AI Infra Finance • DC Builder • Structured Credit/CRE/CMBS • ex-MS & DB. 100x Agency to Scale... Not Financial Advice.
Global
Joined December 2021
- Tweets2.3K
- Following939
- Followers2.5K
- Likes3.7K
Cole retweeted
Citizens: Neoclouds
"CoreWeave has the largest contracted platform and the most future capacity to reprice; Nebius already shows the rate reset in agreements signed at $20–$25 million/MW, with customers prepaying 50–60% of the associated capex; and IREN shows the progression from roughly $10 million/MW on its original Microsoft agreement to above $20 million on newer contracts."
"We are more positive on the Neocloud business model in the near-to-intermediate term. The equities and credit have repriced around the capital required to deliver the build, but demand continues to exceed available supply, delivery remains back-end loaded, and the newest contracts are being signed at higher rates."
$NBIS $IREN $CRWV
Cole retweeted
Revenue per GPU can increase as token prices decline if per GPU token throughput increases enough.
So both Anthropic revenue *and* margins can go up as token prices go down *if* GPU prices go up less than the product of (token price decline * per GPU throughput increase).
But I have super enjoyed all the commentary to the contrary!
Cole retweeted
As someone with a portfolio of energized land assets here is what I am seeing:
1. Neoclouds are disorganized and poorly capitalized. Some are definitely better than others however.
2. Hyperscalers are intractable, bureaucratic organizations that take 3 months to do an NDA and want 18mo escrow periods. We built our portfolio by competing against these terms because of our speed and simplicity. The more future assets orgs like ours win, the bigger the problem for the hyperscalers downstream (see 3).
3. There are delays in the existing pipeline that will catch people off guard in 2027. Hyperacalers will (obviously) react by focusing on their own internal compute needs over any third party. If the third party is a model company, they will tell them to go through their model router where they will also take a non-trivial vig. Them telling Ant they have no compute for them (rumored in Ant’s S-1) if true is a huge tell that this is already happening.
All of this is to say that it does look like there will be a lot of partially stranded compute and compute sitting in inventory in 2027 than is currently forecasted. This is bought and paid for but effectively underearning (below total cost of capital) or collecting dust. We see this as true for everyone - model companies, hyperscalers, neoclouds except ONE (you can guess which one).
This dislocation will benefit two kinds of orgs: the non-bureaucratic big scalers who can build industrial scale quickly and the nimble, well capitalized players feeding them.
2027 will be a turbulent year because it will clear out a bunch of wrong/bad bets. It will also be a consolidation year for a few.
Cole retweeted
Renting Nvidia’s H100 costs 45% more than it did in December.
It went on sale in 2022, and two newer generations have come out since.
Yet, demand for compute is so far ahead of supply that older GPUs are getting more expensive.
Cole retweeted
Zooming out, GPUs are still renting ahead of a standard 6-year depreciation schedule
Rental price vs. schedule:
- H200: +143%
- B200: +111%
- RTX PRO 6000: +73%
- H100: +54%
- A100: +9%
Most of the gap opened up after April, when token demand took off
Charts of the Week: a16z.news/p/charts-of-the-we…
GPUs are aging like fine wine, with rental rates up across the board since January 1st
Approximate change YTD:
- H200: +160%
- B200: +120%
- RTX PRO 6000: +65%
- H100: +35%
Even the 6-year-old A100 rents for more than it did at the start of the year
Charts of the Week: a16z.news/p/charts-of-the-we…
Cole retweeted
GPUs are aging like fine wine, with rental rates up across the board since January 1st
Approximate change YTD:
- H200: +160%
- B200: +120%
- RTX PRO 6000: +65%
- H100: +35%
Even the 6-year-old A100 rents for more than it did at the start of the year
Charts of the Week: a16z.news/p/charts-of-the-we…
Cole retweeted
At @OrnnExchange, we've always been focused on incredible people, innovative product and compressing timelines.
We are taking the next steps to push AI compute markets forward and this team has always been by our side.
Excited to continue building alongside @cdixon and the @a16zfrontier team!
Cole retweeted
$NBIS SECURES $3.5 BILLION FINANCING FACILITY FROM CITI
Nebius announced a new senior secured credit facility to fund its global AI data center expansion and GPU purchases.
The facility has a 364-day maturity, with interest starting at SOFR + 2.75% and potentially rising to SOFR + 3.50%.
Nebius has more than $40 billion in long-term contracts with investment-grade customers, which could support additional asset-backed financing as it expands AI capacity in 2027 and beyond.
Asked how far $NVDA would go to financially backstop neoclouds, or even bail one out, Jensen said:
“We invested in them... that's different than backing them”
He says neoclouds are “completely sold out” & expects them to secure the working capital they need
h/t @TomMackenzieTV
Cole retweeted
Public investors see the supply side of AI (the chips and the hyperscalers), and little of the demand side, because the demand lives in private companies (as my colleague @jrichlive often says).
Since April, token usage on @Vercel's AI Gateway has grown almost 8x, and spend more than 3x’d.
Prices fell by more than half during that time, and developers bought nearly eight times as much. In 30 years, I haven't seen demand like this.
Cole retweeted
SCOOP: Fireworks AI is seeking to raise fresh funding at a valuation between $30 billion and $40 billion, a dramatic jump just months after the AI-infrastructure startup was valued at $17.5 billion.
businessinsider.com/firework…
Cole retweeted
$META +3% AH.
>Revenue $64.6B beat (guide was $61-64B)
>FY capex floor raised: $135-145B vs $130-145B, midpoint $140B.
>Q4 math: YTD capex (incl. finance leases) is $85.8B
>Raised capex guidance implies $49-59Bin Q4, ~$54B at midpoint, +55% q/q.
"Our AI investments are already delivering substantial incremental advertising revenue through stronger engagement, better recommendations, and more efficient monetization."
$META reports after the close. We think 3Q26 earnings could be positive read through for $CLSK & $APLD (among all PSPs and neos)
- Look for Muse monetization and potential compute expansions to support it
- FY 2026 capex guide is currently $130-145B
- FY 2027 consensus capex guide is $193B
Muse early adoption looks strong with 10.9M downloads and ~2.5M DAUs, much faster growth than other launches (although caveats being Muse launched on IOS and Android at once while ChatGPT, Claude and Grok were initially IOS only, and it was heavily promoted taking up most of Metas house-ad impressions). DAU durability is an open question due to this heavy marketing push.
Lot of focus on Muse's monetization path and compute needed to support it, which is relevant for $CLSK which already has a 20-year lease with Meta at Sandersville (175MWs), and a LOI/exclusivity for 885MWs in its TX portfolio. Commentary supporting more 3P leases of capacity would be positive and would support lease execution of this additional power. Would be good to hear that Meta is back focused on utilizing its compute in-house rather than re-selling excess.
Meta is also expected to come to debt markets shortly after earnings to fund capex plans, a smooth process could support their ability to keep expanding capacity via 3P leases (CLSK).
It is also believed that APLD's two newest leases are with Meta (one unnamed hyperscaler), 300MW in Louisiana and 300MWs in North Dakota. Although this is unconfirmed. Also a potential beneficiary around commentary from Meta around wanting more 3P compute.
Cole retweeted
Inference is a beautiful business.
deepseek reports 155T tokens/day from one inference cluster at $38/gpu-hr. I estimate their cluster at ~5 MW, implying ~$160M/MW/year in API revenue and 84–92% margins after $3–6/GPU-hour rent
working assumptions: 96% input-cache hits, ~107 input token/output token ratio -> ~$14.35 revenue per billion tokens at off-peak prices. working backwards from 155T tok/day, 60,000 uncached input tok/sec/gpu prefill + 14,000 output tok/sec/gpu decode, each two-GPU pair handles approximately 1.5m input + 14k output tokens/sec -> 2400 gpus. on blackwell hardware I assume 2 kW/gpu including external networking, cache storage, etc.
No wonder everyone wants be part of the token flow / bid up GPUs :)
🚀 Introducing DeepInfer, our in-house inference framework, now fully open source!
Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable.
🔗 github.com/deepseek-ai/DeepI…
1/4
Cole retweeted
Our $DESKB300 sent to the moon due to open source inference demand. @ComputeDesk
Cole retweeted
Blackwell Ultra (B300) GPU rental prices up 17 consecutive days to a new all-time high, according to Compute Desk's index of verified transactions from U.S. neoclouds.
Cole retweeted
Open-weight models that decrease margins at the model layer are positive for AI infra, exhibit 1001.2
Since the @silicondata “token index,” which captures both price and mix, peaked on May 31:
B300 prices are up 38%
B200 prices are up 10%
Other GPU indices are up *much* more.
Cole retweeted
Old GPUs are turning into appreciating assets
More charts in State of Markets II: a16z.news/p/state-of-markets…
Introducing our State of Markets pt 2, along with a companion podcast where we unpack the data and discuss what comes next.
Tech is the everything cycle.
Supply: putting the buildout in context, just passed railroads as % of GDP. The wisdom of Elon is real: the factory (or the datacenter!) is the product.
Demand: diffusion is so, so early. Median AI vendor spending in the top 1% of companies is 8x that of the top 10%. Only about 30% of S&P 500 companies report a quantified AI impact, which means there’s a substantial opportunity in connecting models to a company's data and workflows. Diffusion into companies is one of the main themes of the next 5 years.
We’re entering the agent work period. Only a few million users today, but applicable to billions of internet users with massive surplus created. META/GOOG monetize US users at $200+ per year today. Agent opportunity is much higher.
Mega-trends the next 5 years: Consumer agents, Robotics, Autonomy, AI x bio, Personal health, Diffusion into enterprise, New era of American Dynamism.
Much more in our SoM report here - a16z.news/p/state-of-markets…
@a16z @sarahdingwang @aleximm @santiago__rdz
Cole retweeted
Older NVIDIA GPUs can remain productive and revenue-generating long after newer generations arrive.
@OrnnExchange's analysis of NVIDIA A100 rental pricing offers third-party evidence of a multi-year earning life for prior-generation NVIDIA infrastructure.
CUDA’s support for diverse workloads, combined with continuous software optimization across generations, helps keep that hardware productive longer.
Learn more: nvda.ws/4yJKOe9
