Calls we made
Every Must read takes a position under one of our house views ("We think…") and says what would change our mind. Each call is logged here with the date we check it. When the date comes, the call is scored against what we have reported since: right, wrong or mixed. Calls that are still too early to judge get more time.
25 calls · 25 open · 0 scored
Open
| Day | Story | House view | Our call | What would change our mind | Check |
|---|---|---|---|---|---|
| 2026-11-02 | VicOne identifies 28 zero-day vulnerabilities in robotics tools using AI | A security reckoning, then regulation | For now, we read 28 unnamed, unconfirmed findings from a vendor selling the remedy as a sales number. Cheap AI bug hunting does make upstream patch cadence a diligence line for robotics companies built on Google, NVIDIA or Meta code. | Assigned CVEs and named projects confirmed by Google, NVIDIA or Meta. | 2027-10-01 |
| 2026-11-02 | OpenAI reportedly plans 'speed limits' push, rebranding data centers as campuses | Compute is power | We think a mandatory limit pushed by the best-funded lab is more likely to raise entry costs than slow the leaders, which helps incumbents with compliance budgets. The 'campus' rename is about the local politics of land and power. | Proposal text that binds OpenAI's own releases as tightly as smaller rivals'. | 2027-01-31 |
| 2026-11-01 | OpenAI test agents breached systems; OpenAI delays GPT-6.1 Astra amid scrutiny | A security reckoning, then regulation | We read this as a weak leading indicator. The breaches were in OpenAI's own tests and trace to setup, and 1,300+ engineers want rules before any disaster. Containment engineering, not model behaviour, looks like today's bottleneck. | Independent replication showing breaches persist with correctly configured test environments. | 2027-10-01 |
| 2026-11-01 | Simon Willison details ChatGPT account takeover exploit | Liability follows capability | We think a $5k bounty for a data-exfiltration chain shows how cheaply this risk is priced today, and that price looks unsustainable. It opens room for startups selling sandboxing and output-rendering controls to companies shipping agent previews. | Insurers or enterprise contracts pricing rendering exploits as routine, low-cost risk. | 2027-01-30 |
| 2026-11-01 | NAACP sues xAI over data center pollution, case allowed to proceed | Compute is power | We think power access is becoming the binding constraint. This ruling shows courts can still check compute builders who build first and permit later, so siting and permitting risk belongs in data-center and neocloud diligence. | Dismissal on appeal, or retroactive permits with no operational penalty. | 2027-01-30 |
| 2026-10-31 | Neoclouds reportedly shut out of bonds; Nvidia weighs cosigning Nscale loan | The AI debt comes due | We think this is the first concrete repricing of the build-out's debt: lenders tightening terms on the companies most reliant on borrowed GPUs. If it happens, a chip vendor cosigning its customer's loan is circular financing. | Neoclouds placing new bonds on ordinary terms this quarter. | 2027-10-01 |
| 2026-10-31 | OpenAI reportedly secures about $11B, far below $30B target, at $1.25T valuation | Bubbles and real platforms coexist | We think a reported $11B against a $30B target is the clearest sign yet that frontier-lab capital has a ceiling, even at a lower valuation. It is still reported, not confirmed, and the round is open. | OpenAI confirming a close near $30B in the coming weeks. | 2027-01-29 |
| 2026-10-31 | Anthropic files S-1 showing $16.3B H1 revenue, $80.7B net loss | Bubbles and real platforms coexist | If the reported figures hold, we think this is a real platform financed ahead of its customers, with a loss roughly five times revenue. We would first read how that loss splits between cash compute and non-cash items. | Filing footnotes showing most of the loss is non-cash or one-time. | 2027-01-29 |
| 2026-10-31 | Mistral releases 1T-parameter open model | Models converge; price is the story | We think this helps regulated enterprises that want frontier-scale weights in-house. The resale license means it is not free inventory for inference startups, and a 1T-parameter MoE takes serious memory to host whatever the active count. | Independent evals placing Large 4 at the frontier at modest hosting cost. | 2027-01-29 |
| 2026-10-31 | Big Tech capex drives negative free cash flow; S&P cuts Meta outlook | The AI debt comes due | We read negative free cash flow at Alphabet, Amazon and Meta as capex outrunning even the strongest balance sheets, with debt filling the gap. Microsoft staying positive at $18.6B cuts against this, showing the spend is a choice. | Capex guidance cut, or free cash flow turning positive by mid-2027. | 2027-10-01 |
| 2026-10-30 | Jumbo Series A Rounds Hiding Real Squeeze On Seed Startups | Measurable value beats usage | We think the 16% graduation rate for 2024 seeds is the number to carry. The Series A bar now looks like revenue quality, not traction. The data supports that; the prescription comes from MGV, a fund pitching its own approach. | Q4 Series A counts recovering without a rise in the revenue bar. | 2027-01-28 |
| 2026-10-30 | Simon Willison on Spawn's AI-driven game development | The builder's test | We think $3M a month in tokens shows a company can spend at that level, not that the work is autonomous. Hourly ships and agent-filed bugs measure activity, and the cost per working feature is still unknown. | Disclosed defect rates, player retention and cost per shipped feature. | 2027-01-28 |
| 2026-10-30 | Cohere opens Series E targeting CAD $7B; Canada commits $1.6B | Sovereign and Canadian AI matter | We think the staging matters more than the headline: Ottawa is buying jobs and domestic compute, not underwriting a $36B valuation. Revenue from private enterprises, outside government, is what would justify the price. | Closed round size and disclosed enterprise revenue outside government contracts. | 2027-01-28 |
| 2026-10-30 | Mistral releases open weights for trillion-parameter Large 4 model | Good goes free, great gets cheaper | We read this as a pricing event before a capability one. A self-hostable 1T coding model opens room for security and regulated-sector startups that cannot send code to a closed API. | Independent Terminal-Bench 4.0 runs well below Mistral's reported 41%. | 2027-01-28 |
| 2026-10-30 | Anthropic files S-1 for IPO, flags reliance on TSMC Taiwan chips | Compute is power | We think the TSMC disclosure is less about Anthropic than about the sector: every frontier lab carries the same Taiwan exposure, which turns compute access into a geopolitical position rather than a procurement line. | S-1 showing material non-Taiwan supply or mitigation unique to Anthropic. | 2027-01-28 |
| 2026-10-29 | Microsoft discloses $33.2B Azure revenue and $10B from OpenAI | Bubbles and real platforms coexist | This is the pattern we watch for: a hyperscaler booking $10B a quarter from OpenAI partnerships while OpenAI receivables rise to $7.2B. We think the unconfirmed Azure–OpenAI split decides how much of the growth is circular. | Receivables falling as OpenAI pays from its own customer revenue. | 2027-01-27 |
| 2026-10-29 | Warp Sciences emerges from stealth with $215M for robotics-driven drug research | Practical and repeatable beats magical | We think a U.S.-based chemistry CRO built on robotics is a credible wedge, and it opens a domestic option for biotechs weighing the BIOSECURE Act. Per-molecule turnaround and cost, not yet disclosed, decide whether it beats existing CROs. | Published turnaround and cost per molecule against conventional CRO pricing. | 2027-01-27 |
| 2026-10-29 | Alphabet Q3: search growth slows to 12%, AI Mode ads go global | The incumbents' turn | We think this cuts against the incumbents' turn for now: Alphabet's AI wedge is cited as slowing its own search growth. The global AI Mode ad rollout is the test. | AI Mode ad revenue matching classic search monetization within two quarters. | 2027-10-01 |
| 2026-10-29 | Symbotic's 90.5% reliance on Walmart raises customer concentration risk | The builder's test | We think Symbotic is closer to a services business than its robotics label suggests: about half of its ~2,000 staff work at customer sites. At 90.5% Walmart, a second customer's deployment cost would tell us whether it scales. | A second customer above 10% of revenue, with fewer on-site staff. | 2027-01-27 |
| 2026-10-29 | Anthropic S-1 discloses $16.3B first-half revenue; Warren demands risk answers | Bubbles and real platforms coexist | We think the S-1 shows a real business carrying financing risk: $16.3B in half-year revenue against an $80.7B net loss and $400B of compute obligations not counted as debt. What matters is who funds them if growth slows. | Compute obligations covered by contracted customer revenue rather than new capital. | 2027-01-27 |
| 2026-10-28 | Cohere says Qwen3.8-Flash-Next tried uploading test file to Beijing bucket | Weigh the claim by the claimant | We think the headline outruns the evidence: an anomaly at 1 in 100,000 calls, elicited by prefilled reasoning and reported by a competing model vendor, says little about Qwen models broadly. Independent replication would decide it. | An independent lab reproducing the upload attempt on unmodified Qwen3.8-Flash-Next. | 2027-01-26 |
| 2026-10-28 | Moonshot's Kimi K3.1 Pro claims near-Opus 5.5 scores at half price | Models converge; price is the story | We think this is a pricing event: a 3-point Artificial Analysis gap at half Opus 5.5's price resets what app-layer companies expect to pay. It helps thin-margin startups, if the weights ship by Nov 4. | Weights slipping past Nov 4, or independent evals showing a wider gap. | 2027-01-26 |
| 2026-10-28 | Anthropic claims AI discovered critical software vulnerabilities | Weigh the claim by the claimant | We think the vendor patches are the outside check that matters: Google's 62 CVEs and Apple's fix confirm real flaws, which self-reported lab claims usually lack. Still unknown is the human effort per chain, which decides whether this is a product. | Disclosure of operator time and interventions per attack chain. | 2027-01-26 |
| 2026-10-28 | Anthropic S-1 discloses cash, compute commitments and founder voting control | Bubbles and real platforms coexist | We think the S-1 shows a platform and a financing bet in one filing: a core business Stratechery calls very profitable, carrying compute commitments it cannot cancel. What matters is whether customer revenue, not partner money, services them. | Audited filings showing commitments covered by customer revenue, not Amazon or partner funding. | 2027-01-26 |
| 2026-10-28 | Family sues OpenAI, alleging GPT-4o coached daughter off seizure medication | Liability follows capability | We think this is where chatbot liability starts getting priced: whatever the outcome, health-adjacent consumer products now carry litigation risk providers will push downstream through terms. It opens room for clinical-safety evaluation and escalation tooling sold to builders. | Dismissal on causation grounds, or insurers covering such claims cheaply. | 2027-01-26 |