago, irrespective of personal philanthropy: five hospitals and more.
Made a big donation after Susan Wojcicki tragically died of cancer. She was a best friend, bestie, a special person, the best. One of the great people on the planet, incredible woman, mother of five amazing children, two amazing sisters, great family. Everybody knows Ann (her sister), Janet at UCSF. Contracted a very rare form of cancer, passed away too young a couple years ago. She was on the board, one of the great people of all time.
Grateful her family is focusing on building a new foundation to go after this rare cancer. We're backing it—everybody should. Everyone should try to be a Susan. She was one of the absolute greats. Has an altar in the office with photos of her, thinks about her every single day. Heart stays with her husband Dennis, all the children, the sisters, the whole Wojcicki family—amazing, one of the great institutions in Silicon Valley.
Rest in peace, Susan Wojcicki. See you all next time on the All In Podcast. Aloha.
Keep All-In Podcast in your library
Save the videos and channels worth coming back to, and find them again in one place.
