Anthropic’s latest numbers from its IPO prospectus. Revenue was $16.3bn in the first half of 2026, about 14x the same period a year earlier. Compute and infrastructure took about 55 cents of every revenue dollar. The headline net loss is $80.7bn, but $72.4bn of that is a non-cash fair-value charge on convertible notes and SAFEs. The operating loss was $693M, and interest income on its cash was $1.1bn. In other words, the interest on its cash pile was larger than its operating loss. Operating cash flow was positive at $3.9bn, a year after burning $1.5bn. The balance sheet shows $112.4bn of cash and $108.6bn of convertible notes. The preferred stock carries a $136.3bn liquidation preference, meaning that much has to be paid back to preferred holders before common shareholders see anything in a sale. Deferred revenue, which is cash customers paid upfront, went from $269M to $1.65bn in six months

Only 3 of the 28 US metros on Apollo’s latest chart are still forecast to be building rentals above their own long-term average: Tampa, Philadelphia and Austin. Earlier this month, Apollo wrote that rents were re-accelerating as higher rates led builders to build less, and called it a “higher rates, higher rent doom loop” for the Fed. Apollo's latest chart shows the other side of that story: The post-pandemic building boom that helped hold rents down is fading, and new supply is back to normal almost everywhere. New buildings take 12 to 18 months to lease up, so Apollo expects upward pressure on rents and shelter inflation into 2027. With inflation already above the Fed’s 2% target, and rent being a big part of inflation, I think this makes it harder for the Fed to cut rates any time soon

Goldman Sachs David Solomon: “The economy is fine, negativity just sells.” “$6.50 for a gallon of diesel is a real headwind, we have to work through that. I understand why people are frustrated about that.’ “If AI is going at a pace where it's creating more disruption in the labor force then the public and private sector have the responsibility to create policy that smoothes that out and provides support." “I don't think AI regulation is a binary choice, we have to allow and be responsible for innovation.” David Solomon admits that the Iran War has been a headwind for gas, transportation, and supply chains.

Shareholder proposals at Japanese companies have more than tripled since 2018, according to Apollo, from fewer than 100 to over 400 a year at the peak. Apollo says that stronger focus on governance and shareholder value is starting to show up in profits. Total ordinary profits for Japan’s non-financial companies have climbed to record levels, at about ¥45 trillion in the latest quarter. It isn’t only the largest companies. Profits at mid-size firms, those with ¥100 million to ¥1 billion of capital, are also at historically high levels, at close to ¥6 trillion a quarter. Is shareholder activism the main driver here, or one of several?

Big Short investor Steve Eisman shoots down Elon’s SpaceX - Tesla merger as a “very bad idea.” “SpaceX is already complicated enough. Now, it will merge with a company Tesla that has an electronic vehicle business that lost massive market share to the Chinese, an autonomous vehicle division that is questionable, and a solar business that is just not that great.” “My concerns about SpaceX is the capex, total capex was 29.6 billion, up 185% QoQ. They are telling you they have massive capex plans in the space segment, grok model and data centers, and Starlink pivoting to compete with T-Mobile.” "For all these plans SpaceX has to raise billions in capital [...] the cost to raise this capital just got more expensive.”

Mergermarket: Defence M&A has hit a record $50bn so far this year, led by deals in the US and Germany. Funding rounds for defence tech start-ups have also hit record sizes globally, averaging $104m this year compared with $67m in 2025. More broadly, crossborder M&A reached $1.2tn in the first nine months of 2026, and inbound M&A into the US rose 4%, with 1,109 deals worth $295.5bn.

When I was an analyst, I worked on a deal where the hardest question wasn’t the building, it was what the stuff inside would be worth in five years. That came back to me looking at this. Add up BofA’s line items and a data center costs ~$43mn per MW, and ~$31mn of that is servers. Power, cooling and construction, the parts that last for decades, are only ~$7mn. For lenders, that’s the key detail. A lot of AI infrastructure financing is described like real estate, but most of the cost sits in chips that can lose value fast as newer generations ship. The vendors in this graphic get paid upfront. Whoever finances the servers carries the risk if they wear out faster than the loan

Notable chart from Apollo today: 80% of the top 1% of AI spenders are VC-backed companies, per Ramp data, while the median spender is only 13% VC-backed. A lot of the heaviest AI spending is coming from investors’ money flowing through startups, not from ordinary businesses paying out of their own profits. Apollo calls it circularity. That’s fine while funding keeps coming, but it’s worth asking how much of today’s AI demand depends on it

Was talking to a friend of mine who's a Banking analyst and he told me he and a lot of his peers are getting pretty annoyed with how AI is expanding the breadth of their coverage but decreasing the depth. For example, they may be asked to do more tasks with the expectation that AI will help them, but it only really does the first few simple steps. In order to keep up, they have to just move on and settle for the superficial work. Not intellectually stimulating for them at all. More work, less depth and worse learning.

Blackstone CEO Stephen Scharzman recounts his early discovery of AI with Alibaba’s Jack Ma: Ma told him about the potential impact of this new technology long before labs became mainstream. He stated that scholars even told him “Jack’s underselling this” as the most consequential technological change of the century. “We bought a very large data center in 2021 for 10 billion dollars, it's now 15 times larger, and we are the largest owner of data centers in the world, putting us in touch with 6 hyperscalers.” His advice and model for Blackstone’s success: “We have a system where we learn information every week, part of it is having curiosity because those people that don’t fit tell you the future, and the future is key to success in the present.” This logic explains Blackstone’s current electricity investments as well.

Very true

Lol

But wouldn’t it come up in reviews

Back when I was an intern we were going through a small group training dedicated specifically to my group. In the excel template we were using, there was an error that the associate running the training pointed out to us. It was something that would be super hard to notice if you weren't looking for it. He chalked it up to "everyone being too lazy to fix it in the master file." I figured it was strange, noted it, and moved on. Come the end of the Summer for final presentations, 2/3s of the interns in my group failed to fix the error in the master template in their own copies. Not all of them didn't get returns, but all that didn't get returns were in that group. When I went back full time, I offered to fix the error in the master doc, but it turns out that it's intentionally left there for the interns to see who can remember the simple stuff. Nowadays I bet Claude would catch it in one pass.

L40 ran the payback math on how much a PE buyer can pay for a software company, and the gap is bigger than most founders expect. Two software companies each make $10M of EBITDA today. A PE buyer can pay about 5x for one and about 11.8x for the other. If EBITDA stays flat, five years of earnings adds up to about 5x today’s EBITDA. If it grows 30% a year, five years adds up to about 11.8x. One active software acquirer told L40 it needs deals to pay back within four to five years, or six at most. So the buyer is paying for the earnings it expects to get back over that window, not for what the company makes today. That’s why two companies with the same EBITDA can get very different offers. Would you pay 11.8x for 30% growth you have to trust will last five years?

Fake it till you make it: > Second year analyst at a good bank > Completely shitting the bed (making mistakes, not meeting expectations) > Inefficient > Despite this, still get neutral/positive reviews from Seniors > Not sure whether to stay or go Sometimes, having a good attitude, showing improvement, and willing to learn is all it really takes

81% of US firms told the Richmond Fed that neither the cost of financing nor access to it has constrained their investment or spending plans, per Apollo. Those are two different things. A company can find rates expensive and still have no trouble raising money, or the other way round. When both are fine for four out of five firms, it’s a fair sign that credit is flowing. That’s a different story from what you’d expect after years of higher-for-longer rates

Don’t shoot the messenger

Another cringe Blackstone video

Middle-aged Korean men with their entire net worth