AI Markets: SpaceX targets mobile carriers, OpenAI and Firmus

SpaceX buys low-band spectrum and sends US carriers tumbling. Two competing figures clash over OpenAI’s revenue, and Firmus walks away from its IPO.

SpaceX buys low-band spectrum, and US carriers slide

On Thursday, October 8, SpaceX announced it is buying all of Grain Management’s 800 MHz spectrum. Shares of AT&T, Verizon and T-Mobile fell in after-hours trading. It’s the latest episode in a saga Five Minute AI has already been following: the EchoStar deal and the FCC’s approval of 15,000 satellites.

The key points:

  • SpaceX is taking over 100% of Grain’s 800 MHz portfolio, or up to 14 MHz of paired low-band spectrum, according to Investing.com.
  • Neither party disclosed the price. A figure of roughly $8 billion is making the rounds in the press, with no official confirmation.
  • The deal still needs final approval from the FCC, the US regulator.
  • The three big carriers lost between 5% and nearly 7% after hours, according to Investing.com. Reuters reported smaller declines. The differences come down to timing and source.

Why this spectrum matters

Low-band frequencies go through walls. That’s exactly what SpaceX was missing: its direct-to-phone satellite service mostly works outdoors. According to SpaceX, most of today’s phones already support the 800 MHz band without any modification.

Grain bought this block from T-Mobile in August, in a swap of cash and 600 MHz spectrum. So the block is changing hands for the second time in two months.

The market angle

What’s changed since our last edition: SpaceX is no longer just going after dead zones, but after the heart of the carriers’ market. Starlink Mobile wants to combine satellites with a terrestrial network.

For a European investor, there are three things to watch. SpaceX stock (SPCX) trades on Wall Street, so there’s no direct access through a PEA. US carriers carry significant weight in many “US dividend” ETFs that are eligible for a standard brokerage account. The FCC decision remains the real unknown: no timeline has been announced.

A word of caution: announcing a purchase isn’t the same as launching a service. The Starlink Mobile timeline, already targeted for late 2027 in previous editions, hasn’t changed at this point.

Sources:

OpenAI: $70 billion or $50 billion in revenue? The number rattling chip stocks

Two figures are clashing over OpenAI’s revenue. Bloomberg reports $70 billion expected by the end of 2026. The Financial Times puts it at just under $50 billion today. Chip stocks got shaken up.

The key points:

  • According to Bloomberg, OpenAI expects to reach or exceed $70 billion in annualized revenue by the end of 2026, driven mainly by enterprise customers.
  • According to the Financial Times, OpenAI told its investors “just under $50 billion” at the end of September, about $20 billion below the figure that had been circulating.
  • The FT attributes the gap to accounting: Anthropic books the full value of sales made through its cloud partners, while OpenAI only counts its own share.
  • Chip stocks fell 3.4% on Thursday, according to Investing.com.

Two numbers, two meanings

$70 billion is a target for the end of 2026. $50 billion is a measurement as of the end of September. So these aren’t two measurements of the same moment.

The FT also believes the $70 billion figure was an investor extrapolation, partly based on press reports. Bloomberg later reported the $70 billion target, citing people familiar with the matter.

What’s changed since our last edition: the debate is no longer just about how big OpenAI is, but about how to measure it. Compare two labs using two different accounting rules and you get two different pictures.

The market angle

OpenAI buys enormous amounts of chips and computing capacity. When its revenue looks weaker, its suppliers take the hit. According to CNBC, the Nasdaq 100 lost more than 300 points in half an hour on Thursday. Nvidia, Oracle and CoreWeave were hit especially hard.

The long-term contracts are still signed. But the market is wondering whether revenue will keep pace with spending. OpenAI is reportedly in talks to raise money at a $1.4 trillion valuation. Its IPO isn’t expected before early 2027, according to media reports. Anthropic could go first.

For Europeans, exposure comes mainly through Nasdaq or semiconductor ETFs. There’s nothing to say Thursday’s move will continue.

Sources:

Firmus drops its Australian IPO: the market’s first rejection of an AI data center builder

Nvidia-backed Firmus abandoned its Sydney IPO on Friday, October 9. It would have been the second-largest in Australian history. The company is turning to private funding instead.

The key points:

  • The offering was aiming to raise about $5 billion, at 11 Australian dollars per share, according to Reuters via Investing.com.
  • That valued Firmus at $30.6 billion, nearly triple its August valuation ($10.5 billion).
  • Firmus says it intends to raise private capital. One person involved mentioned a goal of listing on the Nasdaq after the next round, which the company has not confirmed.
  • Asset manager Maas Group, a shareholder, had already lost more than 20% on Thursday, according to MarketScreener.

Why demand fell short

According to Reuters, investors were worried about the debt (around $30 billion), Firmus’s limited experience with AI data centers and the exit of a partner. The CEO of CDC Data Centres said a 1.6-gigawatt project with Firmus was no longer moving forward. Orders were then pulled.

Another sore point: the lock-up rules would reportedly have let existing shareholders sell more than half of the company on day one.

What’s changed since yesterday’s edition: we were tracking a final price and a listing scheduled for October 23. That option is now off the table.

The market angle

The signal matters more than the company. An AI infrastructure player with big names on its cap table (Nvidia, Blackstone, Jane Street, Coatue) couldn’t find buyers at that price. Markets are now weighing the value of these projects against their debt.

For a European retail investor, Firmus wouldn’t have been available through a PEA anyway. The lesson lies elsewhere: IPOs of heavily indebted AI companies are being judged case by case.

This is a market event, not a forecast: one isolated rejection says nothing about every AI stock.

Sources:

Lumentum: AI’s optical components are sold out through 2029

Lumentum’s CEO says its optical components are “completely sold out” almost through 2029. Demand from AI data centers is outstripping what the factories can produce.

The key points:

  • Michael Hurlston said so in an interview in Tokyo on October 9, according to Bloomberg.
  • Earlier this year, he had pointed to 2028. So the horizon has moved out by about a year.
  • According to the Japan Times, Lumentum won’t be able to meet about 70% of demand for some products next year.
  • On March 2, Nvidia announced a $2 billion investment in Lumentum, along with a multibillion-dollar purchase commitment.

What these parts are

These aren’t compute chips. They’re indium phosphide lasers and components that move data between servers at high speed. An AI data center needs thousands of them. Without them, the processors sit waiting for their data.

What’s changed since our editions on the chip supply chain: the bottleneck is no longer just GPUs and memory, but optical connectivity too.

The market angle

“Sold out” capacity gives visibility into future sales. It doesn’t guarantee margins, profits or the stock price. According to the financial press, other optics stocks (Coherent, Applied Optoelectronics, POET) also rose overnight.

This is a statement from the CEO, not published results. The accounting figures will come with the next quarterly reports.

For European investors, these stocks are listed in the United States. Access is through a brokerage account or UCITS-style semiconductor/AI thematic ETFs. Those funds often hold better-known names, so exposure to optics remains limited.

The opposite risk exists: if AI orders slow down, as Thursday’s OpenAI session suggests could happen, a full order book today is only as good as the customers behind it.

Sources:

Edition prepared by our monitoring agent (data and sources as of Friday, October 9, 2026), reviewed in line with our editorial policy. General information, not investment advice: investing carries a risk of capital loss.