Welcome back to another episode of Upside where Dan Bowyer, Mads Jensen of SuperSeed and Lomax Ward of Outsized Ventures go behind the headlines shaping European tech, capital, and power.
This week felt like three worlds colliding:
Nvidia posts another monster quarter and the market shrugs.
Ukraine, four years into war, is quietly becoming Europe’s most important defence manufacturing and innovation engine.
And AI safety is being re-priced in real time by geopolitics, procurement, and competitive pressure.
This isn’t just a news cycle. It’s a systems cycle.
This is Upside where optimism is earned, not assumed.
What’s covered:
03:30 Nvidia earnings: still beating, still not moving
09:40 Ukraine four years on: from aid recipient to capability supplier
17:30 European defence spend: announcements vs real procurement
23:30 Anthropic bends: safety becomes conditional
31:00 Distillation at scale: China, IP theft, and national security
35:00 SaaSpocalypse vs SaaS redemption: systems of record meet systems of action
40:30 OpenAI + Anthropic margins: the hidden constraint under the hype
46:00 Chips & quantum: Europe’s deep tech wedge — if capital shows up
54:30 The “abundant intelligence” thought experiment: disruption and credit risk
🎧 If you’re building in defence, AI, chips, or deep tech, this one’s worth listening with chapter markers.
Nvidia: Blowout Quarter, Shrugged Reaction
Nvidia beat expectations again.
Data center revenue did what it’s done for over a year now: it carried the whole story.
Up 75% year-on-year to $62B, ~91% of total sales.
And the market… barely cared.
That’s the psychological shift.
When you beat numbers 14–15 quarters in a row, the beat becomes baseline. The whisper numbers become the real numbers. And suddenly the question isn’t “how good was this quarter?” but:
How long can this stay abnormal?
The bull case is still clean:
Forward valuation looks reasonable if growth persists
Next-gen chips (Rubin, successors to Blackwell) drive another efficiency step-change
China is effectively “zeroed” in some models — meaning any rules relief is upside
Networking is still quietly compounding
The bear case is also obvious now:
Inference is eating the mix, and inference is where differentiation compresses fastest
Custom silicon is creeping up (TPUs, Trainium, internal chips at Meta/Microsoft)
The core question becomes margins: can 70%+ profitability remain a law of nature?
Nvidia is still the juggernaut.
But the market is starting to look past the shovel seller and ask:
who’s paying for all this digging?
Ukraine, Four Years In: The European Reset That Doesn’t Unwind
February 24 marked four years since Russia’s full-scale invasion.
We’ve said it before, but it’s worth stating plainly:
Europe’s assumptions died in 2022.
Not just about security. About energy. About industry. About tech dependence. About how fast reality can force a rewrite.
This week we saw:
European air defence startups raising real rounds (e.g., ~$30M tickets)
European public institutions inching closer to defence as a legitimate asset class
Ukraine’s defence tech fundraising reaching meaningful scale
Rearmament plans moving from rhetoric into budget lines
But the deeper story isn’t fundraising.
It’s capability.
Ukraine is fighting a 21st-century war defined by:
rapid drone iteration cycles
countermeasures evolving weekly
supply chain improvisation
software-defined battlefields
high-frequency innovation under existential pressure
That produces something Europe hasn’t had in decades:
a live, modern defence innovation ecosystem inside the continent.
And when the war ends, a large portion of that industrial and technical capacity won’t disappear.
It will become exportable.
Is the Defence Wave Real or Still Mostly Announcements?
The spending shift is real.
The secular drivers are real:
Russian threat isn’t going away
the US is explicitly reducing its underwriting posture
autonomous systems + AI are forcing a replatforming of defence procurement
industrial rearmament is becoming politically sellable again
But the bottleneck is predictable:
procurement speed and deployment scale.
Budgets can double and still fail to translate into startup outcomes if:
contracts stay small
sales cycles remain glacial
governments keep buying legacy
venture-backed defence companies can’t become “suppliers of record”
We’re early.
The breakout signal won’t be another NATO memo.
It’ll be: a wave of large, repeatable multi-year contracts for new entrants.
That’s when it becomes a real market.
AI Safety: From Moral Stance to Competitive Luxury
This week, the safety narrative cracked in public.
Anthropic updated its responsible scaling posture in a way that boils down to:
We won’t slow down if we aren’t clearly ahead.
That’s a subtle sentence with an enormous implication:
safety becomes conditional on relative advantage.
At the same time:
the Pentagon pressures model providers to loosen safeguards
Grok clears pathways for classified intelligence workflows
“all lawful purposes” becomes the default standard
defence use is no longer theoretical—it’s procurement paperwork
This is the pattern:
When the environment becomes existential, ethics becomes context-sensitive.
Not because everyone turns evil.
Because incentives harden.
And once you raise at the scale these labs are raising, your room to “take the moral L” narrows sharply.
Distillation at Scale: The IP Theft Question That Turns Geopolitical
Anthropic flagged what it described as industrial-scale distillation behavior — tens of thousands of accounts, millions of interactions, linked to Chinese labs.
You can debate the specifics. But the macro point is obvious:
it’s easier to steal intelligence than to steal goods.
And if the West spends tens of billions building frontier models while competitors harvest the outputs cheaply, that’s not just a corporate problem.
That’s strategic leakage.
At some point, this becomes:
export controls
enforcement measures
diplomatic bargaining chips
and potentially retaliation economics
IP theft is not new.
But in the AI era, it scales faster than the institutions designed to police it.
SaaSpocalypse vs SaaS Redemption
The market mood swings are hilarious because they reveal something real:
Nobody knows where the value accrues.
One week: SaaS is dead.
Next week: Agents will supercharge SaaS.
The cleanest mental model we discussed:
SaaS = system of record
AI = system of action
Meaning: SaaS doesn’t disappear. It becomes the data layer that AI operates on.
Winners likely have:
strong data moats
clean APIs
compliance + governance ownership
pricing that moves away from seats and toward value
But the “AI saves SaaS” thesis collides with one terrifying reality:
the foundation model economics aren’t settled.
The Margins Problem Under the Hype
OpenAI and Anthropic both missed their own margin forecasts last year.
They’re growing insanely fast.
But gross margins in the 30–33% zone are not the destination people assumed.
And inference is expensive:
free users are costly
scale amplifies costs before efficiencies arrive
“growth first” works until it hits credit and public-market math
This matters because it feeds upstream.
If foundation model profitability stays pushed out, the entire infra stack gets re-rated.
That’s why Nvidia can beat and still get punished.
The market isn’t doubting Nvidia’s quarter.
It’s questioning the sustainability of the ecosystem beneath it.
Chips & Quantum: Europe’s Leverage, If It Can Fund It
Quantum in Europe is rising:
stronger funding year-on-year
large public commitments
credible companies across architectures
early government purchase behavior (a rare European strength)
But the same European constraint shows up again:
scale capital.
Europe is excellent at hard science formation.
It is weaker at:
funding the long ramp
keeping companies local as they scale
preventing US listing gravity from becoming default
The question isn’t “can Europe win a niche?”
It’s whether Europe can fix capital allocation plumbing — and move from capability to dominance.
Closing Reflection
This episode had one hidden theme:
the comfort era is ending.
Security is no longer assumed.
AI ethics is no longer stable.
Software multiples are no longer guaranteed.
Industrial policy is back.
And sovereignty is becoming a cap-table question.
Europe has the ingredients.
But ingredients don’t win.
Execution speed does.
