Muse is obviously very cool. But the idea that OpenAI should "exit consumer" because of Muse has to be one of the most ridiculous things I've ever read on X, which is saying a lot. By that same logic, OAI should have exited consumer because of Gemini (lol) last year. And should have exited enterprise because of Anthropic earlier this year.. I mean Nat Friedman quite literally said that Muse is "heavily inspired" by Openclaw.. Guess where the founder of OpenClaw works these days?

$50B for what is essentially a single-site, single customer data center development project (with zero MW operational) was a pretty absurd valuation to target.. If anything, pretty healthy sign that there wasn't much demand at that price. Just look at how Fermi's $FRMI done since listing, lol.

OpenAI’s forecasted $278B of cumulative cash burn from 2026-2030 looks absurd on the surface.. but when you evaluate their funding requirements with some additional context things don’t look nearly so dire. They started 2026 with $40B of cash (Sarah Friar confirmed as such on Squawk Box at Davos in January). And then raised $122B in March.. So the forecast implies a $116B funding need. Investors reportedly “approached OpenAI” about another pre-IPO round which suggests the existing shareholders are happy to step up with more capital. I don’t think it’s unreasonable to believe they can raise $150B across another private round and IPO in aggregate. Which more than plugs the $116B hole. It’s also unclear if part of this funding gap can be addressed via NVIDIA’s $500B financing platform with Blackstone, Goldman, etc… NVIDIA strongly implied that much of this capital was intended for frontier labs on its latest earnings call (mirroring the Broadcom, Blackstone, and Apollo structure to finance Anthropic’s TPU purchases).. “The frontier labs have enormous demand for training and inference compute, but they are growing faster than what their balance sheets and credit profiles can support.. To support the frontier labs infrastructure build-out, we recently announced partnerships with 6 of the world’s leading infrastructure capital providers.. to establish financing platforms that will raise over $500B of third-party capital.” So it’s plausible NVIDIA expects to direct capital from this platform towards plugging OAI’s reported funding gap. The $278B cumulative cash burn number is also reported to be an improvement from what was expected in May ($305B). So it’s possible that with more momentum across Astra, Codex etc. the projected burn continues to come down (as reported financials come in better than expected). It’s also possible OAI is deliberately presenting aggressive levels of spend to justify another enormous raise only 6 months after the $122B in March. (Dylan Patel even alluded to them being cash flow breakeven / positive in Q3 on Dwarkesh a few weeks ago.) Now whether or not all of this investment is a good idea is a different question.. but I think the idea that OAI is almost certain to fail to meet its commitments based on Friday’s reported cash burn projection is a bit misguided. $CRWV $ORCL

Pretty big fumble for Anthropic not to have done a massive round in July-August.. Makes little sense that they raised just $65B (lol) in May after OAI did $122B a few months earlier, even though Anthropic had all of the momentum at the time..

CleanSpark $CLSK Founder and CEO Matt Schultz was also at H.C. Wainwright.. Some decent color on the importance of securing a top general contractor, what lenders are looking at when underwriting data center financings, and the current status of the sites they have in Texas under exclusivity (with their same signed tenant in Georgia). "For us, our tenant actually came with a specific request for an EPCM, and that EPCM has built for them before, and they also manufacture a lot of the mechanical, electrical, and plumbing components internally. It enabled us to get certainty on supply chain. It enabled us to push some of the potential risks and liquidated damages over to the construction site.. The ability to build.. has quickly been de-risked largely because of the fact of that one lump sum turnkey contract at the request of our (tenant)." "In the beginning, the analysis for high yield or investment grade was, who's the guarantor? What's the credit wrap? Who's going to give certainty that the rent payments are being made? That evolved to.. what's the certainty on construction, and what do the political headwinds look like." "We feel very confident that the exclusivity and the option for that will ultimately be executed, and we're seeing price go up in parallel with demand."

Some interesting quotes from Core Scientific $CORZ Co-Founder and CDO Russell Cann at the H.C. Wainwright conference earlier this week.. - On liquid vs air cooling: "Direct liquid cool and air cool were just not the same thing. The legacy data centers we had that were air cooled, it just made no sense to try to convert those. We had to build everything from the ground up." - On trends in $/MW CapEx costs: "The first deals.. late in 2023, we had good hard estimates of around $4.5MM a megawatt.. By the time we got around to inking the deal in June, that same facility was just under $8MM a megawatt. Now, the things that we actually turned on this year, the other sites, they were floating around $10MM to $10.5MM a megawatt.. The stuff that we have turning on in 2027.. is between $12MM and $13MM a megwatt. Most of that increase is labor, and then.. the gear, switchgear, the transformers." - On labor constraints: "If you have a kid graduating high school right now, they need to go become an electrician. Journeyman electricians right now on one of our sites make about $250k - $350k a year, and master electricians are bringing in $750k a year." - On AMD's expansion option: "We agreed to give them the next 2,000MW at three of our sites in Hunt County, TX, Pecos, TX, and Muskogee, OK.. I would say unless the market just really goes south, AMD will take all 2,000 of those MWs. They have already announced who their customers are, so you guys can figure it out."

Blackstone and Google getting their neocloud JV started with a $22B GPU loan is pretty amusing.. Have to respect $BX sticking to their PE roots and using as much debt and as little equity as possible to scale the business. But also ironic that no one bats an eye when $BX and $GOOG raise debt to build the infrastructure required to support long-term contracts. Maybe non-stop equity and convertibles issuance isn't the most optimal way to finance a neocloud?

Yup.. imo, the labs’ compute capacity for any given period of time should be viewed as fixed, and has to be allocated to either R&D or inference (COGS). Model safety/alignment is a sub-segment of R&D. So all else equal, if they allocate more compute from within the R&D bucket to safety/alignment (e.g. shifting R&D spend away from areas like pushing the frontier on RSI).. Or from the inference bucket to R&D for safety/alignment.. it’s probably net negative for near term revenues. But over longer period of time, safety/alignment should be a capability representing a source of incremental demand for compute.. Zuck articulated it well that safety/alignment really is just a standard product feature. And seems pretty clear the labs don’t have enough inference (or R&D) capacity as it is. So they’re going to need even more net new compute.

It’s probably safe to say $META ‘s capex budget isn’t getting cut after the initial reception to Muse.. Will be interesting to see if they issue equity to accelerate their spend. Now might not be a bad time to do so, with Anthropic and OpenAI still yet to IPO.

“Galaxy has been very upfront, very open and very sensitive to the needs of or to the concerns of people relative to water.. My experience right now and what I've learned in dealing with this space is that we can be proud that Galaxy is really a good example to follow.. They seem to have everyone's interest at heart.” - Dickens County Judge Kevin Brendle “I can see that the small businesses that we have here, they're all doing much better business and I can see that through the sales tax that comes back into the city.” - Dickens City Secretary Lillian Atkinson $GLXY

Following Hut 8's rather diabolical Friday-afternoon-before-Labor Day tweet that their Beacon Point project (the ~1GW gross, 704MW IT load campus leased to NVIDIA) received a conditional "Base Load" classification from ERCOT, I'd imagine early next week we get a flurry of similar updates from the other public data center developers building in Texas. In particular, confirmation of "Batch Zero - Base Load" for certain $CIFR $CLSK $CORZ $GLXY projects could be a catalyst for this broader bucket of names as it takes the projects closer to final authorization to energize, and likely removes any doubt as to whether the projects can be commercialized via a tenant lease. As the Hut 8 - NVIDIA lease showed, projects that were (i) approved by ERCOT prior to the Batch Process but (ii) are still subject to the Batch Process (because they have not yet been energized) can in fact be leased.. Yet the confusion caused by the hodge-podge of Batch Zero definitions, process steps, and ongoing delays in implementation of the Batch Process itself has definitely been an overhang on these stocks. Taking a step back, ERCOT's Batch Process was established in response to the overwhelming amount of load requests from data center developers (and speculators) seeking to connect to Texas's grid (400GW+ total..!). Rather than approve or deny each request separately, ERCOT's Batch Process studies all of the projects as a group to understand and allocate grid and transmission capacity. Late last week, ERCOT provided classifications on requested loads to transmission service providers (who then pass along those classifications to the relevant data center developer behind them). In short, a "Base Load" classification is the best outcome, and indicates that ERCOT has treated the project's existing studied power allocation as valid and effectively reserves the capacity for it. "Base Loads" are essentially grandfathered in. (Note that ERCOT had previously approved certain loads prior to the Batch Process being formalized. "Base Load" classification on these is more or less a reaffirmation of the prior approval.) "Studied Load" means that the project is eligible for the first "batch" of interconnection approvals (Batch Zero), but the requested power is not reserved or guaranteed as it requires further evaluation. "Studied Load" could result in full, partial, or no approval of the requested MWs. "Excluded Loads" are not eligible for Batch Zero and must wait to apply through a future ERCOT Batch Process. To summarize on $CIFR $CLSK $CORZ and $GLXY's Texas projects: - $CIFR hinted that its Colchis (1GW gross), Mikeska (500MW), and McLennan (500MW) sites are expected to receive "Base Load" classification, noting on its August 4th earnings call that "All three sites have necessary deposits funded, land secured, and their requisite studies and executed FEAs were submitted to ERCOT on time. We have strong conviction that all three sites will be included in Batch Zero." $CIFR also stated its 900MW Apollo project and 200MW expansion at Stingray "have been submitted as studied loads in Batch Zero." - $CLSK's Sealy site (285MW) came with existing ERCOT approvals when acquired in 2025. Similarly, 300MW at its Brazoria campus received ERCOT approval in 2026 before Batch Zero was formally effective. The second phase at Brazoria (another 300MW) is expected to be either base or studied load in Batch Zero. - $CORZ's 430MW at its Hunt site was approved in 2024, and a portion has already been leased to AMD (indicating a Base Load classification is highly likely). For its 300MW of requested expansion at its Pecos site, $CORZ did not provide explicit classification guidance. - $GLXY's 830MW (gross) of incremental power at Helios (Helios II) was already approved by ERCOT in January (before the Batch Zero process was codified), and the company labels it as "available to contract." As such, I view a "Base Load" classification as extremely likely. With respect to Galaxy's Caspian site, the company stated on its August 5 earnings call that "Caspian has a potential total gross power capacity of 700 megawatts and is eligible for Batch Zero Base Load classification based on the milestones we've already completed, including executed interconnection agreements and the posting of known and identified required financial commitments for transmission and distribution upgrades. That puts Caspian on one of the most advanced development paths within our expansion portfolio." Accordingly, I think a "Base Load" classification for Caspian is highly likely as well. Galaxy's Helios III (1GW) and Selene (900MW) are expected to be "Studied Loads." $CIFR $CLSK $CORZ $HUT $GLXY

After rereading the Q2 earnings transcripts for the AI data center developers (e.g. powered shell providers, including $APLD $CIFR $CLSK $CORZ $GLXY $HUT $RIOT $WULF).. The message was pretty consistent across the board. Demand for big blocks of data center capacity from a variety of tenant types (hyperscalers, neoclouds, frontier labs, NVIDIA and AMD directly, etc.) remains strong. A handful of examples: - Applied Digital (APLD): "(We) expect.. new capacity to command higher pricing. We expect (expansions at our existing campuses) to be executed on.. materially higher lease rates than the existing leases and possibly longer duration.. We've seen other contracts out. It's great to see pricing moving up in the industry. Just gives a really strong indicator of where demand is out there." - Cipher Digital (CIFR): "The demand environment is very strong and extending.. I have never seen a better environment for us in terms of how lease terms are evolving with higher rents, longer time periods, triple net structures, etc.. The demand environment has never been stronger.. We want to make sure we strike the best possible deal. Best possible deal includes both great structures, great terms, long-term leases, triple net structures, high rental rates.. It's just a matter of getting to a place with picking our dance partner. I expect all those available megawatts will end up leased." - CleanSpark (CLSK): "We see the demand profile on the tenant side of the house continue to be relentless." - Core Scientific (CORZ): "As we look at the demand picture really for developers that are having direct conversations with the counterparties that can sign contracts right now, it's really starkly contrasted against what you're seeing in the media headlines. What we're seeing on the ground is still a significant amount of demand coming out of the hyperscale channel and the AI labs. I think in terms of what we're going to see over the coming months and through the remainder of this year are continued new deal announcements as it relates to large scale infrastructure commitments." - Galaxy Digital (GLXY): "Probably most importantly, while forward market demand for power and compute is stronger today than it's ever been, forward market supply of new projects, while ambitiously headlined, is very quickly being constrained now by both physical delivery and political headwinds." - Hut 8 (HUT): "Demand is robust from all the conversations we've had. Demand is real. Demand is there. Everyone wants capacity yesterday." - Riot (RIOT): "Following our first data center lease with AMD earlier this year (at Rockdale), demand for large-scale data center capacity has continue to accelerate. At Rockdale, subsequent to quarter end, we executed a lease with one of the world's leading frontier AI labs (Anthropic) for 191MW of critical IT capacity.. At Corsicana (RIOT's other campus), interest in our 1GW of utility power has been substantial. From a broad field of prospective tenants, we have narrowed our focus and entered into a non-binding letter of intent with a single tenant for the entirety of the site." - TeraWulf (WULF): "At our Muskie site (the site TeraWulf is currently in market to lease), we have the most active data room we've ever had, and with world-class credits as potential customers." $APLD $CIFR $CLSK $CORZ $GLXY $HUT $RIOT $WULF