Oct 28, 2026 · 1h 4m
GPU rental rates defy the AI commoditization narrative
Why Compute Prices Are Still Rising | Steve Hou of Silicon Data on GPU Rental Indices, Fed Meeting, and Anthropic S-1 (LIVE)
The episode tests whether strong compute pricing reflects durable AI demand or a temporary profit peak vulnerable to depreciation, financing costs, and model commoditization.
- 1Rising rental rates for both new and older GPUs point to constrained supply and sustained demand, despite falling token prices.
- 2The central investment dispute is future AI usage and returns, not whether data centers currently generate positive operating profits.
- 3Anthropic’s adoption, open models, and semiconductor consolidation frame competing views on pricing power and durable AI infrastructure advantages.
Don't miss
Steve Hou explains why rental rates for older GPUs remain strong alongside newer chips, challenging the assumption that AI hardware rapidly commoditizes.
The brief
Steve Hou of Silicon Data opens with a puzzle: data centers and GPU rentals can be profitable while investors still question whether those profits survive depreciation and heavy capital needs.
Falling token prices support two opposing readings—broader AI adoption or accelerating commoditization. The discussion turns to whether rising usage can outrun declining prices and preserve attractive economics.
Hou’s rental-curve data shows elevated rates across H100, Blackwell, and older chips, pointing to demand and constraints in power, land, construction, and equipment supply.
The market’s skepticism appears in weak neocloud stocks and high-yield spreads, even as hyperscalers expand capacity. The key question is whether future returns justify today’s financing and investment.
Anthropic’s enterprise adoption, open and Chinese models, and switching costs sharpen the debate over whether intelligence becomes a commodity or remains embedded in reliable workflows.
A closing discussion of potential semiconductor consolidation and strategic interest in optics broadens the thesis: AI infrastructure scarcity may extend beyond GPUs into the supply chain.
What was said on this episode
31 statements · 15 positive · 9 negative · 3 mixed · 4 neutral
Existing data-center activity currently has positive margins.
“when you ask the question of whether or not data centers are profitable, it's a bit like looking at this rolling profitability of the labs. On the one hand, the existing activity, the margins are all positive.”
Listen at 2:19
Falling token prices can indicate healthy AI proliferation and increased usage.
“this is exactly what AI proliferation should look like. and prices of tokens coming down and people using the cheaper models to do a lot more stuff”
Listen at 4:36
GPU rental rates are currently surging.
“the rental rates of gpus are surging”
Listen at 4:57
Users are substituting cheaper open models for expensive proprietary models.
“there is still a, within the sample that we're observing, you know, a substitution away from expensive proprietary models towards the cheaper, the open models”
Listen at 6:48
Anthropic models have achieved substantial enterprise adoption.
“the S1 is showing an extraordinary amount of enterprise adoption of Enthropic models”
Listen at 7:03
Hyperscalers are unlikely to keep raising CapEx without stock-market rewards.
“It's hard to imagine that they will keep raising capex if they're not seeing that activity being rewarded by the stock market.”
Listen at 9:06
Hyperscalers will likely exercise greater CapEx caution or discipline.
“there's going to be, you know, some degree of caution or discipline, you know, on the side”
Listen at 10:55
Rising on-demand GPU rental rates indicate a GPU shortage.
“GPU rental rates on demand is going up, right? And that speaks to shortage”
Listen at 11:34
Current demand for GPU rentals is extremely strong.
“at the moment all we can see is that uh uh demand for uh gpu rental is extremely strong”
Listen at 12:34
Power and land availability constrain data-center expansion.
“the constraint is on the physical side of actually, can you actually find that power and land and whatnot?”
Listen at 14:04
Data-center component costs are experiencing extraordinary inflation.
“the underlying components are seeing extraordinary inflation”
Listen at 15:35
Interest rates alone are unlikely to stop AI investment.
“I actually don't really think the interest rate itself is going to stop anyone here per se.”
Listen at 20:03
AI-driven demand and inflation could make the Fed an enemy of AI investment.
“if the fact really comes to see that the ai economy to the extent that is strong and is holding up the economy and creating another level of aggregate demand um and creating inflation that it becomes directly an enemy”
Listen at 20:36
Bond markets have not believed AI would produce miraculous productivity.
“the bond market was never AI-pilled, much less AGI-pilled”
Listen at 22:24
Low valuations reflect expectations that memory and compute earnings are unsustainable.
“if people believe that the earnings cannot be sustained, that eventually this is going to come to an end”
Listen at 23:47
An upward-sloping compute rental curve indicates rising future rental rates.
“the upper sloping curve is actually indicative of expectations of rising future rental rates”
Listen at 33:05
Observed GPU residual values exceed accelerated-depreciation assumptions.
“this is actually all lying well above what would have been implied right by such depreciation schedule”
Listen at 35:06
GPU depreciation is occurring more slowly than some expected.
“depreciation is happening, just maybe not as quickly as some people have thought”
Listen at 37:21
Newer chips currently produce tokens that sell above production cost.
“if in fact you know newer chips are so much better at producing tokens that uh you're certainly producing you know those tokens are certainly selling well above cost”
Listen at 41:07
Token-serving margins are currently positive.
“margins for serving tokens is very much positive”
Listen at 42:10
Anthropic has strong enterprise adoption and customer retention.
“they have very strong enterprise adoption, and they have very strong retention”
Listen at 47:06
Anthropic may face internal cannibalization if Opus 5.5 dominates usage.
“they do suffer the potential risk that Opus 5.5 is just so good that nobody ever thinks about using Fable”
Listen at 48:01
Low current AI spending can support either bullish or bearish forecasts.
“you can read it as a bullish statement. You can read it as kind of a bearish fact.”
Listen at 49:05
Convenience and low prices reduce incentives to switch from American AI models.
“there is a certain degree of inconvenience, right? It's already so easy for you to click a button and it's not that expensive for you to be using the American models.”
Listen at 51:03
User experience and ecosystem integration create technology switching costs.
“It is the user experience, right? It is the seamless experience of being able to use your MacBook, your iPhone, be able to text your friends with blue bubbles and whatnot.”
Listen at 52:29
AI agents can make users largely indifferent to the underlying model.
“agents actually are making you almost model agnostic”
Listen at 53:47
AI systems may develop a moat through accumulated user context and personalization.
“there seems to be some mode i think the mode lies in the fact that over time your ai is going to know get to know you better”
Listen at 54:47
Large-scale cost savings could motivate enterprises to switch AI models, but scalability is uncertain.
“if the economics is there uh especially if you use a such large scale uh the question the question is just you know Does that actually really scale out?”
Listen at 55:45
Semiconductor supply-chain concentration has increased pricing power.
“the pricing power and of the semiconductor supply chain is is higher and is more, you know, oligopolized”
Listen at 58:57
EUV, etch, and deposition markets are highly concentrated among few suppliers.
“there's one company that does EUV. There's only three companies really that do etch and deposition”
Listen at 59:31
Marvell’s reported Lumentum interest signals confidence in optics’ future.
“That shows you somebody having conviction in the future being optics”
Listen at 1:01:55
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.
Books & mentions
Some links are affiliate links — PodLume may earn a commission if you buy.
Weekly email
State of the Pod, every Monday
The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.