Michael hochstat

Knowledge is free knowledge.dotadda.io/session…

A $22 billion credit manager is getting hit hard. The bigger story is what happens when buyers of distressed debt stop buying. Arini Capital Management’s flagship fund is down roughly 26% in 2026, including a 12% loss in October through October 23. The fund had $7.5 billion under management at the end of July. It is on track for its first annual loss since launching in 2022. Hamza Lemssouguer built Arini into one of Europe’s most aggressive distressed-credit investors. Now, according to Bloomberg, he is focusing on existing positions rather than aggressively pursuing new ones. That matters because markets need buyers when debt starts falling. Look at the underlying positions. Aston Martin bonds fell from roughly 55 cents on the dollar at the start of October to around 47 cents. Ritz-Carlton Yacht Collection debt dropped from above 92 cents at the end of September to below 75 cents. Altice International and Brightspeed are also under pressure. These are signs of stress in the weakest part of the corporate debt market, where leverage, refinancing risk and collateral values can determine whether creditors recover their money or suffer substantial losses. Arini is not alone. Sona Asset Management’s flagship fund reportedly lost about 3.1% in the first half of October, erasing its gains for the year. Here is what matters. When a major distressed-debt buyer retreats, liquidity can disappear just as troubled companies need it most. Other investors become reluctant to buy the same securities. Prices fall. Financing gets more expensive. Refinancing becomes harder. Paper losses can become permanent losses of capital. European CCC-rated corporate bonds have suffered an unusually prolonged decline. Investors are reassessing heavily indebted businesses facing higher borrowing costs, weak fundamentals and, in some industries, AI disruption. This is not yet proof of a systemic credit crisis. Stress remains concentrated in the riskiest borrowers, and a fund’s losses do not automatically mean it faces a liquidity crisis. The question is whether the damage spreads. Watch credit spreads, refinancing failures, distressed-debt prices and whether other large credit funds are forced to cut exposure. In a bull market, leverage and concentrated bets can make a credit manager look brilliant. When the market turns, those same positions can become impossible to exit without pushing prices lower. The real risk is not that Arini lost 26%. It is that a major buyer is pulling back just as the weakest borrowers need capital. That is how distress in one corner of the debt market can spread.

Wall Street’s tax-loss selling trade: Which stocks could get hit next? The data has always been public. Stock prices, year-to-date performance and the biggest losers are not secrets. The edge is knowing how to read them. As October ends, investors are watching the stocks that have been crushed this year. Those losses can create selling pressure as investors harvest tax losses, rebalance portfolios and clean up positions before year-end. Here are five S&P 500 stocks to investigate: AppLovin (APP): −61.4% YTD CoStar Group (CSGP): −59.0% Fair Isaac (FICO): −58.7% Boston Scientific (BSX): −55.5% Lululemon (LULU): another major decliner flagged in tax-loss-selling research. The percentages are reported figures, not verified live quotes. The ranking is a starting point, not proof that mutual funds are selling these stocks specifically for tax reasons. Then there is Google. Alphabet (GOOGL) sold off sharply Thursday following earnings. The market is reassessing the cost of AI infrastructure, the cash required to fund it and the returns shareholders can realistically expect. That is an earnings-driven repricing, not necessarily tax-loss selling. But it belongs on the same research screen: a major decline, a change in expectations and a market trying to determine what the business is worth. Here is what matters. The opportunity is not finding stocks that are down the most. It is finding stocks where forced or calendar-driven selling may push the price below intrinsic value. That requires separating three things: Tax-loss harvesting and portfolio rebalancing. Deteriorating business fundamentals. A valuation reset that may have gone too far. The market data is public. The analysis is where the work begins. And one important correction: October 30 is not the universal tax-loss deadline. U.S. investors generally have until December 31 to realize capital losses for the 2026 tax year. The question is simple: Which of this year’s biggest losers are genuinely impaired businesses, and which might become opportunities if year-end selling pushes them too far?

Palantir CEO Alex Karp is raising a question the AI industry would rather avoid. Who benefits when companies get cheap AI tokens? Karp argues that discounted tokens may come with a hidden cost: access to proprietary data, workflows and business know-how that can make the model better. That changes the economics of AI. A company feeds a model its internal information, operating procedures and proprietary methods. The model provider potentially learns from that interaction, depending on its data-use terms. The customer gets cheaper inference. The provider may gain a better product. Who captures the long-term value? For an investment firm, it could be a research process. For a manufacturer, years of operating expertise. For a software company, the workflows that differentiate its product. These aren’t just data points. They can be competitive advantages built over years. But there’s an important distinction: sending data to an AI provider does not automatically mean that data is used to train its models. The answer depends on the provider’s policies, contract, product and configuration. That is precisely why disclosure and control matter. The AI buying decision cannot stop at the price per million tokens. Companies need to understand what happens to their data, whether it can be used for training, how long it is retained and who controls the resulting intellectual property. Cheap tokens may be a great deal. Until you discover what else you’re giving away. The real question isn’t just what AI costs. It’s what your company is contributing to someone else’s competitive advantage. #AI #EnterpriseAI #DataSovereignty

Governments aren’t banning AI. They’re deciding which AI their employees are allowed to use. Sweden is the latest example. As of October 26, Sweden’s Government Offices blocked ChatGPT and Claude on staff computers, directing employees toward Microsoft Copilot instead. The distinction matters. This isn’t necessarily a rejection of the technology. It’s a question of who controls it, what happens to the data, and which services have been approved for government work. The pattern is showing up elsewhere. The European Parliament restricted built-in AI tools on lawmakers’ work devices over concerns about the security of information sent to AI providers. In 2025, governments and public agencies also restricted DeepSeek on official devices, citing security concerns. But the picture isn’t uniform. The US government has also pursued deals to make commercial AI models available to federal agencies. Here’s the bigger issue. Employees want the best AI tools available. IT departments want security, oversight and predictable data handling. Governments want all three, without losing control of sensitive information. That creates a new dividing line in the AI market. The winning product inside a government agency or large corporation may not be the smartest chatbot. It may be the one that passes security review, meets procurement requirements and fits the organization’s existing software contracts. There’s an important distinction between banning an AI model and blocking access to its public website. An organization can prohibit employees from using ChatGPT directly while still permitting approved AI services under different terms. That’s why this is worth watching. The next phase of AI adoption may be determined as much by procurement departments, security policies and data agreements as by model performance. The question is no longer simply which AI is best. It’s which AI your organization will allow you to use.

The important angle here is that AI is moving beyond answering questions toward doing the work inside a single environment. The distinction is no longer just how smart the model is, but whether it can use your files, operate your tools and keep working after you walk away. ANTHROPIC IS TURNING CLAUDE FROM A CHATBOT INTO A WORKER The most important AI product shift isn’t another model benchmark. It’s the move from answering questions to completing jobs. Anthropic is combining Claude Chat and Cowork into one experience. You can ask a quick question, then hand Claude a larger assignment in the same conversation. Have it research a subject, read files, run code, analyze data and produce a finished report or presentation. It can continue working in the cloud after you close your laptop, while accessing local files and apps when you explicitly grant permission. Anthropic is also bringing document editing, slide creation and design into the workflow. Why this matters Today, most software still expects a human to connect the dots: open applications, find the data, move information between tools, create the analysis and format the result. AI agents are beginning to take over that sequence. The product is no longer just the model. It’s the combination of the model, tools, context, permissions and ability to deliver a finished result. That changes the competitive question. Who has the best chatbot matters less if the customer increasingly wants a completed task rather than a conversation. The business question Can Anthropic turn this into a reliable daily workflow that customers will pay for—and keep using? The hard problems remain execution quality, security, operating costs and trust. A polished demo is easy to admire. A system that consistently finishes real work is much harder to build. But the direction is clear: AI is moving from a destination you visit to a layer of work that follows you across applications. The next battle isn’t just over who has the smartest model. It’s over who becomes the place where the work actually gets done. Anthropic described the unified Claude experience in its product announcement and help documentation. Claude by Anthropic +1

Just had one of the worst bagels I can remember at Zaro’s. The cream cheese left a horrible aftertaste, and the bagel itself was awful. Seriously, what happened to the bagels? They used to be something you could count on. This one was disappointing from start to finish. Bad bagel. Bad cream cheese. Terrible aftertaste. What happened to Zaro’s?

HIGH-YIELD CREDIT: THE WARNING SIGNS ARE AT THE BOTTOM October 28, 2026 The U.S. high-yield market looks relatively calm. Beneath the surface, credit risk is becoming increasingly concentrated. The headline spread tells one story. Credit quality tells another. The ICE BofA U.S. High Yield Index yields approximately 8.1%, with an option-adjusted spread near 330 basis points—wider than the 2026 tights, but still below the long-term average of roughly 500 bps. Now look at the breakdown: BB: 190–200 bps Single-B: 330–340 bps CCC and below: 1,250+ bps The weakest borrowers face yields in the high teens, yet CCC debt represents only 8–9% of the index. That limited weighting helps mask the stress beneath the aggregate numbers. This is credit decompression—and a potential late-cycle warning. Three developments deserve attention: 1. Refinancing risk. Companies that borrowed cheaply must refinance at today’s higher costs. For highly leveraged borrowers with weak cash flow, the math can become unforgiving. 2. Rising credit stress. Reported U.S. high-yield default rates have been around 2.6%, with leveraged-loan defaults near 3.9%. The critical question is whether distress remains concentrated in CCC credits or spreads into single-B and BB issuers. 3. Debt versus cash flow. AI-related financing and heavy issuance create another test: Will future cash flows justify today’s borrowing? Capital expenditure is not cash flow, and revenue growth does not guarantee debt repayment. What does this mean for investors? Carry remains attractive, but yield alone is not a margin of safety. An 18% yield may signal opportunity—or a serious probability of default and poor recovery. Higher-quality BB and stronger single-B credits may offer a better risk-reward balance, provided valuations and issuer fundamentals support the investment. My view: The market looks late-cycle, but not yet in crisis. The real warning will come if stress migrates from CCC debt into the broader credit spectrum. Watch refinancing failures, downgrades, interest coverage and spread widening—not just the headline index yield. In credit, the most important signal is often not what the average borrower pays. It’s what the weakest borrower can no longer afford. Market figures are approximate as of late October 2026; default statistics vary by provider and methodology.

THE PUBLIC CLOUD ISN’T DEAD. THE CLOUD-FIRST ERA IS. For 20 years, the technology industry sold companies the same idea: move your data to someone else’s computers. Shut down your servers. Rent infrastructure. Let somebody else handle the complexity. Then AI changed the economics. First problem: latency. AI inference, industrial automation, medical imaging and autonomous systems cannot always wait for data to travel to a distant data center and back. Physics has no customer-service department. Every network hop adds delay. Moving enormous datasets costs money. Sometimes bringing compute to the data is faster and cheaper than shipping the data to the compute. Second problem: control. Companies are reconsidering where their most valuable assets live: proprietary models, customer records, training data and intellectual property. Public cloud can be secure. On-premises infrastructure is not automatically safer. But outsourcing infrastructure creates dependencies, third-party risks and legal complications. The U.S. CLOUD Act also means that storing data overseas does not necessarily shield it from qualifying U.S. legal demands directed at providers subject to U.S. jurisdiction. For sensitive workloads, control over the infrastructure matters. AI makes both problems harder to ignore. The industry is building enormous data centers to centralize computing while simultaneously developing applications that benefit from moving computing closer to the user, the machine and the underlying data. That tension matters. The future is not a wholesale return to corporate server rooms. Hyperscalers still have enormous advantages in scale, capital and infrastructure. The likely outcome is a more deliberate mix of public cloud, private infrastructure and edge computing. Some workloads will stay in the cloud. Others will move closer to home. The investment question is not whether cloud spending continues to grow. It is whether the assumption that every workload belongs in a hyperscale data center survives the AI era. Cloud-first is giving way to workload-first. And that could change where the next generation of computing infrastructure gets built — and who gets paid for it.

AMAZON IS REASSESSING ITS AI BET Amazon’s AI retreat is becoming harder to ignore. The company has closed its San Francisco AGI Lab, reportedly pulled back from several flagship Nova models, and announced that Amazon Q Developer’s IDE plugins will be sunset in April 2027 as developers are directed toward Kiro. But the details matter. Amazon has not abandoned all Nova models or AI coding. It launched Nova 2.5 Sonic in October. The bigger question is what Amazon believes it can win. Training foundation models is expensive. Keeping up with frontier-model competitors requires enormous investment, while the commercial payoff remains uncertain. Amazon has another option: sell the infrastructure. AWS can make money providing compute, custom chips and access to models built by Amazon and its competitors. It doesn’t necessarily have to build the best model in every category to benefit from AI spending. That creates a strategic trade-off. Does Amazon need to win the model race, or does it make more sense to sell the picks and shovels to everyone running it? The distinction matters for investors. AI spending can keep growing even as individual models, research teams and products get cut. The winners may not be the companies training the most models. They may be the companies that can turn AI demand into durable cash flow. Amazon’s restructuring doesn’t prove the AI boom is ending. It does suggest that even the biggest players are making choices about where the economics justify the investment.

THE INDEX BUYING MACHINE Apollo CEO Marc Rowan sees a potential pattern in the next wave of AI IPOs. His example is SpaceX. Rowan says the company moved from a private-market valuation above $1 trillion to $2 trillion as it went public, then reached $3 trillion as investors filled out their positions. His explanation: index buying. Once a company enters a major stock-market index, funds tracking that index must buy it. That creates a powerful source of demand, regardless of whether investors can agree on what the business is actually worth. Now Rowan sees the possibility of a similar dynamic with Anthropic. His point is worth examining. “We’re talking about companies that are literally changing the world and whose business models are not predictable at this point.” Then comes the key line: “There’s no right price.” That is the tension at the center of the AI IPO market. The technology may be transformative. The businesses may eventually generate enormous cash flows. But neither fact establishes what investors should pay today. And when index inclusion creates a wave of forced buying, market price and fundamental value can move in different directions. The question isn’t whether AI changes the world. It’s how much investors will pay for that possibility — and who will be left holding the stock when the buying subsides.

Autopilot Blues An island shuffle — steel drums, a lazy backbeat, and a little salt in the air. Verse 1 Well, I’ve got a little helper, never sleeps, never tires, Keeps the heading steady while I’m sitting up higher. Reads the instruments, holds the line, makes flying look like fun, And I’m kicking back and cruising like the hard part’s done. Chorus Oh, autopilot, autopilot, carry me along, Like a trade wind rolling to the rhythm of a song. But when the sensors ice up and the numbers don’t make sense, You’d better know your way around the controls and the instruments. Verse 2 June of ’09, from Rio, through the midnight blue, Flight 447 was crossing where the Atlantic winds blew. Ice crystals fouled the pitot tubes; the airspeed readings strayed, The autopilot disconnected. Now the pilots had to fly. Chorus Oh, autopilot, autopilot, carry me along, Like a trade wind rolling to the rhythm of a song. But when the sensors ice up and the numbers don’t make sense, You’d better know your way around the controls and the instruments. Verse 3 There were alarms inside the cockpit, confusion in the night, The aircraft climbed, then lost its lift, beyond the pilots’ sight. The stall warning sounded, but the danger wasn’t clear, And the ocean kept on waiting in the darkness far below. Bridge — slow, steel drums drop out Two hundred twenty-eight souls, Lost above the sea. A lesson written in the dark For you and me. Trust the tools that help you, Let them lighten what you do. But when the machine says, “Take control,” You’d better know what to do. Verse 4 Now we’ve got these thinking engines, writing words and drawing plans, Reading all the numbers, lending knowledge to our hands. They can help us find the answer, they can save us hours a day, But a confident machine can still get lost along the way. Final Chorus — full island groove Oh, autopilot, autopilot, help me on my way, Chart the course and clear the skies; make the work a brighter day. But keep your eyes upon the gauges, keep your judgment in command, ‘Cause the tool can help you fly, but it can’t replace the man. Oh, autopilot, autopilot, let the steel drums play, I’ll take the help you offer, but I’ll think along the way. When the numbers stop making sense and the answers turn to lies, I’ll be more than just a passenger who’s staring at the skies. Outro — steel drums fade Easy now… easy now… Let the machine help you along. Just don’t forget who’s flying. Fade out with the song.

If AI is like the autopilot on Air France Flight 447, the winners won’t necessarily be the companies with the best AI. They’ll be the people who know when to trust it, when to question it, and when to take control. That is the connection between AI and Flight 447. On June 1, 2009, Air France Flight 447 crashed into the Atlantic. Ice crystals temporarily disrupted the aircraft’s airspeed readings. The autopilot disconnected. The pilots had to take over, but their responses to confusing instrument readings contributed to the aircraft entering an aerodynamic stall. The plane had functioning systems, but the crew failed to recover from the stall. The official investigation identified multiple interacting technical and human factors. The lesson for AI is striking. Who wins in the AI economy? 1. The people who understand the underlying business. AI can summarize a company’s earnings call. A good analyst knows whether management is dodging a question about cash flow. AI can write code. An experienced engineer knows when the code is fundamentally wrong. AI can recommend an action. Someone with domain expertise understands the consequences of getting it wrong. 2. The companies that own the workflow, not just the model. A model that produces impressive answers is not necessarily a profitable business. The durable opportunity may belong to companies that integrate AI into real work, verify its output and deliver a measurable result. 3. The people who know when the autopilot is wrong. Automation creates a dangerous paradox: the better it works most of the time, the less prepared people may be when it fails. The competitive advantage isn’t simply using AI. It’s knowing when not to trust it. 4. The investors who distinguish productivity from profits. A company can spend billions on AI without earning an adequate return. The winners will be those that turn automation into higher cash flow, better margins or revenue that customers actually pay for. The technology’s capabilities are not the same thing as its economic value. My conclusion: AI is the autopilot. Human judgment is the ability to recognize when the instruments no longer make sense. The winners will combine both. The losers may be the people who confuse a machine that usually works with a machine that cannot fail.

AI IS ON AUTOPILOT. FLIGHT 447 SHOWS WHAT CAN GO WRONG. On June 1, 2009, Air France Flight 447 crashed into the Atlantic Ocean. All 228 people aboard died. The Airbus A330 was flying from Rio de Janeiro to Paris when ice crystals obstructed its pitot tubes, disrupting airspeed measurements. The autopilot disconnected. The pilots had to take control. They struggled to interpret the conflicting information. Incorrect control inputs pushed the aircraft into an aerodynamic stall. The crew failed to recover before the plane hit the ocean. The entire sequence lasted only minutes. The official investigation identified a chain of failures involving unreliable airspeed indications, pilot responses, training, cockpit design and the failure to recognize and recover from the stall. This was not simply a case of technology failing. It was a failure of the relationship between automation and the people expected to take over when automation stopped working. Now consider the AI industry. AI can produce an answer in seconds. It can analyze thousands of documents, write a report and sound completely confident. But what happens when the data is wrong? What happens when the model reaches the limits of its knowledge? What happens when the human using it no longer knows enough to recognize the mistake? And what happens when a company builds its entire workflow around a system nobody knows how to operate without? Flight 447 offers a warning, not a direct equivalence. An aircraft autopilot is not a modern AI model. But both raise a fundamental question about automation: Are we building systems that make people better at their jobs, or systems that leave people less prepared to act when something goes wrong? In aviation, the ability to recover matters as much as the ability to automate. In business, the same principle applies. Use AI to accelerate research. Verify the evidence. Understand the limitations. Keep a human accountable for the decision. Because the most dangerous moment may not be when the machine stops working. It may be when the human being no longer knows what to do next. #ArtificialIntelligence #Automation #RiskManagement

It usually doesn't crash in one dramatic moment. The failure is quiet: a wrong answer that looks just like a right one, so nothing signals the problem until the damage is done. The crash point is when three things line up: •High stakes: the output feeds into a medical, legal, financial, or safety decision. •No checking: nobody verifies it, either because it's automated or because people have learned to trust it. •Compounding: one error becomes the input for the next step, or thousands of people rely on the same mistake at once. Autopilot has the same pattern. Planes rarely go down because the system fails loudly. They go down when it fails subtly and the pilots have stopped watching the instruments. Air France 447 is the classic case, where the crew was handed control back in a confusing situation after years of relying on automation. So the practical line is how much damage one unchecked error can do. For brainstorming or drafting, it's low, so let it fly. For anything where a wrong answer is expensive or irreversible, keep a human hand on the controls.

AUTOPILOT FOR THE APOCALYPSE The robot is flying the plane. The pilot is checking the dashboard. The passengers are arguing about the algorithm. And somebody in First Class is selling subscriptions to the seat belts. Data aggregation, workflows, and alignment may be the cruising speed of AI. The machinery that keeps the whole operation moving while everyone stares at the glowing screen. But here’s the funny part. When you’re 35,000 feet above the Atlantic, you don’t particularly care how elegant the autopilot is. You care whether somebody is watching the engines. You care whether the instruments are telling the truth. And you care whether the person in the cockpit knows what to do when the computer starts hallucinating a runway. AI can gather the data. AI can build the workflow. AI can write the memo. AI can even explain, in 14 beautifully formatted paragraphs, why the plane is absolutely, positively not going down. But who checks the work? Who questions the assumptions? Who notices that the numbers don’t add up? And who has the authority to grab the controls? Autopilot is a useful metaphor. Right up until somebody mistakes it for a pilot. So, by all means, let the machines fly. Just don’t fire the crew, sell the cockpit, and tell the passengers that human judgment is an obsolete feature. And please, for the love of all things silicon, keep your seat belt fastened. Turbulence ahead. The robots are in charge of the playlist.

Why Boeing stock is sinking today Boeing isn’t facing an immediate collapse. Investors are reacting to something simpler: the recovery is taking longer than promised. The biggest concerns: • Boeing’s cash-flow recovery has effectively shifted about a year to the right.
• 737 production may not stabilize at 47 aircraft per month until 2027.
• The 52-per-month target has moved to the second half of 2027.
• 787 deliveries are tracking toward the low end of guidance.
• Commercial Airplanes still reported a $697M operating loss and -6.5% margin. There were positives: Boeing generated $464M of quarterly free cash flow, advanced key certifications and ended with a record $745B backlog. But the market no longer gives Boeing credit for long-term promises. It wants consistent production, improving margins and cash generation now. Bottom line: Today’s selloff is less about Boeing’s survival and more about execution credibility. The backlog is there. The question is how quickly—and profitably—Boeing can turn it into delivered aircraft and cash. #Boeing #Aerospace #Investing #Markets

Google’s Anthropic deal may be one of the smartest structures in AI. Google didn’t just sign Anthropic as a cloud customer. It created multiple ways to win: 1Investment upside: Google invested through equity and convertible notes, giving it exposure to Anthropic’s rising valuation. 2Cloud revenue: Anthropic plans to access up to 1 million Google TPUs, potentially generating substantial infrastructure revenue for Google Cloud. 3Strategic insurance: If Claude gains share against Gemini, Google can still benefit as an investor and infrastructure provider. Alphabet disclosed that, during the first nine months of 2025, it converted $3.0 billion of convertible notes into equity and recognized a $418 million conversion gain. It also made another $1.1 billion of convertible-note investments. The filing does not identify how much of these amounts related specifically to Anthropic, so the figures should not all be attributed to the company. But they illustrate why the structure is powerful. The potential flywheel looks like this: Google provides capital → Anthropic scales → Anthropic purchases more compute → Google earns cloud revenue → Google retains equity upside. This isn’t just a supplier relationship. It is a combination of: •Venture investment •Infrastructure contract •TPU ecosystem expansion •Strategic hedge against a major AI competitor Google is effectively positioned to benefit whether customers choose Gemini or Claude—as long as more AI workloads run on Google infrastructure. Amazon also has significant exposure through AWS and Trainium, so this is not an exclusive Google win. But structurally, Google found a smart way to monetize both sides of the AI competition. #Google #Anthropic #ArtificialIntelligence #CloudComputing #VentureCapital #TPU #AIInfrastructure #Investing