How much revenue is needed to justify a $1 trillion valuation? How much extinction-level risk is too much? And what happens if the frontier AI labs decide to pace themselves while open models nip at their heels?

Such existential quandaries bedevil the AI industry’s many stakeholders in the lead up to Anthropic and OpenAI’s expected mega IPOs, especially in light of a truth nuke last Thursday that the latter startup’s revenue was growing slower than previously thought. That’s the bad news backdrop for ASML and TSMC as they prepare for their next earnings reports on Wednesday and Thursday this week, respectively. The good news? It’s boom times for the hardware kings.

What a Difference $20 Billion Makes

Last week, the Financial Times reported that OpenAI had begun telling investors that its annualized revenue came in at around $50 billion in September (the company later confirmed the figure to The New York Times). That’d be all well and good, had it not been previously reported that the company was telling investors that its annualized revenue was closer to $70 billion. The $20 billion chasm from a preeminent software seller sent shockwaves across the entire AI trade. Shares of Oracle dropped more than 6% on Thursday. Nvidia-backed cloud computing startup Firmus placed its IPO plans on hold. The Philadelphia Semiconductor Index, for instance, is now down 7% since last Wednesday’s close; TSMC and ASML are down 5.5% and 3.5% since then, respectively.

On the other hand, there’s no letup in demand for compute and the chips behind it — and the machines that make such chips, and the machines that make those machines. And both companies have already shared tidbits that inspire continued confidence:

• TSMC gave an all-important teaser for its full earnings report, announcing on Thursday that its third-quarter revenue came in at a gargantuan $46.7 billion. That’s another record for the chipmaker, as well as a 50% year-over-year increase and above most consensus expectations for the period; Thursday’s call will now reveal if the company can notch its sixth straight quarter of record profits to go along with its revenue beat.
• ASML, meanwhile, is reportedly planning a 10% price hike for some component parts of its lithography machines at the start of next year; in July, CFO Roger Dassen told analysts the company has room to flex its pricing power.

Token Tumble: In another sign of a growing divide between AI hardware’s relative resilience and AI software’s relative softness, the cost of LLM tokens has roughly halved since May, while the cost of compute rental prices continues to climb, according to market intelligence firm Silicon Data. Translation: open-weight AI is undercutting the model market, while GPU demand is still outstripping supply.

Shares of major telecom companies fell off a lunar cliff last week after SpaceX announced a plan that could expand Starlink beyond satellite-connected internet into cellular coverage. AT&T and Verizon both had their worst day since the early 2000s on Friday, while T-Mobile hadn’t fallen so low since 2013.

But investors may have jerked their knees too quickly; Citi analysts chimed in Friday to basically say to buy the dip, predicting that SpaceX won’t pose a threat to the Big 3 telcos’ operating profits until at least 2029.

Keeping the Lines Open

SpaceX plans to buy a portfolio of low-band spectrum licenses from investing firm Grain Management. The move could put its Starlink service, which already connects 12 million customers in more than 160 countries to the internet, in direct competition with major telcos. Starlink already has high- and mid-band spectrum, which work best for fast data transmission — SpaceX bought $17 billion worth of mid-band spectrum from EchoStar last year. Low-band spectrum isn’t as fast but is better at penetrating walls, helping connect devices inside buildings and signaling a move from broadband to cellular service.

While that sounds like a layup for Starlink, the company won’t turn the Big 3 into a Big 4 overnight:

• Traditional telcos have an infrastructure advantage on the ground. If SpaceX wants to expand beyond its satellites, it’ll need to acquire infrastructure of its own, whether that means building it or partnering with companies that already have it.
• The process is expected to take years, time that telcos could use to figure out how to keep customers. Citi analysts expect AT&T and other companies to unbundle plans and create more affordable tiers. Telcos could also hang onto customers for longer as shoppers wait longer to buy new phones.

Today’s Winners: Smaller picks-and-shovels telco companies are seeing shares surge. Cell tower operators, which could soon gain a huge new account in SpaceX, gained momentum last week — shares of Crown Castle, American Tower and SBA Communications all climbed. FCC Chair Brendan Carr told CNBC there’s another beneficiary from SpaceX’s ambitions: American consumers. More competition in the spectrum space could mean more options, wider coverage and cheaper prices.

“Love it, hate it, fear it. It is already here.” That’s not the tagline of a breakout low-budget horror film creeping into a theatre near you.

It’s how International Monetary Fund Managing Director Kristalina Georgieva describes the economic impact of artificial intelligence, which she likened to a “slow-moving tsunami” at the organization’s annual meeting on Monday. She suggested the amount of movement toward capital and away from labor that the IMF forecasts could prompt a rethink of national taxation models.

Downgraded by Growth

Earlier this morning, the IMF hiked its forecast for global growth this year to 3.1%, up from 3% in July but well below the 3.5% recorded in 2025. In the months since its last update, a sustained bond sell-off made borrowing the most expensive in decades for many governments, including the US. Georgieva, a Bulgarian economist, warned officials in advance of this week’s meeting that they should rein in spending because the “underwhelming” pace of growth this decade may be enough to relieve the increasing debt burden. Public debt has almost reached 100% of global GDP, she noted in an address last week, calling advanced economies “the worst offenders.” She also noted central banks can only do so much, as hiking rates to stem inflation is “feeding directly into the cost of short-term debt.”

“We don’t see decisive action in the high-debt advanced economies where the need of the hour is for credible medium-term fiscal consolidation plans, supported in some cases by upfront fiscal measures, including to take some of the pressure off monetary policy,” Georgieva said.

While on the subject of growth on Monday, she flagged the positive and negative disruption potential of AI. On the one hand, the IMF forecasts the demand and investment fueled by global AI spending — which Gartner forecasts will increase 49% this year to $2.7 trillion — could significantly boost growth. For example, it could add as much as 0.5% to 1% growth in advanced Asian economies and 0.2% to 0.8% in emerging Asian economies. On the other, what it does to the job market could prompt a rethink of the entire taxation model governments are built on:

• Georgieva said the IMF estimates 40% of jobs will be “augmented, downgraded or eliminated” by AI, with that level hitting a dramatic 60% in advanced economies. Evidence is already emerging: an IMF study published earlier this year found that one in 10 US jobs require new AI or tech-related skills.
• A pivot from labor to capital on this scale would require massive changes in how governments generate revenue through taxation. It’s also aimed straight at young workers: In the US, Georgieva said demand for graduates has gone down for the first time in a century while entry-level career jobs are shrinking and low-paying jobs are increasing.

Jobs as Policy: Singapore President Tharman Shanmugaratnam, who spoke on the same panel as Georgieva, said he thinks governments need to develop incentives on par with the COVID and Great Recession eras to encourage the private sector to hire graduates and early-career professionals. These, he said, should be part of “the menu of structural policies” rather than thought of “as a crisis measure.”