Upside is a weekly podcast designed to look behind the headlines that will affect European venture, startups and investing.

Below are the notes from this week’s episode. Episode links above to tune in and stream wherever you pod.

Musk v Altman, and the export-control mirror

Musk v Altman went to court. My favourite line came from OpenAI’s lawyer William Savitt: “We are here because Mr Musk didn’t get his way at OpenAI. So he quit, saying that they would fail for sure, but my clients had the nerve to go on and succeed without him.” Bait-y. Designed to be.

Musk also said under oath that Tesla is not currently pursuing AGI. He said the opposite on X a few weeks ago. Mads’ take: you can have a difference on horizon. My take: also a Trumpian habit of saying whatever calls a reaction in the moment.

The substantive story is China blocking the sale of Manus to Meta. Manus is the agentic front-end that turned heads last year, originally Chinese, redomiciled to Singapore for optionality. Benchmark led a US round, Meta lined up an acquisition, China killed it, citing IP transfer without approval and slapping exit bans on the founders. The IP cannot leave. The founders cannot leave. It is hard to sell an AI company under those conditions.

This is the export-control mirror image. The Americans say you cannot buy our chips. The Chinese say you cannot buy our engineers. Anyone underwriting a cross-border AI deal needs to bake retroactive Chinese IP and personnel controls into diligence, not just US chip export rules.

The EU AI Act did not get watered down

12 hours of trilogue this week. No resolution. The proximate fight was whether AI inside a medical device sits under the AI Act or the Medical Device Regulation. Without a fix, the original 2 August deadline holds.

The omnibus rewrite was meant to soften the parts of the Act that stop European AI champions from competing, and to touch GDPR where it bites. Industry, governments and large parts of the Council all wanted some version of that softening. Parliament killed it. Several MEPs framed the rewrite as big-tech overreach and deregulation that would hurt normal people.

So the watering-down narrative is wrong. The Act is not being quietly buried. It is being defended.

There may still be a deal in May or over the summer. There has to be. If we kill our industry, there is no tax base. If there is no tax base, there is no good stuff. None of it.

Same day, in London, Liz Kendall announced a £500m sovereign AI fund, a national hardware plan, and a middle-powers alliance with France, Germany and Canada. Her line on regulation: pragmatic, not dogmatic. We will see how that holds up under contact. If the EU stalemate persists, the biggest beneficiary is the UK. Already the most attractive place in Europe to build an AI lab. If trilogue stays broken, it becomes the only one.

Could SMRs solve the UK’s energy crisis?

Quick framing. UK domestic energy is second only to Germany on price. Commercial is top quartile in Europe. Industrial is the killer: roughly 63% above the IEA median and around 125% above the EU-14 median. Heavy industry pays the bill, and it shows in where projects get built.

Small modular reactors are the right kind of bet. Self-contained, transportable, faster to permission than a full Hinkley. Drop them on a brownfield site near a substation and you skip a decade of new-build planning theatre. France is doing real work on modular designs. The UK should be in the conversation.

But the honest answer: the first SMR-powered electrons in a UK data centre are early-to-mid 2030s. That is a decade of nothing. SMRs do not solve this crisis. They eventually solve the next one.

The all-of-the-above answer is correct. Gas peakers now. Life extension on Hartlepool and Heysham, even with the engineering case for retiring them. Clear the queue of zombie projects clogging the grid. Apprenticeships for electricians and pipefitters, because the FT had a great piece on US data centre slippage and the constraints are mostly people, not silicon. We need more economic activity. We need the trades to deliver it.

Quick hits from the rest of the week

Earnings: Five big beasts reported. Microsoft cloud up 18% year on year, Alphabet up 22% with Google Cloud specifically up 63%. Meta got punished for raising CapEx without a cloud business and dropped 8% on the day. Apple printed iPhone sales up 22% YoY, Greater China up 28%. Universal CFO message: capacity-constrained, would grow faster with more GPUs.

Memory: The constraint has shifted from power to memory. Three suppliers worldwide: SK Hynix, Samsung, Micron. New fabs take years. Even if Chinese labs make their own accelerators, the memory queue runs years long.

Europe is shipping: Ineffable raised $1.1bn at a $5bn+ valuation in London. David Silver, the architect of AlphaGo and AlphaZero, leading. The pitch is bold: scrap LLM-on-internet training, do reinforcement learning from scratch. Recursive Superintelligence raised $500m on AI that improves its own architecture once deployed.

Models: Grok 4.3 (beta, $300/month). Kimi K 2.6. GPT 5.5 (API token cost double the previous model). Opus 4.7 is more token-hungry by design and consumption thresholds came down. A price hike in everything but name. Anthropic on track for $100bn revenue this year. Mistral shipped a new model and Vibe agents, cheaper than Sonnet, real candidate for sovereign AI demand inside Europe.

SpaceX: S-1 filed confidentially. IPO mid-May, targeting $1.75-2T at a ~$75bn raise. The largest IPO in history. On 2025 revenue of $15-16bn, that is 110x trailing revenue and 220x earnings. The bull case is space compute. People who bet against Musk on Tesla and Falcon 9 lost money.

Predictions

Mine: France becomes Europe’s OPEC. Net exports hit 92 TWh last year, the highest ever. About 8% of UK consumption is French electrons. They have been pricing surplus at marginal cost. Watch that flip to strategic value as transport infrastructure across the Pyrenees and the Channel matures. (Mads pushed back: electricity is hard to store, you cannot tank it like oil. Fair. The OPEC analogy stretches. The political leverage does not.)

Mads: Anthropic should IPO and probably will not. $50bn round in motion at OpenAI-like multiples. Family offices everywhere are putting capital in. Pressure to stay private gets stronger every round. A board decision is expected in May.

Deals of the week

Mads: Ineffable. $1.1bn, $5bn+ valuation, London. Sequoia, Lightspeed, Index, EQT. The proper who’s-who.

Mine: A-Tech. £680k from Loveball. Making hardware prototyping more accessible. A few orders of magnitude smaller than the round above. The market needs both ends of the barbell.

Notable Quotes

“We are here because Mr Musk didn’t get his way at OpenAI. So he quit, saying that they would fail for sure, but my clients had the nerve to go on and succeed without him.” – William Savitt, OpenAI counsel.

“It is hard to sell an AI company when you block the transfer of the IP and you block the movement of the people.” – Mads Jensen, on China’s response to the Manus deal.

“If we kill our industry, there’s no tax revenue. If there’s no tax revenue, we can’t afford any of the good stuff.” – Mads Jensen, on the EU AI Act omnibus failing.

“Even the magic of AI can’t wish away all these other things.” – Mads Jensen, on capital, commodities, energy and labour markets still mattering.