📉 Broadcom, 🚀 SpaceX and 🏎️ MiniMax: AI's contradictory signals

Broadcom projects $16 billion in AI chips for its next quarter — and the market is disappointed because it expected $17.2 billion. SpaceX starts its IPO roadshow, the largest IPO in history. And MiniMax launches a code model that rivals Opus 4.7 at 1/40 of the price. Three stories that seem unrelated but tell the same story: AI sends contradictory signals — and all are true from their angle.

📉 Broadcom cools expectations: $16B in AI chips, below expectations

On June 3, Broadcom presented its Q2 fiscal results and delivered a Q3 outlook that disappointed the market. The company projects $16 billion in AI semiconductor revenue for the quarter ending in July, below the $17.2 billion analysts from Bloomberg expected.

CEO Hock Tan didn't raise his $100 billion AI revenue target for 2027, despite Google and Meta expanding their commitments with Broadcom. This was interpreted as a sign that AI chip market expansion might be progressing slower than anticipated.

But the details paint a more nuanced picture. Broadcom plans to ship 10 gigawatts of compute capacity in 2027 and "much more" in 2028. AI semiconductor revenue in Q2 was $10.8 billion, slightly above consensus. And TSMC, the manufacturer of Broadcom's chips, maintains its 30% growth forecast for 2026, with its CEO saying customers "are very positive about AI prospects."

The real message isn't "AI is slowing down" — it's "the supply chain can't keep up." Demand is still there, but TSMC's manufacturing capacity is the bottleneck. And Broadcom, as the middleman, feels the pressure first.

From chip supply to public offering: while Broadcom deals with capacity, SpaceX prepares the largest stock market debut in history.

🚀 SpaceX starts IPO roadshow: the largest IPO in history

On June 4, SpaceX started its IPO roadshow with a target valuation of $1.75 trillion. It's the largest initial public offering in history — and comes in a context where Anthropic has already filed its S-1 and OpenAI is preparing its own.

The roadshow runs until June 12, when shares are expected to start trading under the ticker SPCX. SpaceX plans to sell approximately 10% of the company, which would mean an injection of around $175 billion. Funds will go toward Starship development, Starlink expansion, and the AI computing infrastructure it already rents to Anthropic for $1.25 billion a month.

The context is historic: never before have three tech companies of this caliber — SpaceX ($1.75T), Anthropic (~$965B), and OpenAI (~$1T) — prepared simultaneous public offerings. Wedbush analysts call it "the opening of the IPO market floodgates." If all three go public, we'd be looking at the largest concentration of IPOs in history, with a combined value exceeding $3.7 trillion.

We already covered Anthropic's S-1 and how the triple IPO will define the next decade. With SpaceX joining the dance, the question is no longer when, but at what price.

And while the giants go public, a Chinese startup proves efficiency can break any prediction.

🏎️ MiniMax launches a code model that rivals Opus 4.7 at 1/40 the price

And the story that breaks the narrative: Chinese startup MiniMax launched a code generation model that, according to independent benchmarks, matches the performance of Anthropic's Opus 4.7 on programming tasks — but at 1/40 the cost per token.

The model, whose technical name hasn't been made public, was evaluated on the SWE-bench benchmark (fixing real bugs in open source repositories) and achieved results comparable to Opus 4.7. The cost, however, is drastically lower: approximately $0.25 per million tokens compared to ~$10 for Opus.

Just a week ago we analyzed how open source was losing ground and the gap with closed models was widening. This launch shows the story isn't so simple. MiniMax isn't open source, but its proprietary model competes on price and performance. The gap isn't between open and closed — it's between those who can afford the most expensive compute and those who optimize for efficiency.

MiniMax's move also has geopolitical implications: a Chinese model rivaling the best Western models at a fraction of the cost. In a context where China restricts its researchers' travel, that its startups keep launching competitive models suggests talent and innovation don't stop by decree.

Broadcom says "caution, capacity is the limit." SpaceX says "more, the public market is ready." MiniMax says "cheaper, efficiency is the key." Three contradictory signals from the same ecosystem — and all three are true. AI isn't an industry with one clear direction: it's a field of forces where demand, supply, capital, and innovation pull in opposite directions. Whoever understands that complexity will have an advantage. Whoever looks for a single answer will miss it.

— Max