I talk about batteries and clean energy a lot. Personally managing alpha to Valhalla. Internet rando. Not financial advice. 🌊
Silicon Valley
Joined August 2015
- Tweets64.4K
- Following2K
- Followers6.4K
- Likes43.7K
$EOSE
Two key data points to confuse you all even more:
1. Shipments increased 23% quarter over quarter (~307 MWh in Q2 to ~378 MWh in Q3 by my count).
2. Line 2 contributed 17% of total battery production.
By itself, this doesn't make perfect sense.
If the batteries from Line 2 were shipped, that means Line 1 only produced 314 MWh. But if Line 1 had simply continued its average June run rate then I would expect a number like 378 MWh from that line alone.
Only a few ways to explain this:
- Line 1 had problems and does truly suck
- Line 1 was intentionally starting to ramp down
- A significant amount of battery modules produced (from Line 2 or from either of the lines) were not delivered and instead went into WIP
- Some combinations of the above
Using the few clues they give us to figure out what actually happened is always exhausting 😅. But the actual outcome has a significant impact on how fast Line 2 needs to ramp in Q4 and how long Line 1 must continue to operate.
$EOSE
Prelims released. My top thoughts:
- revenue in line with Street estimates
- Gross Margin is on track per Alessandro’s cost out plan
- WIP I suspect is a critical number which we don’t yet know
- Going from partial shift to 3 shifts in Q4 sets us up for more EOY drama; another December to Remember.
- Implied net new orders are $40.5 m. Golden Dome is part or all of that
$EOSE
Prelims released. My top thoughts:
- revenue in line with Street estimates
- Gross Margin is on track per Alessandro’s cost out plan
- WIP I suspect is a critical number which we don’t yet know
- Going from partial shift to 3 shifts in Q4 sets us up for more EOY drama; another December to Remember.
- Implied net new orders are $40.5 m. Golden Dome is part or all of that
$EOSE
Messaged some people this weekend and had a long phone conversation this morning to share notes. This has given me an opportunity to learn some new things, and also reinforce some existing beliefs:
- Indensity has backlog and might be launching in Q1; new orders today for cores can likely be fulfilled no earlier than Q3
- Waiting for the next best mouse trap (Indensity) may be hampering sales of existing cubes; that's where FPUSA comes in
- The company and Cerberus is very focused on achieving profitability first before chasing volume
- More deals like Google are in the works.
- The company is very aware of the PA, and is both horrified and confused
- I believe a pre-release may still happen.
$EOSE
So anyway, here's a thought experiment on Q3 numbers. They previously said that to make plan they would just need to continue the June run rate through to the end of the year. That was 1.5 GWh/yr on average, and in June that was almost exclusively from Line 1.
Let's say Line 1 doesn't perform as well as June, and they have problems, and only average 1.2 GWh/yr from it during Q3. That's 300 MWh produced, which is less than they even produced in Q2 (308 MWh).
And let's say Line 2 ramps from essentially 0% to 30% during Q3 (15% average), keeping in mind they want to get to get to full production by EOY. So 15% is very much lowballing it.
And let's say no service revenue comes in because they haven't officially completed commissioning on some projects that they wanted to. And whatever WIP they produced in Q2 still isn't recognized in Q3.
Between Lines 1 and 2 that's 300 (L1) + 75 (L2) or 375 MWh total. Assuming a product ASP of $220/kWh that's $82.5M right in line with expectations.
That outcome is, to me, an execution fail, imo, compared to what they should be capable of easily performing. But even in that scenario they would still be on a credible path to making year end guidance.
They have the workable backlog with FPUSA. They have the previously achieved performance metrics from June that they can just maintain. They would really have to f*k things up to miss this bigly again. After a year on the job, is Mahaz really incapable of pulling this off? Are the line workers just complete idiots/slackers who with redundant machinery in place now to mitigate downtime, and additional laser checks to ensure quality, and after a year of training -- all that and they still can't manage downtime and quality?
I'm not making any predictions. I've posted extensively how disappointed I was about how the RO was handled. And how disappointed I was to learn that Project AMAZE and Factory 2 works are very behind schedule. I always get a heap of hate from people who apparently can't even read everything I've been posting over the last 8 months, as if I predicted them to exceed things in the first place. I am not predicting that. I'm on record giving them a 40% chance of making street numbers, based on their past track record. But, I would really be at my wit's end if they can't pull Q3 off without falling on their faces. I would be satisfied to take the rest of my winnings and move on.
Imo, for Q3 they have an opportunity to really blow through these numbers. On paper it seems very straightforward. But that would be the first time in Eos history where they company has actually done that.
I would be very pleased to see $90M revs and -50% gross margin as solid, objective evidence they are on track to achieving the cost out plan recently presented by the new CEO. I am mentally prepared for much less.
Reasonably Approximating 🇺🇸 🇺🇦 🔋 🅰️ retweeted
I spend a lot of time in the private markets, and the recent madness around financing challenges for DCs and neoclouds is pure FUD.
Yes, there is a shortage of traditional lender capital, but many pristine balance sheets in Silicon Valley, elsewhere, and across sectors have strong incentives to participate and fill that void.
AI is a moment when everything as we knew it changes.
Every single business…
in every single sector…
across every P&L line item…
will be profoundly impacted…
by this new god-like capability.
Every balance sheet in the world will contribute to AI development over time, and AI datacenters benefit because of their lock on power.
So why the debt scare? The datacenter/neocloud market is being crowded out by public sector debt issuance. If you agree with my earlier proclamation about AI affecting everything, private sector enterprises will contribute their balance sheets to AI development in two ways:
1) As the next source of debt/equity financing for DC developers and compute providers (probably not voluntarily), thereby lowering DC cost of capital;
and/or
2) By consuming tokens far faster than supply, raising $/GPUhr for compute and lease rates for DCs.
Path 1, lowing the cost of capital, seems logical, but massive public sector debt woes will keep financing costs elevated for a long time, making path 2 equally a necessity:
Higher debt/equity rates are offset by higher DC lease rates.
Higher lease rates are offset by higher $/GPUhr or a larger share of those $/GPUhr economics.
In an expanding TAM, prices rise quickly as demand outstrips supply. If AI’s TAM is expanding, $/GPUhr should be high, right? Sort of. It’s a tale of two cities: the haves and the have-nots.
Most of today’s AI economics (outside GPU makers) go to the incumbents: hyperscalers. They have large, highly creditworthy balance sheets from decades of capital-light models, diverse and growing revenue, and strong margins, serving the world’s largest and successful enterprises. Hyperscalers are the most trusted providers of enterprise-grade/scale compute—an expertise built over decades.
Neoclouds, on the other hand, must build a balance sheet from scratch with creative financing, only to serve the hyperscalers’ leftover TAM, and all for a new commodity (compute tokens) whose value is not yet established.
No wonder neoclouds live on a razor’s edge: supernormal revenue growth fueled by equal or greater capex, hoping commodity prices soon make investors very happy. I call it “a dream business (to short).”
But wait—there’s more.
In a nascent, rapidly evolving commodity market, large pricing inefficiencies arise from swings between oversupply - think $CRWV in 2024–2025 bringing large compute online for what was only a handful of large buyers at the time - and undersupply when many buyers enter at (what feels like) all at once (Jevons’ paradox accelerating). Buyers and sellers are constantly left wonder g who holds the negotiating power in these conversations.
Things eventually stabilize, with better matching of land, power, talent, construction, procurement, GPUs, compute, financing, buyers/sellers of stabilized assets, used GPUs, and more.
Then $/GPUhr accelerates (as long as TAM grows). The most recent compute pricing charts suggest we might be there now.
If so, neoclouds are here to stay. The strong ones will grow fast; the weak get acquired and their old contracts renegotiated to much higher current rates.
A much healthier ecosystem overall.
But remember this:
Be it the hyperscalers of the neoclouds, the DC developer always makes bank!
$NUAI $IREN
Reasonably Approximating 🇺🇸 🇺🇦 🔋 🅰️ retweeted
Quoting from Anthropic's Michael Sellitto at the IMF Artificial Intelligence and Escalating Cyber Risks panel today. This is the first time Anthropic has given a projection for what percentage of frontier model training will be done by Claude in 2027. In March this number was 1%. By next year they are projecting 80-90%.
'Last year, about zero percent of our code was written by our model Claude. By August of that year, about 50% of the code inside the company was written by the model. That's because we released a new model in August of last year, and the engineers said, ‘Actually, this is the first time that Claude is writing code at my level.’ Anthropic only hires the best software engineers in the world, and it was at that point that Claude finally got to their level. In August we were at 50%. By the end of last year we were at 80–90% of the code in the company being written by the model.
What percentage of training the next-generation models is Claude leading today? This is following a similar step change. At the beginning of the year it was zero percent. In April it was 3%. In August, 26% of the process of building the next generation of models is being led by Claude. And we can project out - and we had a blog post about this not too long ago - that sometime next year it could be 80–90%.'
That new Tesla Roadster is simply awesome.
Even if I did plonk down the money for it, I doubt I would get that actual thing that we saw today in 2028.
But, even if I did, I doubt the government would allow me to drive it around town.
Consensus revenue numbers for $EOSE are $81-85 million. Where do you think they'll land?
44%< $81m
24%consensus
31%> $85m
143 votes • Final results Reasonably Approximating 🇺🇸 🇺🇦 🔋 🅰️ 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.
I am once again asking you to do the math.
Market caps:
$OKLO: $5.91B
$SMR: $2.79B
$EOSE: $0.87B
2026 full year revenue consensus:
$OKLO: $3.8M
$SMR: $24.5M
$EOSE: $305M
Down YTD:
$SMR: 49%
$OKLO: 56%
$EOSE: 79%
Projected profitability per management:
$SMR: early 2030s
$OKLO: 2029
$EOSE: 2027
We made too much money last year and being made examples of from hedge funds coordinating against us and trying to wreck our spirit.
Use your brains and act, people. I am personally buying $3 2028 leaps.
Reasonably Approximating 🇺🇸 🇺🇦 🔋 🅰️ retweeted
Billions of dollars’ worth of contracts predicting oil would fall and the S&P 500 would rise were placed about 15 minutes before Trump announced a U-turn on Iran in a post on Truth Social. CFTC staff began to investigate-- then were told to lay off. bloomberg.com/news/features/…
Happy anniversary!
October 14, 2025 was in my estimation the combined top of high beta growth and crypto. Leveraged crypto, a leading indicator, then got liquidated, followed by beta, and it was the beginning of a year long slide in many successful names. We are coming up on the one year anniversary of that. Time flies when you’re having fun.
Like crypto earlier this year, things always seem to get much worse than you think is possible before it gets better.
Market caps:
$OKLO: $5.91B
$SMR: $2.79B
$EOSE: $0.87B
2026 full year revenue consensus:
$OKLO: $3.8M
$SMR: $24.5M
$EOSE: $305M
Down YTD:
$SMR: 49%
$OKLO: 56%
$EOSE: 79%
Projected profitability per management:
$SMR: early 2030s
$OKLO: 2029
$EOSE: 2027
We made too much money last year and being made examples of from hedge funds coordinating against us and trying to wreck our spirit.
Use your brains and act, people. I am personally buying $3 2028 leaps.
Reasonably Approximating 🇺🇸 🇺🇦 🔋 🅰️ retweeted
The man arrested for planting a secret recording device at OpenAI's headquarters is a Venture Partner of a VC firm alongside the son of a senior Saudi intelligence official:
