a16zAI, Infrastructure, and the Next Investment Cycle
a16z
Introducing the State of Markets Discussion 0:00
David George opens the A16Z podcast with colleagues Sarah Wang, Alex Emerman, and Santiago Rodriguez to walk through slides from the firm's yearly state of markets presentation. The discussion covers macro trends like capex and data center demand, along with on-the-ground takeaways about models, apps, and vertical deep dives. The framing question throughout is whether the current tech boom, despite its size, constitutes a bubble.
Tech Dominates the Investment Cycle 3:02
High-tech equipment, software, and research and development now account for roughly 55 percent of US capital spending, and tech makes up almost 40 percent of the total value of the US stock market. Eight of the top ten most valuable companies in the world are US tech companies, and AI model companies alone have raised over 350 billion dollars. This buildout has just surpassed railroads as a share of GDP, and the group expects the current numbers to look 20 times higher within five to ten years.
Why This Differs From the Dot-Com Bubble 5:03
Even as the market hits new highs, trading multiples are actually down about 20 percent while stocks are up about 20 percent, meaning gains come from real earnings rather than inflated valuations. The S&P 500 trades below 20 times earnings, unlike dot-com era companies that traded near 100 times earnings. Since ChatGPT launched almost four years ago, the market is up 90 percent, or 17 percent annualized, which feels aggressive but looks different when paired with earnings growth rather than multiple expansion.
Hyperscaler Spending Keeps Outpacing Forecasts 6:31
Combined capex from Alphabet, Amazon, Meta, Microsoft, and Oracle is projected at about 780 billion dollars in 2026, up from 416 billion in 2025, with expectations of surpassing a trillion dollars annually from 2027 onward. Forecasts for this spending keep getting revised upward faster than anticone expects, driven by demand that keeps outstripping supply. Sam Altman and Sarah Friar's aggressive compute commitments, once criticized, now look prescient, especially after OpenAI had to pause new subscriptions on its 2,000 dollar pro plan two weeks earlier due to overwhelming demand.
Signs of Durable Demand and Returns 10:02
Microsoft, Google, and Amazon hold a combined 1.7 trillion dollars in cloud backlog, suggesting real customer commitments even as free cash flow stays depressed during the buildout, with recovery expected from 2028 onward. Amazon's earnings call described a J-curve dynamic where GPUs and TPUs have long useful economic lives even though building data centers and acquiring chips takes time upfront. Spot market pricing for GPUs remains strong, and hyperscalers are earning attractive returns even as chip costs run much higher than they did one to two years ago, generating substantial value for both vendors and users.
A New Age of Physical Infrastructure 13:02
Global infrastructure investment needs are estimated at 90 trillion dollars through 2040, covering far more than AI and data centers, including power, water, roads, and transit. Examples across the portfolio include Anduril's manufacturing facility spanning the equivalent of 87 football fields, Waymo's expanding depots, and SpaceX's hundred billion dollar investment in Louisiana. This physical buildout requires different skills than software investing, including financing, vendor management, and capacity forecasting, and Elon Musk's idea that "the factory is the product" helps explain why founders from Tesla and SpaceX have proven especially skilled at scaling manufacturing as a competitive advantage.
Data centers helping the grid 15:00
A power grid has shared fixed costs like poles, wires, and substations. A large stable customer such as a data center helps spread those costs across more units of electricity, which is why residential rates in one example went down by 40 basis points. Meta's work in Louisiana is cited as a real case where the company partnered with the community to lower electricity costs, showing this is more than just talk.
Costs plummeting, revenue exploding 17:00
Costs for AI are falling fast, with new models making agents economical for a much wider range of work, sometimes at two orders of magnitude cheaper, echoing the Jevons paradox where cheaper resources drive more total use. Combined annualized revenue for OpenAI and Anthropic has climbed to levels that surpass the best net-new revenue estimates of the greatest software companies in history, showing both unusual scale and unusual speed.
Adoption is still early 18:31
Even though 69% of S&P 500 companies report live AI deployments, only 30% see quantifiable impact, and just 2% track results over time, showing enormous room to deepen use inside organizations. Power users are pulling far ahead of everyone else, with the top 1% of vendor spending roughly eight times the top 10%, and inside AI-native portfolio companies top users spend $7,500 to $9,000 a month versus $200 to $400 for the median user.
Real results and cheaper agents 25:31
Public companies are now reporting concrete gains: Chime cut its cost to serve by nearly 50% over four years with help from Decagon, and Shopify's AI sidekick raised the share of merchants reaching five orders within 15 days by 8%. ServiceNow has passed a billion dollars in AI annual contract value with a 9x rise in agentic deployments. Falling costs, caching, routing, and fine-tuning, as seen at Databricks, Harvey, and Eleven Labs, are making agents cheaper, faster, and more reliable, while consumer AI subscriptions remain early, at just over 2% of US households, but show unusually strong retention.
Consumer AI still tiny by scale 29:31
The speakers note that consumer AI subscription numbers look small next to giants like Amazon Prime, with over 200 million households, or Netflix, with 70 million. They argue this shows AI adoption is just getting started, even citing a view that 97 percent of AI-using households are not yet paying for it. They also compare time spent: Facebook, Instagram, TikTok, and Snap all capture 30 to 60 minutes a day, while future AI assistants may become so persistent and proactive that they actually reduce time spent on screen, making engagement harder to measure the old way.
Agents and marketplace advertising tension 33:31
New consumer AI tools like Muse and Instinct, alongside ChatGPT, are pulling queries away from traditional search, forcing marketplaces to consider their next moves. Amazon recently declined to work with Muse while Instacart agreed, reflecting how much each company's profit depends on owning the customer relationship versus gaining incremental orders. Amazon's advertising business exceeds 70 billion dollars and is highly profitable, so losing clicks to agents is risky, whereas Instacart, with grocery still under-penetrated online, could gain more orders through agents. The group debates whether this shift is a zero-sum loss for platforms or a positive-sum gain, since lower friction could increase overall consumption, and early market reaction has actually been net positive, with gains at companies like Meta outweighing marketplace losses.
Software winners, vertical AI, and private valuations 36:00
In public software, only about 30 percent of companies grow at 20 percent or more, and just a handful exceed 30 percent growth, compared to private companies where most exceed that pace easily. Cybersecurity, observability, and vertical software have held up better than horizontal applications, since AI creates new security needs while replacing some general workflows. Vertical AI companies like Harvey and Bridge and Lease are growing faster than any prior precedent in their industries. Stripe's private data shows SAS customer growth accelerating into 2026, a trend some call a renaissance. Meanwhile the top six private companies, including Anthropic, OpenAI, Databricks, Stripe, Wayland, and Revolut, are valued around 2.4 trillion dollars combined, more than all IPOs of the past decade excluding SpaceX, showing why many founders now choose to stay private longer.
Staying Private For Bigger Bets 44:00
Companies often stay private longer because they can take bigger swings with longer payback periods, away from public market scrutiny. Meta's stock dropping below 100 dollars during skepticism about its AR and VR spending is cited as a case where public markets punished long-term bets. The speakers describe the IPO as simply another financing event, and note that some companies, like Databricks or Stripe, run themselves with public-company-level discipline while private, while others, like Navan, benefit from going public because disclosed financials build enterprise and consumer trust.
Secondaries And Tender Dynamics 45:30
Private capital pools are large but can eventually be outgrown, pushing companies toward public markets. Secondaries are a major focus, with tender offers letting employees get liquidity while staying private. Notably, tender participation tracked by Carta has dropped to 58 percent, meaning employees increasingly choose not to cash out because they trust their company's trajectory. Frequent valuation resets help keep private stock attractive for hiring and give companies fresh currency for acquisitions. Secondary market discounts to last-round prices, once meaningful from 2021 through 2024, have now shrunk to nearly zero.
Where AI Opportunity Now Sits 48:00
AI-related companies now make up 86 percent of US VC deal activity in 2026, up from 65 percent in 2025, and the opportunity has broadened well beyond OpenAI and Anthropic into enterprise apps, consumer apps, semiconductors, power, and defense.
Robotics Autonomy And Beyond 49:00
Long-running consumer agents could handle unwanted tasks quickly. Robotics may eventually surpass large language models in scale, arriving three to five years later. Self-driving already works, with Uber and Lyft miles at only about 1 percent of US miles traveled, expected to grow tenfold as networks prove fourteen times safer than human drivers, with all 17 million yearly new car sales potentially autonomous within a decade. Other exciting frontiers include AI-driven drug discovery, personalized health advice, deeper enterprise AI adoption beyond coding, and American dynamism, where newer defense vendors like Anduril, Saronic, and Castellion still represent under 5 percent of military spending but are expected to grow sharply.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
