Editor’s Note
This week sharpened the market’s central AI question: not whether demand is real, but whether current prices already assume too much perfection. Across mega-IPOs, AI infrastructure, frontier-tech forecasts, and Microsoft’s platform strategy, investors were forced to separate durable earnings power from liquidity, narrative, and capital intensity. The result was not a clean rejection of the AI trade, but a clear shift toward proof.
Mega-IPOs and the Liquidity Test
The week opened with Seth Klarman’s warning that the coming wave of mega-IPOs could test market liquidity and raise the cost of capital. SpaceX was the focal point, but the bigger concern was what might follow: OpenAI, Anthropic, and other highly valued private companies seeking public-market capital while early investors and employees look for liquidity.
The issue is not simply headline valuation. Even if SpaceX is valued at $2.5 trillion, only $75 billion may initially come to market. Other mega IPOs could begin with $50 billion to $100 billion of float before releasing more shares over roughly 12 months. That staggered supply matters because institutions may have powerful incentives to sell once private holdings become liquid, especially when 10 or 15% of some endowments’ total assets may be tied up in one company.
Klarman’s caution was grounded in valuation discipline. Goldman estimates discussed in the interview suggested that some parts of SpaceX would need growth on the order of 100 X over a long period to justify current prices. His response was blunt: “Those projections have a way of not happening.”
By Monday, the market was already testing that logic. SpaceX fell 16% Monday, extending its three-day decline to nearly 24% after opening at $150 on June 12, above its $135 offering price. At its recent $3 trillion market capitalization, SpaceX traded near 150 times sales, despite generating $19 billion in sales and a $9 billion net loss. Microsoft, by comparison, produced $318 billion in sales and $125 billion in net income; Amazon generated $743 billion and $91 billion, respectively.
SpaceX’s cash position is substantial: $100.8 billion in cash and cash equivalents as of June 19. But losses remain real, including $4.9 billion in 2025 and $4.28 billion in this year’s first quarter. The company’s compute deals add a credible AI infrastructure angle, including Reflection AI access to Nvidia GB300 chips through Colossus for $150 million per month from July 1, 2026, through 2029, a potential $6.3 billion contract. Google separately agreed to pay $920 million per month for capacity involving roughly 110,000 Nvidia GPUs from October 2026 through June 2029.
The bull case is expanding. The valuation risk is, too.
Paper Wealth, Control, and Concentration
Elon Musk’s estimated $1.1 trillion net worth illustrated another version of the same theme: market value is not the same as liquidity.
SpaceX now accounts for roughly 75% of Musk’s fortune. He owns approximately 4.8 billion shares and 350 million options, representing a 42% - 43% economic stake. At SpaceX’s $1.75 trillion IPO valuation, Musk’s position was worth approximately $780 billion, including options. After the company’s first trading day, its estimated value approached $850 billion.
But selling even 5% of his SpaceX stake would place roughly $43 billion of stock on the market. That could depress the share price and signal reduced confidence. Lockups add another constraint: SpaceX’s S-1 reportedly prevents Musk from selling his shares for 366 days after the company’s June 12, 2026 listing, keeping them locked until approximately June 13, 2027.
Tesla supplies most of the remaining wealth. Musk’s shares and options give him an economic interest of around 20%, currently worth approximately $280 billion. His reinstated 2018 compensation package granted options to acquire 304 million Tesla shares, registered in April 2026, and the package alone is now worth $116 billion.
The point for investors is broader than Musk. Concentrated ownership can create spectacular mark-to-market wealth, but liquidity, taxes, leverage, lockups, and control all shape what that wealth actually means.
AI Infrastructure: Growth Is Real, Economics Are Uneven
AI demand remained strong, but the week made clear that not all AI revenue is equal.
Nvidia’s B200 rental price fell from $6.11 per hour on May 30 to $4.22 on June 21. Nvidia shares are up approximately 12% in 2026 but down roughly 3% over the past month, badly trailing the VanEck Semiconductor ETF’s 84% annual gain. Leadership rotated toward memory and infrastructure: Micron and Sandisk each gained nearly 60% over the past month, while Micron has advanced 280 percent year to date.
Cerebras Systems (CBRS) showed the trade-off clearly. Quarterly revenue rose 94% year over year to $193 million, while gross profit more than doubled to $86 million. Operating loss narrowed to $15 million from $28 million, even as R&D expense rose to $75 million from $52 million. Wall Street expects revenue could reach $1 billion in quarterly revenue by year-end 2027, implying a roughly $4 billion annualized run rate.
The problem is margin and cash flow. Core gross margin reached 46.5%, up from 42.1%, but management guided next quarter gross margin to only 36% - 38%, far below Nvidia’s roughly 75% - 76%. Operating cash flow improved from a $54 million loss to a $12 million gain, but Cerebras also spent $131 million on property, plant and equipment against $193 million of revenue.
The company’s strategic pitch remains compelling: “Fast tokens are the most valuable tokens because they get more work done in less time.” OpenAI signed a definitive agreement on December 24 to purchase more than $20 billion of Cerebras compute over several years. Still, a roughly $45 billion valuation demands more proof that growth can become durable free cash flow.
Frontier Tech Needs Evidence, Not Just Ambition
The market also saw a familiar frontier-tech pattern: huge forecasts, sparse methodology.
Nine companies were highlighted as expected to more than triple revenue over five years. The more established names included Nvidia ($NVDA) at 302%, Oracle ($ORCL) at 325%, Broadcom ($AVGO) at 364%, AMD ($AMD) at 402%, and Palantir ($PLTR) at 763%. The more speculative tier was much larger: Iren ($IREN) at 2,556%, Ondas ($ONDS) at 3,024%, AST SpaceMobile ($ASTS) at 4,900%, and Nebius ($NBIS) at 8,217%.
Those are revenue-growth estimates, not earnings, cash flow, returns, or valuation conclusions. Without the source, methodology, starting revenue, analyst assumptions, or dates, the percentages are a diligence prompt rather than an investment case.
Quantum carried the same issue. President Trump signed two major executive orders on quantum technologies, saying they will strengthen America’s position as the “world leader.” IonQ ($IONQ), Rigetti Computing ($RGTI), Infleqtion ($INFQ), D-Wave Quantum ($QBTS), IBM ($IBM), Alphabet ($GOOGL), and Microsoft ($MSFT) were all cited as relevant names. But the article did not provide titles, timelines, funding levels, policy provisions, or company-specific awards.
Ouster ($OUST) offered a more tangible physical-AI thesis. The company is being reframed less as a self-driving car lidar stock and more as a robotics, industrial automation, drones, and smart infrastructure company. Adoption examples included Utah DOT in 100+ intersections, Chattanooga in 120+ intersections, Komatsu autonomous mining, Balyo forklifts, Trombia street sweepers, Microavia avalanche drones, and Argus counter-UAS drones.
But valuation is demanding. At a $3 billion market cap and $196 million revenue, the author estimates a forward price-to-sales ratio of 15.3, above IREN at 6.7, NBIS at 8.3, MU at 7.3, and Nvidia at 12.4 by the comparison used. With the stock up 67% in one month and 157% in three months, entry discipline matters.
Earnings Decide the Next Leg
By Friday, Tom Lee’s framework gave the week its cleanest market takeaway: 2026 has been about “E,” not “P/E.” Consensus 2027 S&P 500 earnings rose from $352 per share at the start of the year to $399 by June 18, a $48 increase. Over the same period, the index rose from 6,846 to 7,365, while the forward 2027 P/E compressed from 19.4 to 18.4. Lee’s punchline: “Today, the market is cheaper today than it was on January 1.”
That matters after a sharp AI pullback. Semiconductors fell 7% in one day and memory stocks dropped 14%, but semis are still up 150% over the past year and memory stocks are up 750%. Since 2011, there have been 17 instances where the semiconductor index fell 6% or more in one day. One month later, the median gain was 12%, with an 88% win ratio at the one-month, three-month, and six-month marks. Six months later, semis were up 38% on median.
Microsoft captured the debate. Commercial remaining performance obligations nearly doubled year over year to $622 billion, but 45% is tied directly to OpenAI. At the same time, Azure and other cloud services grew about 40%, constant-currency growth was 39%, backlog excluding OpenAI grew 28%, and revenue growth accelerated to 18% in Q3 fiscal 2026. Microsoft 365 still has 450 million daily active users.
The concern is the $190 billion CapEx plan, including $25 billion tied to higher component pricing. The opportunity is that much of this spending may be growth CapEx, with about two-thirds going toward servers and networking equipment that could support near-term revenue.
Looking Ahead
Next period, the key watchpoints are AI pricing, IPO supply, and earnings revisions. Monitor SpaceX lockup dynamics and post-IPO trading behavior, Nvidia and AWS GPU rental pricing, Micron and memory-stock follow-through, Cerebras margins and capital intensity, and whether Microsoft can prove that OpenAI dependency is manageable inside a broader multi-model Azure strategy.
The AI trade is not gone. It is becoming more demanding. The companies that can convert infrastructure spending into earnings, cash flow, and defensible returns should separate from those still relying on ambition alone.