Aug 31, 2026 · 1h 14m · a16z
Why AI Demand Is Outrunning Compute Supply
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this in-depth discussion on The a16z Show, investor Gavin Baker and host David examine why surging enterprise AI demand is rapidly outpacing compute supply, analyzing data center unit economics, NVIDIA's hardware supremacy, and the emerging viability of orbital computing with SpaceX.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The host as informed peer 6.5 Guest teaching 2.5 Guest disagreement 1.6 The host pushing back 1.4
speaking balance: gold is the host, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 32:10 Gavin dismisses misconceptions on open source token costs
Gavin forcefully rejects the common narrative that open source tokens are free, explaining they require identical compute and highlighting licensing revenue share models.
Hardest push from the host ▶ 10:05 David George pushes back on lab profit assumptionsDavid George interrupts Gavin's profit reinvestment thesis to point out that frontier labs are currently heavily subsidizing token consumption on their first-party products.
Biggest teaching moment ▶ 37:15 Gavin explains orbital compute unit economicsGavin educates listeners on orbital data centers, breaking down the 50 billion dollar per gigawatt math and showing how reusable launch costs beat terrestrial inflationary cooling and power.
The host holds their own ▶ 15:05 David George reveals proprietary a16z enterprise token metricsDavid George displays authoritative expertise by citing granular data from the a16z portfolio on token spend ratios compared to human compensation in legacy versus AI-native enterprises.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The host as informed peer | Guest teaching | Guest disagreement | The host pushing back | Why |
|---|---|---|---|---|---|---|
| Episode Preview and The a16z Show Title Sequence | 5 | 1 | 1 | 1 | David George sets up the episode discussing public market tech drawdowns versus accelerating private AI developments. Gavin Baker notes that despite public market volatility, no enterprise operator he has spoken with reports worsening business metrics. | |
| Frontier Lab Competition, Accounting True-Ups, and Mission Alignment | 6 | 2 | 1 | 1 | The conversation explores Anthropic and OpenAI revenue reporting and IPO readiness. David George actively complements Gavin's points regarding revenue added definitions and mission-driven cultural alignment. | |
| Compute Allocation Dilemmas and the Non-Zero-Sum Ecosystem | 6 | 2 | 1 | 1 | Gavin and David discuss the capital allocation dilemma frontier labs face when trading off training compute versus inference revenue. David George emphasizes how this represents an additive 'and' market rather than zero-sum competition. | |
| Infrastructure Payback Cycles, Lab Subsidies, and Private Financing | 7 | 3 | 2 | 2 | David George intervenes to highlight first-party product subsidies and presses on payback calculations. Gavin lays out neocloud unit economics and private credit debt structures backing GPU deployments. | |
| Demand-Side Diffusion, Enterprise Token Spend, and Agentic Automation | 8 | 2 | 2 | 2 | David George cites specific internal a16z portfolio data regarding token consumption as a percentage of human compensation across tech-forward firms and banks. Gavin matches with Atreides' internal consumption scaling. | |
| Historical Market Bubbles, Physical Constraints, and Data Center Revitalization | 7 | 2 | 2 | 1 | Gavin draws parallels to historical market bubbles and physical bottleneck constraints like copper and power. David George reinforces the point by citing data from Loudoun County, Virginia regarding data center tax benefits. | |
| Re-Industrializing America, Severe Compute Shortages, and Open Source Nuances | 7 | 3 | 2 | 2 | David and Gavin explore the risks of compute undersupply leading to token price increases. Gavin clarifies the distinction between open weights and true open source, noting Kimi's 30 percent revenue cut requirement. | |
| The Feasibility and Unit Economics of Orbital Data Centers | 6 | 4 | 2 | 2 | David George frames the physics versus economic constraints of orbital data centers. Gavin breaks down the capital expenditure comparison between terrestrial power/cooling and reusable Starship launch economics. | |
| SpaceX Infrastructure, Asteroid Mining, and Mars Colonization | 5 | 3 | 1 | 1 | David asks Gavin for his most futuristic long-term SpaceX thesis. Gavin details asteroid mining of precious metals on Asteroid Psyche and humanoid Optimus deployments on Mars. | |
| Microsoft's AI Strategy and the Battle for the Enterprise Abstraction Layer | 8 | 2 | 2 | 2 | David George and Gavin Baker dissect the multi-model enterprise router architecture and the high-stakes battle to own the corporate abstraction layer between hyperscalers, Databricks, and application startups. | |
| NVIDIA's Ecosystem Strategy, Data Center Financing, and Custom Silicon | 7 | 4 | 2 | 1 | Gavin details NVIDIA's comprehensive nine-chip stack and residual value guarantee financing frameworks. David George and Gavin discuss why building custom ASICs is fraught with hardware risk compared to plugging into the NVIDIA ecosystem. |
Statements from this episode (30)
Tech Founders Reported No Deteriorating Business Metrics in July and August
“My standard question is, can you tell me one quantitative data point in your business that's getting worse? Just one. That's my standard question, and it's at least in July and August, I haven't been able to find a single person.”
AI Adoption Accelerated Across OpenAI, Open Source, and Grok in Summer
“Now, if we're being honest, you know, Anthropic is you know, in a quiet period, so maybe they've slowed down a little bit, but I do think the rest of the world has accelerated, you know, OpenAI is clearly accelerated. Open source, I think, has accelerated more…”
Anthropic Cleaned Up Accounting to Make Revenue Comparable to OpenAI
“If you're anthropic, one, I think they probably trued up and cleaned up some accounting. You would, you'd rather do that. So you rebased and now you're comparable to open AI.”
Anthropic's Next Financial Disclosure Will Likely Show Re-Accelerating Revenue Growth
“And then, you know, I would hypothesize, because they've executed well, probably the next disclosure is a re-acceleration, and then there's always this kind of funny game between the frontier model companies.”
Anthropic Asks Candidates If They Are Fine With Equity Going to Zero
“Anthropic is now in their culture interviews saying, how would you feel if the equity went to zero? Because we're looking for people who are mission aligned.”
Frontier AI Lab Revenue Is Determined by Internal Compute Allocation
“A lot of the revenue is kind of under their control based on what checkpoint they release. Yeah. Where they price along this, you know, kind of Pareto curve, and then how much they allocate between training and inference.”
The AI Market Will Be Non-Zero-Sum Across Labs, Open Source, and Apps
“Like, this is not an or thing. It's an and thing, right? Like, this is an and thing. Frontier's gonna work really well. Like, N minus one models are gonna work really well. Open source is gonna work really well. There's gonna be a bunch of application companie…”
Frontier AI Labs Will Not Generate Free Cash Flow Anytime Soon
“They, I, for sure, I don't think they will generate free cash flow anytime soon. I think they're gonna generate a lot of operating cash flow, and then they'll use that to buy a lot of, you know, GPUs XPUs, whatever, whatever I'm gonna call them.”
Nebius and CoreWeave Compute Payback Periods Are Nine to Ten Months
“We calculate, you know, Nebius and Corweave both gave some interesting disclosures. But you can kind of get to a nine to 10 month payback for Nebius”
SpaceX Achieves Sub-10-Month Compute Payback on Large Rapidly Built Clusters
“And then SpaceX, because they build these really big clusters, and I think a really important point is they bring them on fast. They have an even faster payback, and they can monetize it, you know, higher”
Private Credit Finances NVIDIA GPU Purchases at Low Cost of Capital
“Particularly if you're buying NVIDIA GPUs to a lesser extent TPUs, you can finance these. And there's a very sophisticated, you know. Yeah, it's a very low cost of capital to finance them today.”
Top a16z AI Startups Spend 10% of Human Compensation on Tokens
“Across the A-sixteen Z portfolio... Best companies spending on tokens per month relative to human compensation... High single digits, some at 10%, like some of the very AI native ones, like 10% plus. And so, you know, and then old economy companies are spendin…”
Atreides Management's Internal AI Token Spend Grew 100x in Five Months
“At Atreides, our internal token consumption has gone up a hundred X from the month of March, March through August, 100 X. Yeah. Our token spend.”
Top a16z Portfolio Engineers Now Write Over 90% of Code With AI
“Our most sophisticated engineers, you know, were doing whatever, 20% of their code, you know, with AI to like, you know, whatever, 90 plus.”
Operating Cash Flow Funds the Majority of Current AI Infrastructure Buildouts
“Even today, a majority of this is still being funded out of operating cashflow, which I think is really helpful.”
Behind-The-Meter Data Centers Can Increase Local Town Tax Revenue 10x
“Particularly with behind the meter power generation, when a data center goes in, It transforms a town like tax revenue. It doesn't double it like 10 X's and it is revitalizing all of these like dying small towns all over America.”
CCP Launders Anti-Data Center Campaigns Through TikTok, Asserts Gavin Baker
“Like there is an organized CCP funded campaign. I think against data centers here in America. Like I think a lot of it gets laundered through TikTok and it's just tragic because the other thing that's happening is this is re-industrializing America.”
AI Compute Capacity Will Remain Massively Undersupplied Through 2028
“So it seems more likely than given that fact pattern, if you go back to just the sort of macro situation that we're in, that we're, we under build on the supply side. For like through 28. And by the way, like there's no capacity available with all the forecast…”
Open-Source Models Require the Same Compute as Equivalent Frontier Models
“It takes the exact same amount of compute. All else equal to make an open source token as a, you know, frontier token for a comparably sized model. Now there's a lot of nuances there, but that's broadly true. It's just a question of what are the margins that a…”
Kimi Model License Demands 30% Revenue Share on Downstream Usage
“And even then the Kimi license, something that I don't think a lot of people appreciate is the Kimi license stipulates a 30% share of any revenue. So, like, Kimi has taken a 30% cut of all the revenue generated on its, and this is because it's open weights, no…”
All Incremental Global Internet Capacity Will Come From Space
“And any incremental internet capacity, like, is not going to be built in a traditional sense on, it's going to come from space.”
Starship Reusability Drops Orbital Compute Launch Costs Under $1B per Gigawatt
“What you have to compare it to is the cost of launch, and with Starship reusability, that goes to under a billion, so the economics just instantly flip.”
Musk and Huang Co-Designed an Orbital NVIDIA Rubin Rack for 2027
“Elon said that he and Jensen have co-designed a Reuben rack. And there, it's gonna launch in the fourth quarter of 27.”
Microsoft Failed to Build Competitive In-House Frontier AI Models
“They clearly tried to make a frontier model. They failed. You know, Satya said, we're gonna have our own models that are very competitive. Like, I think he said that 18 months ago. They don't have their own models that are competitive.”
American Open-Source AI Led by NVIDIA Will Reach Near-Frontier Capabilities
“I do think you're going to see American open source led by Nvidia get really close to the frontier.”
Kirkland & Ellis Committed $500M to Build Internal AI
“Kirk Glenn and Ellis said, we're gonna spend five hundred million bucks to build this ourselves.”
Enterprise AI Providers Must Vertically Integrate Compute to Be Low-Cost
“You're just simply not going to be the low cost provider. If you're not vertically integrated, if you don't own your own compute over the very long. Long-term.”
Every One Percent Market Share in AI Accelerators Is Worth $100 Billion
“My rule of thumb for accelerators, every one percent share today is probably worth a hundred billion. So there's no need to go head on with NVIDIA.”