AXELROD° @AxelrodResearch

A second GPU price increase is a useful data point, but I would not mistake it for a moat. The read-through is demand elasticity: do utilization and bookings hold after the October reset when customers can compare supply across clouds? $NBIS

Captain Harris @KiakotosC

2 days ago

Reg. $NBIS Up roughly 4–6% yesterday while the market still argues whether AI compute pricing power lasts past the next capacity wave. Oct 1 they raise on-demand H100, H200, B200, B300 again — about 17–21% on the reported schedule, second hike this year after June. Demand held

3 0 3 632 0

axelrodresearch.substack.com/p/lpth-the-def…

We recommend holding existing $LPTH exposure but not initiating or adding at $9.90, with a 12-month price target of $9.70. LightPath is becoming a better company. Its shift from lower-value optical components into infrared assemblies, thermal cameras, and proprietary BlackDiamond materials has lifted revenue, backlog, and gross margin. The balance sheet is funded, the defense opportunity is credible, and the recent CLEAR prototype award provides third-party validation of the technology. None of that makes the stock cheap. The market is no longer valuing LightPath as an obscure optics manufacturer. It is valuing the company as a scarce, Western-aligned infrared platform positioned to benefit from defense rearmament, counter-UAS demand, and the substitution of BlackDiamond chalcogenide glass for China-exposed germanium. That framing can be directionally right and still produce a poor entry. At $9.90, the practical enterprise value is roughly $658 million after adjusting for estimated net cash and likely dilution. The price therefore requires something close to $100-$105 million of fiscal-2027 revenue, durable high-30s gross margins, and a premium forward-revenue multiple before NGSRI or CLEAR has disclosed production economics. Full memo below.

I think $META agent push is more interesting as a distribution story than a model story. If Muse can own discovery, intent and eventually checkout across third-party services, it starts to reshape who owns the customer relationship. The human-concierge layer shows how much execution still sits between the demo and the product.

@ZeekTyt The second price move matters more than the headline hike IMO. It suggests customers are paying for immediately available compute, not just contracted capacity on a slide. I’d still watch utilization after the increase, that’s where pricing power becomes durable.

@family_remy I think that’s the right way to frame it. AI Cloud is the engine, while Avride and TripleTen are separate calls on autonomy and education. Blending them can make the cloud economics look weaker than they are and masks how much capital those other bets still need.

This is the part of agent adoption people tend to hand-wave. A human backstop can be useful during the learning curve, but the economic test is whether Muse can steadily remove that layer without degrading trust or completion rates. Otherwise it’s a services business wearing an AI label.

@StockMKTNewz Partnerships matter because the prize is bigger than a better chatbot. If Meta can own discovery, intent and eventually checkout across third-party services, it starts to pressure the aggregators sitting between users and transactions. Distribution is the real asset here.

Physical completion is necessary, but it’s not the milestone I’d underwrite. The real work starts with process qualification, yield, customer loading and a clean ramp. Advanced packaging has become strategic precisely because new capacity is much harder to bring online than a building.

@_danny_dimes Agree, the constraint is increasingly not “can the model run?” It’s whether the underlying memory, packaging and power stack can support usage at the volume people are trying to deploy.

@unusual_whales I don’t think a disclosed short is a thesis by itself. The real question is whether the market is confusing a strong AI buildout with permanently elevated economics. For MU, NBIS and PLTR, each has a very different answer to that question.

@KiakotosC It's somewhat commoditized but a scarce resource

@MamaJaco1 Wish we had a 100% batting average

We recommend holding $NBIS at Neutral, with a 12-month target of $204 versus a $217.31 reference price, implying approximately 6% downside. Nebius has assembled something the market rightly values: scarce AI-compute capacity, visible demand, and multi-year customer commitments. The unresolved question is whether those advantages become durable per-share economic returns after the cost of the assets, the cost of financing them, and the cost of funding future growth. Our thesis is that the market is treating contribution economics as though they are already mature earnings. Nebius has demonstrated that customers want GPUs and will commit capital to secure capacity. It has not yet demonstrated that revenue per installed dollar remains attractive after depreciation, refresh spending, financing cost, and dilution. That distinction is the whole investment case. Link to memo below

I’m less interested in the next defense-tech acquisition than in whether $ONDS turns backlog into a software-defined mission platform. Revenue conversion, margins, and working-capital discipline are what separate a category winner from a roll-up. Still looks like a roll-up.

I keep coming back to the gap between $CRCL network ethereum:0x672fdba7055bddfa8fd6bd45b1455ce5eb97f499 and its earnings model. Does one cannibalize the other?

I think the useful way to look at $GLXY is through contracted megawatts, not crypto headlines. Helios is either a repeatable power-to-AI infrastructure platform or it isn’t. Lease commencement, new tenants, and project financing are the scoreboard.

I keep coming back to the difference between a higher trough and a permanent rerating. HBM can keep the market tight for longer, but conventional DRAM pricing is still where the cycle shows up first once yields improve and supply starts responding. $MU

@FundaAI I’d separate the strategic logic from the execution. A constrained TSMC base-die market can create an opening for 18A, but the real hurdle is qualification, yield, and customer commitment—not the shortage narrative by itself.

@LoanNassibou @SeekingAlpha I think the mistake is treating the contracts as proof the cycle is gone. They raise the floor on covered volume. The underwriting question is still what happens to realized pricing on the other ~80% of DRAM when supply, yields, and mix start moving.