Kash (@KashRamki) header
Kash (@KashRamki) profile picture

In search of value

Joined October 2010
Photo by Kash (@KashRamki)Photo by Kash (@KashRamki)Photo by Kash (@KashRamki)Photo by Kash (@KashRamki)Photo by Kash (@KashRamki)Photo by Kash (@KashRamki)Photo by Kash (@KashRamki)Photo by Kash (@KashRamki)
Agency + Urgency = Extraordinary outcomes. Agency: do what needs to be done, regardless of permission. Urgency: continuous learning and execution like your life depends on it. That’s it.
1
2
49
26,648
Nothing has really happened to the AI trade. Ignore the noise.
This sell-off has been driven by a perfect storm created by: 1) The Blue Tidal Wave win and fears about an anti-AI/anti-datacenter DC, 2) losing the global leader of the AI Revolution Jensen in his black eye moment, 3) China driven regulatory concerns across the rest of semis/memory stocks (e.g. AMD short report yesterday), 4) multiple market clearing events which have just led to more selling, 5)Anthropic IPO being a massive success but marginally reducing liquidity from other tech names, 6) significant fund P&L pain diminishing buying power, and 7) sticky inflation and hawkish Fed speak with concerns around more hikes and constricting financial conditions….The $4 trillion AI buildout is still happening and our tech thesis into 2027 remains.
3
4
71
2,946
An update to my post from earlier this year. Three things stand out to me. 1) I am always surprised by how little AI capacity has actually been energized to-date. Adding up the table, these 9 companies have ~33GW of AI capacity announced (using the floor figures) and ~0.5GW energized. That's about 1.5%. $NUAI has 0MW energized, but so do $FRMI, $HUT and $KEEL, at $1.3-14B EVs. Most of this group is still valued on capacity that isn't online yet. 2) The NUAI analyst targets look like Phase 1 values to me. The 5 targets run $6-$17, avg $11. That's an implied EV of ~$676M-$2.2B vs ~$283M today, or 2.4x-7.9x. Phase 2 is very much in play with TCEQ now issuing permits once again. As a reminder, Phase 2 is 2.7x of Phase 1, in terms of MWs, and much more in terms of equity value. 3) Nobody is paying for New Mexico. NUAI holds an option on ~3,500 acres in Lea County, NM, announced in Nov 2025 as a future 7GW AI hub. It's only an option: $200K paid, no tenant, power contract or permit on record, and it isn't mentioned in the Q3 10-Q, release or call. It doesn't belong in a valuation today, and I haven't seen an analyst note that includes it. At ~$283M EV the market isn't paying for it either. Will let others draw their on conclusions from this, but this has been an interesting view of the market, over time. Not investment advice.
I prepared this chart to illustrate the severe mispricing in $NUAI. Current EV: ~$260M Peer average EV: $8.1B That gap? Massive mispricing for a company with 8GW announced AI capacity - Expected energization of Phase I in 2027 - pure behind-the-meter power buildout, and no grid queue nonsense. Peers like $IREN ($15.7B EV, 4.6GW), $APLD ($9.9B, ~1.5GW), $CIFR ($8.9B, ~4.2GW), $HUT, $WULF etc. are getting huge credit for announced hyperscaler deals. But reflect on this: Even though they've inked those deals, the vast majority of their capacity remains in construction, with energization timelines mirroring or a year ahead of $NUAI's 2027 expected timeline. Execution risk cuts both ways—yet the market prices $NUAI for total failure while rewarding peers well before energization. Downside protected by powered land value alone. Upside explosive on first hyperscaler deal. The yellow arrows and the dotted red line tell the story. $NUAI
3
7
1
57
15,453
Feels within reach for $NUAI
1
2
56
2,829
Wonderful alpha from the $NUAI X community. Slowly and then suddenly. That’s how value is created.
15
6
94
6,581
Go Raptors! $NUAI
3
1
40
3,542
I agree with @jiahanjimliu on the hit to Chinese compute capabilities. The China diversion story has a second-order effect that gets less attention: near-term oversupply risk for NVIDIA. Pre-ban, China was ~20% of NVIDIA revenue (FY2024) and ~95% of the advanced AI accelerator market there. Official sales have since collapsed toward zero for restricted chips. On volume, Epoch AI’s median estimate is ~660k H100e smuggled through end-2025 (customs data points higher, ~1M+). Against ~9M H100e of AI compute NVIDIA shipped in 2025, that is roughly 7–13% of a peak year’s output — material, but not the majority. NVIDIA has simultaneously extended large residual-value guarantees and capacity backstops across the ecosystem (including major commitments tied to OpenAI and neoclouds). Reporting from the WSJ and others shows Jensen drove much of this financing activity to keep demand clearing the supply pipeline. When a non-trivial slice of production finds its way into gray-market or remote channels while official China demand is largely offline, the remaining buyers have to clear a larger share of the output. That increases reliance on credit support and residual-value guarantees to sustain the flywheel, and pressures incremental unit economics. I continue to believe GPU/hr prices will continue to rise because the bottleneck isn’t chips. It will be power and datacenters, the foundation of compute. $IREN was smart for sure for waiting, but to be clear, it was a gamble, albeit on very good odds. The installed base is still heavily $NVIDIA. The marginal economics look more fragile.
Impact of China GPU Tightening The US will tighten both export to China and China remote access to GPUs in SE Asia. @KashRamki dug into the figures and 60-60% of Chinese compute is Nvidia. New shipments have fallen, market share of shipments pre-ban 95% and fell 34% but 34% is still significant! This is: 1. Bearish Chinese open source labs. 2. Bearish GPU prices but neutral GPU rental prices. - All 2027 powered datacenters have been planned, any planning now will not result in 2027 datacenters! Thus the amount of GPUs available on market will be limited by datacenters which were already set in stone. - This is bearish semiconductor manufacturers and bullish Neoclouds. Neoclouds earn the spread of GPU rental prices - GPU prices. If the amount of GPUs available on market is constant from datacenter availability and GPU prices come down, everyone who did not "secure" their VR200s, will benefit! In $IREN's earnings call, @danroberts0101 talked about securing GPUs late because prices may not fall. What a genius!
6
2
1
37
6,077
I casually dug into how much China has benefitted from NVIDIA chips and I was surprised to find that China still runs ~60-65% of its accessible AI compute on NVIDIA/US-designed chips. IFP end-2026 estimate: ~4.6M H100-equivalents available to Chinese entities. The breakdown is as follows: • Domestic Chinese chips: ~1.7M (37%) • Legal US imports: ~850k (18%) • Smuggled restricted NVIDIA: ~1–1.2M (22%) • Remote/offshore (mostly SE Asia): ~1M (22%) Diversion + remote access exceeds everything Huawei, Cambricon et al. have produced combined. Epoch AI median: ~660k H100e smuggled through end-2025 (range 290k–1.6M). New shipments have fallen (IDC: NVIDIA 34% share H1 2026 vs Huawei 28%), but the installed base is still heavily NVIDIA. Pre-ban it was ~95%. Again, not the latest generation of chips but enough volume to stay somewhat relevant over the past few years. Hoping there are no more leaks of the Vera Rubins and beyond going forward, so as to keep China squarely in the rear view mirror. #Nvidia #NVDA #CHINA #USA
3
7
3
37
29,432
Grok bot is truly phenomenal. Burning through tokens as I ramp up usage. Jaw droppingly easy to configure and use. It's scary to think I might have just replaced my accountant in the past 2 hours. And that's just the beginning. @elonmusk I can't wait to a broader number of suggested bots!
1
2
1
22
2,836
A few quick thoughts on the $NUAI earnings call: As expected, one tenant, under exclusivity, and management says there is "full agreement on what the economics are going to look like." If it's anything like Riot's, which they confirmed was the case on the call, then we're in for some pretty stellar economics. First modules September 2027, ready for service Q4 2027, in line with expectations for power from Vistra. I was pleased to hear that the JV developer was removed at the tenant's request. The tenant wants to deal with NUAI directly, and NUAI keeps the economics a developer would have taken. That's a big shift from some sort of 50-50 economics split with Stream. We keep you keep 95% of the pie, with the other 5% going to a partner who is now motivated to deliver more power for a Phase 3 at TCDC. That's pretty huge! Also, Macquarie's lease deadline is no longer an issue. Macquarie waived its repayment right for the missed Oct 8 deadline, and the PIPE paid off $60M of loans. About $20M of debt is left, fully cash-backed. Of course, this hurt investors to raise at the high $2s. A self-inflicted and unexpected wound. This one was interesting: some of the $118M tied up in the Luminant LC could come back. The CFO said that once the lease is signed there are options to restructure the LC and release restricted cash. If ~$35M comes back, most of the near-term dilution goes away. There is a lot of FUD on X around Phase 2. It's clearly moving, and fast! Mgmt expects the air permit by year-end as TCEQ is processing permits again and turbines arrive Q4'27-Q1'28. That last part tells me everything I need to know. The stock has been responding very favourable in light of this commentary. Even the unfortunate news around Jensen couldn't hold $NUAI back. Steady as she goes. NFA
$NUAI has released its Third Quarter 2026 Business Update. Highlights include: ✓ 207 MW of firm, contracted power ✓ Power availability expected in Q3 2027 ✓ Initial ready-for-service targeted for Q4 2027 ✓ In advanced negotiations on the Phase 1 lease documentation under a non-binding LOI covering TCDC Phases 1 and 2 ✓ Site preparation underway ✓ Strengthened liquidity to support TCDC development ✓ Proposed Phase 1 modular build: preliminary cost estimate below half that of a comparable traditional build “Responsible development has guided our approach from day one. That means designing power arrangements to avoid additional transmission costs for ratepayers, using water responsibly and giving local communities a voice. We want Texans to have a seat at the table and local residents and businesses to share in the opportunities TCDC creates,” said Charlie Nelson, Chairman and CEO of New Era. Management will host a business update call and webcast on Thursday, November 12 at 8:00 a.m. ET to discuss the Company’s recent business developments and operational progress. Register here: edge.media-server.com/mmc/p/… investors.newerainfra.ai/new…
7
14
1
117
15,936
Good call outs from the $NUAI earnings call.
Nothing has even happened yet $nuai I think the most overlooked part of the $nuai call yesterday included the incremental 99 acres for TCDC and the differentiation from the $riot deal that @Carboncharlie touched on regarding construction. @JasonLes_ mentions with @blockspace that their deal with @AnthropicAI provided electrical pads to their client but $nuai will provide construction services as well. The clues are beginning to point to @AnthropicAI as the tenant at TCDC and it’s very telling that one of the largest AI frontier labs in the world will entrust construction to a company with a sub $500m market cap. Additionally, the focus remains on Phase 1 and 2 but it’s becoming more clear that @VistraCorp will uses its 1.17 GW plant to power Phase 3 and those additional 99 acres could be worth 100-200 extra MW Finally, @JasonLes_ on his earnings call mentioned the initial LOI for a Tier 3 DC would produce 756 MW IT but converting to a modular build format and increased IT load to 910 MW There’s a chance that 1) $nuai earns more per MW due to construction 2) $nuai earns more than sell side is guiding due to cap ex and power flexibility by the tenant 3) TCDC is 1.5-1.6 GW; and incremental 100 MW that assumes 70 IT is likely worth an extra billion on equity value 140m (70x2.00) x 10 EV Mult = 1400m less 420 (70x6) 980m equity value
1
5
67
4,177
I spent the day analyzing $RIOT's Corsicana lease with a Frontier AI Lab, and what a similar deal could mean for $NUAI. Here are the facts on Corsicana: - 910 critical IT MW on 1 GW gross. $130/kW-mo, escalating 3%/yr, billed on the full 910 MW whether used or not. Tenant can flex usage seasonally within the 1 GW load. - $28B over the 15-year initial term ($54B with extensions). ~33% average unlevered yield on cost (!!) - Capex: ~$2M/MW for the HV/MV power pad (~$1.8B), plus up to $4M/MW reimbursed to the tenant for modules (~$3.6B), paid per phase only after modules are delivered. - Delivery: 400 MW Sep 2027, 200 MW Jan 2028, 310 MW Jul 2028. Rent starts no later than 60 days after each block is delivered. - Equipment: first 400 MW substation is operating. No backup generation, so no air permits needed. - Financing: base case is 80-90% LTC non-recourse project debt ($4.4 to 4.9B) on tenant credit. I believe they will have negotiated a fallback strategy if financing doesn't come to fruition. This deal is a paradigm shift and the new gold standard for every datacenter developed out there. Here's my read-through to $NUAI (my napkin math): - TCDC in Ector County has 207 MW of Luminant power come Q3 2027 (Vistra gets 5% of the Phase 1 project company) and ~500 MW more planned on behind-the-meter gas. - On the Riot structure at ~690 IT MW and $125/kW-mo: ~$19B of contract value, ~$1.3B average NOI, ~31% unlevered yield on cost. Tenant funds the modules, project debt funds most of the pad, and the equity check is under $100M. - At a 10x NOI, with a 40% haircut on Phase 2 for the air permit, turbines and timing, I get ~$27/share in a base case, against ~$2 today, with ~10% dilution at current prices. That's $11/sh on Phase 1 alone. Of course, it all hinges on what mgmt can deliver. NUAI still has to sign a tenant and Phase 2 needs an air permit that Texas won't process until the ERCOT audit wraps December 10th. Q3 results and call on tap tomorrow at 8am ET. Looking forward to it! NFA
15
16
4
90
12,978
This is good commentary from @level39 on the distinction between $NUAI and Thunderhead, the actual applicant for the air permit (assuming they own the turbines on their balance sheet and are the power generation partner). The big picture is that TCEQ's current freeze on air permit issuances is a tip of the hat to Gov. Abbott. TCEQ is under no obligation to comply or to continue the freeze beyond Dec 10th. The current freeze affects all air permits, not just those linked to datacenters, though that is what started all of this. For what is effectively a petrostate, that's a hell of an ask of Texans. Rest assured that come Dec 10th, the freeze will be lifted and the spigot shall be opened.
Replying to @KashRamki
TCDC's air permit is likely to get approved when Abbott's pause lapses in mid-December. Thunderhead's permit uses a 4911 code, which obfuscates its connection to a data center from TCEQ's perspective—only connected through press/investor clippings.
6
3
1
57
6,682
Amen
Jensen’s affair is a blip for Nvidia. Sure, it’s embarrassing, but it doesn’t change the company’s insane moat. If it hadn’t been a Chinese spy, the tech CEOs would have let him have sex with one of their wives. Watch the interviews. Whenever Jensen comes up, brash billionaires suddenly sit up straight and the brown nosing starts. If anything, this spy story is the ultimate testament to Nvidia's competitive advantage. The Chinese, like all the tech CEOs, are desperate for access to his chips, which are limited and extremely expensive. Why? Because it makes or breaks their ability to compete. Want to buy a state of the art Nvidia GPU? Good luck. Maybe in 2028. Their order book is full. There's a reason China rallied its national security apparatus to compromise Jensen and get H200s. They NEED these GPUs or they lose the AI race. It turns out the "Sputnik" moment we all got hot and bothered about (DeepSeek) was complete BS. There hadn’t been a breakthrough at Huawei as people feared. The Chinese were using the same Nvidia GPUs that we were and distilling Anthropic’s and OpenAI’s models, with a lag of 6–8 months (way behind in AI's compressed time scale). They hadn't solved the chip problem, they had simply been trafficking some sweet, sweet Nvidia GPUs. Nvidia has the sort of moat that Warren Buffett drools over: ✅ Network Effect (CUDA) ✅ Pricing Power (A product nobody can compete with) ✅ Predictability (YEARS of demand) ✅ Switching Costs (Moving workloads is a pain) Sure, given the huge reward of competition, Nvidia's advantages will likely erode over time as alternatives like Google's TPUs, Amazon's Trainium, and AMD's Instinct GPUs improve, but (HUGE BUT): 1. Chips that are seriously competitive with Nvidia's are likely 5–7 years out. 2. The AI labs are in a desperate race for ASI and we may be only 2–3 years away from recursive self-improvement (the singularity) — nobody wants to gamble on moving away from Nvidia. What about the risk that Jensen gets pushed out? He’s not going anywhere. As per usual, he’s going to hang his head in shame for two weeks. We’ll all forget about it, and life goes on. Trump has already publicly praised him since the story came out, and even said he might invite him on his tour of Asia. And even if he did get fired, that wouldn’t erase CUDA, the installed base, or switching costs in the near term. The stock is trading at about 11.2x estimated earnings for its next fiscal year, and I wouldn’t be surprised if it goes lower in the coming weeks. For context: Amazon: 24.8x Apple: 32.9x Microsoft: 22.3x Alphabet: 19.4x Meta: 18.0x And a fun note: Nvidia assumes ZERO revenue from Chinese datacenter compute in its forward earnings. (Plus any cancelled Chinese espionage revenue is likely to get gobbled up by a North American AI lab.) There’s a great Charlie Munger line: “Show me the incentive and I’ll show you the outcome.” Let’s think about the incentives: The AI labs need Nvidia chips to win. They will not move to AMD due to morals. And the US government needs Nvidia (and Jensen, who is viewed as necessary to its success) to keep building them. Not only to maintain dominance in AI, but also to ensure the stock market doesn’t implode. This is embarrassing for the poor guy, but if anything, it’s a ringing endorsement from the Chinese that Nvidia’s moat is fortress level. I'd love to hear the bear case. What am I missing? *Not investment advice, do your own research! I don’t currently own Nvidia, but I might in the future.
1
28
3,418
Welcome to the $NUAI cap stack @mikealfred!
I bought the stock for the first time this morning.
4
14
2
135
20,057
Rumblings of republicans calling for a lift of TCEQ moratorium early. My base case is that it gets lifted on Dec 10th.
7
3
1
123
15,228