Aaron Chan

I am so proud of the work my wife Ladan has done (and continues to do) for the tens of thousands of cats she has rescued across Southern California. It is an endless journey, but it makes me so happy that the @JHudShow recognized her for the tireless effort she and the @FriendsofNormie team put in every day. As she says, "Open your bathroom, save a life." youtube.com/watch?v=S0UJTWtH…

A bunch of rate sensitive names have had a sneaky bounce off the bottom in the last two weeks. Maybe a bit of rotation away from AI, maybe some inflation and rates cooling. Lots of stocks +15-30% pretty quickly.

$CVNA In September 2026, the SEC informed us that it had concluded its investigation and did not intend to recommend an enforcement action against us in connection with the matter.

What if, god forbid, AI growth capex was funded only by reinvesting operating FCF from this point forward instead of burning through new equity and debt capital?

It's a hot topic but the amount of political backlash against data center projects is significant. Over 25% of the tracked AI-related data center projects are under some form of political contention. Some relatively minor, some very major. 3.3 GW have been killed altogether.

It will be interesting to see what kind of core network $SPCX will need/build to stand up a wireless business given its existing and unique vertical infrastructure. Standard wireless toolkit likely does not apply. Should be incremental to towers and fiber, but how much?

70% of the Nasdaq composite is in a bear market and the S&P 500 and Nasdaq are within 2% of all-time highs. It's pretty incredible.

I am making good progress on my AI data center tracker. It's cool to see how much capacity is coming quarter by quarter, and where. Built this to analyze potential wavelength growth for $CCOI and others, but interesting more broadly as well. DM me if you want to try it.

Looks like $CVNA is refinancing $1.6b of 9% notes with $1.66B of TLB at SOFR+225 (3.63%+2.25% = 5.88%). Should be a nice tailwind to earnings. Expect to see further reductions in cost of capital over the coming years.

I'm getting a lot of inbound on $CCOI. The inevitable -30% move on earnings has happened yet again. Growth has been consistently and astonishingly disappointing every quarter. I have been consistently wrong on this - I thought we should have seen meaningful movement by now. This is not financial advice of course, but I'm not selling because I haven't been able to convince myself that it's not timing. In most ways, it would be easier to sell and move on. The EBITDA bridge below is my early attempt to model all the moving pieces described on the earnings call, assuming pretty anemic growth going forward (~1% q/q on net growth, no material acceleration in waves). Could be wrong of course. There seem to be surprises every quarter, but I'm trying to get to the normalized numbers. Dave's balance sheet strategy relied on waves ramping quickly and building toward a big enough revenue and EBITDA base as the TMUS payments ran out. This acquisition always required a strong growth story to emerge. Since growth is very late, leverage looks way worse than expected and the reflexivity between equity and debt is causing problems on both fronts. AI has created incremental demand, but the short-term surge has created logjams and delays across the whole ecosystem since last year. I think the 7-year opportunity is larger than expected going into the Sprint acquisition, but the first two years have created a lot of odd situations which have piled up into significant delays. Cogent and other carriers talk about the robust demand out there, but it hasn't shown up materially in revenues for anyone yet. This may be a bad analogy, but here it goes. Imagine you have a sensor measuring traffic through a certain section of a highway. The market is pricing Cogent as if the highway is pretty empty, which is why volume is low. Cogent is arguing that there is so much traffic that the highway is jammed bumper-to-bumper and only one lane is open due to an accident, which is suppressing volume measured by the sensor. If they are right, at some point all those cars will get through which will result in accelerating volume growth in the wave business. That's the entire bet on Cogent. Is there an accident causing delays, or is the road empty? The rest of it (DC sales, IPv4 monetization, etc.) is a bunch of tactical maneuvers to bridge the company through this period because of its over-levered balance sheet, but they do little to help the terminal value of the business.

Maybe it's fun to have a wish list for $CCOI's earnings, ranked loosely by surprise potential: 1. They communicate a willingness to sell the Sprint IPv4 addresses to deleverage faster. Unlikely to happen, but I'd love it. 2. LOI for more DC sales. Possible, but unlikely. They didn't close the last deal until June 29th, my understanding was that they had to work hard on the handoff until close. I doubt all the rest will go as one bundle, it might take 2-3 deals to sell the rest. 3. Waves installed up q/q. I think this is likely, but it's been so lumpy it's hard to have high confidence. The ~1,500 signed but not billing orders in backlog haven't disappeared. Waiting on equipment, on greenfield DCs to complete, on customers to perform, etc. Probably the same messaging again, could be good, could be bad depending on the quarterly blow-by-blow. 4. Overall growth inflection to positive q/q. Was supposed to happen in Q1, but they keep shedding bad revs from Sprint which has been a drag. GP growth will be healthier than revenue growth. EBITDA should step up nicely in Q2 based on normal seasonality q/q. 5. Small stock buyback in late Q2/early Q3. Maybe 1m shares? Nothing that would move the needle materially vs. the board's preference for deleveraging, but a nice opportunistic buyback with $225M of new cash on the balance sheet. 6. Refi coming - the market is watching this closely, I think it's coming shortly after earnings. The DC sale closed so late they couldn't launch before quarter end. Then it made the most sense to wait until after Q2 numbers. 7. Capex will remain elevated due to cost inflation, but pricing trends will be better, creating a growth offset. They should get better top line trends as carriers raise prices and/or add surcharges now that capex/mb of capacity is going up for the first time ever.

$CCOI reports tomorrow and everyone I know is bracing for impact based on recent history. Even $CVNA in 2022 wasn't as bad as Cogent on earnings.

I own some $AMZN so I’m not trying to hate. But a 3 year payback on a 5 year asset justifies levering up to let bottlenecks get 85% GMs?

I'll riff on $CVNA a little. Q2 was a decent quarter with some pretty interesting moving pieces. It wasn't perfect by any means, and it's disappointing not to see operating leverage show through. I think much of that is a choice they are making - keeping growth high while fixing things in parallel. Recall that they missed Q4 2025 due to recon issues, with some facilities really falling behind the average. It took them until about late May/early June to turn the corner on those. Q1 saw some early benefit and Q2 got some more. Q3 should be the first "clean" quarter since they centralized a lot more and deployed new tools that reduce IRC variance and improve aggregate efficiency. Growth was choppy across 1H because of this, which showed up in weekly 3P data. IRCs that were on track got ramped earlier; those that fell behind in 2H 2025 ramped later. Most are ramping pretty well now, which is why we see inventory/selection expanding now. This is not substantiated by anything, but it feels like the dynamic of inventory growth < retail unit growth probably resulted in a ~300-500 bps headwind on retail units, all else equal. So they probably should have been in the low 40s were it not for the recon issues suffered in Q4 last year. They said so many times on the call that they are ramping and inventory will grow faster than sales. If true, inventory acceleration should drag the growth rate higher and we should see acceleration in unit sales. It's early, but this is starting to show up already. This acceleration also should come at superior unit economics given the improvements in recon. Finance GPU was hurt by a mix effect and a timing effect in Q2. Higher prime mix = lower GoS %. Rapidly rising rates in March/April compressed GoS for loans that were seasoning on their BS. Neither is a long-term concern or a change in underlying economics on a like-for-like basis. $CVNA's guidance philosophy is silly and obviously they are targeting ~10% above the top end, which is reasonable given the unit economics and growth expectations. They've used this silly policy since 2024. Each Q2, the same questions come up. "Why does your guidance imply the business worsening in 2H?" Because the guidance policy is silly, plain and simple. They give a range even though the bottom end is almost impossible. Midpoint is ~2x 1H EBITDA even though they will grow units q/q in Q3 and Q4. The business has positive operating leverage and Q3 should see better GPUs, all else equal, because of the recon improvements in units produce since late May/early June. The BS is in an interesting place. 1x net debt/EBITDA with this kind of cash generation is very strong. Ratings agencies seem intent on discounting finance-related EBITDA, so the company shifting GPU from finance to retail should help them, all else equal. They are on a path to IG which should open up the business to lots of different financing options. They have $3.6B of inventory while using $126m of floor plan. It's absurdly conservative. I'm not sure they get proper credit for this. Lastly, the new car initiative is really interesting and I can't wait until they share more. NPS for Sell to Carvana is the highest in the company. Excellent prices x best experience = huge win for customers. No reputational risk to CVNA - just a great transaction. I've done it a few times - it's amazing. New cars are similar. Best prices w/ no negotiations x best experience should = big NPS. I see an average of ~$6k off MSRP. No reputational risk on unit quality, no need for scaling production. Both mega-NPS businesses generate nice sources of inventory, to be sold through the used business. New car unit economics are probably ~similar to used, but the mix is different. On balance, I would guess the delta vs. used GPUs would be: Lower retail GPU, higher finance GPU (due to higher ASPs, partially offset by higher prime mix), higher other GPU (OEM incentives, higher attach rates on other products, etc.). It will be interesting to see what else they can build on top of this platform. The network nodes should get closer and closer to end customers the bigger they get. It should improve NPS across every business, improve unit economics, and unlock new businesses and capabilities.

Hermes almost back to flat over the last 5 years. Luxury has had some big multiple compression. Quality at any price doesn’t work so well, it turns out.