"Sport is tremendous, I had a long career but my passion for sport is intact, my passion for human beings is intact."
He changed football and still considered a mentor by the world stars who played for him.
A man who truly understands #Laureus19 Sport for Good: Arsene Wenger
Very important podcast that I think is a must listen given the current moment.
>Rising GPU rental prices are being read as AI's most bullish chart, but tighter credit is doing much of the work. Silicon Data's B200 benchmark rose 23.5% from August-end to November 19 ($5.62 to $6.94/GPU-hr, peaking at $7.10 on Nov 17). On Odd Lots, Stanley Lee (Liquid Compute) and Paul Kedrosky (SK Ventures) lay out how a financing squeeze can push spot prices up while making the underlying buildout more fragile.
How credit stress gets misread as spot demand (Stanley Lee)
>Lenders are pulling back on neocloud financing. Midmarket GPU loans (several hundred million dollars, 3-year fully amortizing) saw advance rates drop from ~80% over the summer to ~70% post-Labor Day, and now 50-60% at mid-to-high-teens borrowing costs.
>The equity math is getting harder. On a $100M cluster, a drop from 70% to 50% LTV increases the capital required outside senior debt from $30M to $50M (+67%). Some stressed operators are now seeking 3-to-6-month bridge loans just to stay funded.
>Lenders are flagging both execution and concentration risk. A take-or-pay contract is only viable if the neocloud can deliver usable capacity and meet SLAs, leading at least one lender to decline first- and second-deployment operators. Lenders are also seeing the same few AI offtakers across multiple neocloud loan books, creating concentrated exposure even across different operators.
>Failed term deals are funneling buyers into the spot market. Lee gives an example of a buyer willing to pay $4.30/hr for B300s with 30% upfront whose contract the operator still cannot get financed. That buyer still needs compute today, so they get pushed into short-term capacity.
>Buyers are also hoarding cash. Startups are choosing to pay mid-to-high-$5/hr for B300s with 0% prepayment rather than $4.40/hr for 3 years with 30% upfront. Preserving cash lets them deploy compute immediately and show faster growth ahead of their next VC raise.
>The combined effect obfuscates the real signal. When operators cannot finance long-term contracts and buyers refuse to lock up upfront cash, demand crowds into the short-term market while fewer new deployments get funded. The market sees a +23.5% move in spot B200 rates and reads it as unconstrained demand, when it is partly a byproduct of financing friction.
=Lender recovery is much harder than holding a hardware lien. Recovering value in a default requires a working cluster that can be re-let in place, which means facility access, an operator to keep it running, and enforceable rights to transfer contracts (while existing customers may demand non-disturbance clauses to keep below-market rates). Liquid Compute is fielding 1-3 private-credit inquiries per day for indicative pricing on OTC B300 rental-index puts, but an index floor only hedges the reference rate, not downtime, operating costs, facility access, or basis risk.
Buildout returns and enterprise adoption (Paul Kedrosky)
>Kedrosky's stress test puts ~$10T in cumulative capex against a $3-6T annual revenue hurdle by ~2032. Consumer subscriptions cannot carry that load, meaning enterprise spend has to either displace existing budgets or generate new output.
>Sizing enterprise TAM off today's white-collar wage bill overstates the opportunity, because AI lowers the unit price of the tasks it automates.
>Coding is a poor template for broader enterprise rollout. Software has fast, deterministic feedback loops (tests and execution), whereas most corporate workflows have fuzzier quality metrics. Productivity gains exist outside coding (a study of 5,172 support agents showed +15% issues resolved per hour with AI), but scaling autonomous workflows is slower.
>Moving into vertical applications like legal, finance, and healthcare (e.g., Anthropic) can improve near-term inference monetization, but it creates channel conflict with cloud and application partners and forces trade-offs against frontier training spend.
The same credit shock hits GPU rents in opposite directions across the cycle. Initially, buyers shut out of term financing bid up scarce spot capacity while fewer new clusters get funded. If VC fundraising later slows enough to curb buyer spend and force operators to re-let defaulted clusters, demand drops right as available supply rises.
The underwriting test for this cycle is how much of the compute bill, including recurring training spend and fixed capacity commitments, can be paid from collected third-party customer cash if the fundraising market dries up.
Martin Ødegaard is the only player in the runs the game quadrant in the 2026/27 Uefa Champions League season.
He's been the best ball progressor and creator in the UCL so far.
🚨 Mikel Arteta when asked if Ødegaard can be next player to sign new deal: “Yes, I want Martin Ødegaard to stay here”.
“He’s a key player for us. Martin is having an upward trajectory here at Arsenal”.
To give another example of how much pens warp public perception wrt scoring from midfield:
Over their PL careers to date, Bruno Fernandes has scored NPGs at a lower rate p90 (0.20) than Ødegaard (0.22).
Øde is only 9 NPGs behind Bruno in >6k fewer minutes of football.
@alexfrco@Blahovic@theiconic15@WindUpWounded Also the way people think of Ramsey as a goalscoring midfielder but Odegaard as not someone who provides goals is completely incongruent if you look at the actual data..
@alexfrco@GMCG9899 Also doing this in a predominantly 4-4-2 is much tougher than a 4-3-3 or 4-4-1-1 attacking position that Gerrard or Lampard did it in
@SomethingRibald They end up having some pods which are very treasury/ hedge fund focused that go over my head and I skip but they are very good for the majority of it
@SomethingRibald For instance, they did a pod the day before midterms where Nate Silver came and basically said Dems are going to do a whitewash when no one had the guts to say it. And then next episode they were debating how Megaspeed operates and it was a really good pod explaining it
@TomCantonMedia Literally the last season when he didnt play he still led in assists and you think there is a team where we have worked without O.. Please name the starting XI..
Big decisions to be made on Kai Havertz and Martin Odegaard contracts amid major push to renew Arsenal squad - @Football_LDN
Official talks have yet to begin with either player. Both happy at the club but a key moment in high-earning 27yo's futures.
https://t.co/R20UMYaD5Q