• Avoid treating Nscale’s proposed IPO at a $30 billion valuation as an actionable opportunity: reported execution delays, power constraints, heavy losses, and high-cost debt create substantial risk, while its contracts depend on facilities being built.
• Approach AI data-center IPOs cautiously: Firmus reportedly canceled its offering, while Nscale and SB Energy faced weak investor demand; verify operating capacity, permits, power access, and financing before investing.
• The insights provide no specific buy, sell, or short recommendation for NVDA, MSFT, SoftBank (9984.T), or BTC; avoid inferring one from concerns about data-center developers.

Nscale (Private; proposed IPO)

• The UK-based AI data-center company reportedly sought an IPO at a $30 billion valuation, but investor demand was described as weak.
• The host argued that Nscale’s record of project delays raises doubts about whether it can deliver its plans:
  • Its announced UK data center was reportedly canceled after Microsoft withdrew, citing insufficient electrical power. The host said construction had not begun and necessary permits had not been obtained.
  • At the Glomfjord, Norway, site, only 6.4 MW of a planned 30 MW was reportedly operating by October 2026, well behind schedule. The transcript also describes cooling-system leaks, a fire, and safety violations.
  • The host questioned the credibility of an image on Nscale’s website, alleging it was AI-generated and made construction progress look more advanced than it was.
• The company reported $140 million in revenue and an operating loss of nearly $500 million in the first half of 2026. As of June 30, 2026, it reportedly had $137 million of debt, including financing at 11.3% and 22.7% interest.
• Nscale claims substantial future contracted demand, including agreements reportedly worth about $44 billion with Microsoft, $44.6 billion with Anthropic, and $3.5 billion with Figure AI. The host emphasized that those revenues depend on Nscale actually building the data centers.
• The company’s proposed Monarch campus in West Virginia is intended to provide more than 1 GW of AI computing capacity. The host questioned whether Nscale could execute a project of that scale given its reported difficulties with a much smaller facility.

Takeaways

• The discussion presents Nscale as a high-execution-risk business, with project delivery, power availability, financing, and high borrowing costs all central to its prospects.
• Treat announced contracts and backlog as conditional rather than equivalent to revenue: the transcript says customers’ payments depend on facilities being built.
• The host’s view is strongly bearish, but the transcript does not provide an independent verification of the allegations or a specific recommendation to buy or short the company.

Firmus (Private)

• Firmus, an Australian AI data-center company, reportedly canceled its IPO after failing to attract enough investor demand.
• The host described Firmus as one of several NVIDIA-backed companies trying to benefit from the AI investment boom.

Takeaways

• The failed offering is presented as a possible sign of cooling investor appetite for AI data-center IPOs. It is not, by itself, evidence that demand for AI infrastructure or Firmus’s business has disappeared.

SB Energy / SoftBank Group (9984.T)

• SB Energy, described in the transcript as a subsidiary of Japanese technology conglomerate SoftBank, was reportedly pursuing an IPO.
• Media reports cited by the host said investor skepticism and weak demand put the offering at risk, though it had not been formally canceled.

Takeaways

• The transcript offers a cautious-to-bearish signal about investor interest in AI infrastructure listings, but provides no specific financial details about SB Energy or a direct investment recommendation for SoftBank.

NVIDIA (NVDA)

• The host said Firmus, SB Energy, and Nscale are all backed by NVIDIA and are seeking to capitalize on the AI investment boom.
• The transcript does not discuss NVIDIA’s own financial results, valuation, or outlook.

Takeaways

• The discussion’s concern is about the ability of AI data-center developers to execute and secure financing—not a direct analysis of NVIDIA’s stock. It does not establish a specific bullish or bearish view on NVDA.

Microsoft (MSFT)

• Microsoft was reportedly expected to rent capacity from Nscale under long-term agreements described as worth about $44 billion through 2033.
• The host said Microsoft withdrew from the proposed UK data-center project, citing insufficient electrical power. The transcript characterizes Nscale’s broader customer agreements as dependent on the facilities being built.

Takeaways

• The discussion highlights a key distinction for infrastructure contracts: a large announced commitment may not translate into usable capacity or revenue if projects face power or construction problems.
• No specific view or price target for MSFT was provided.

Anthropic (Private)

• Anthropic reportedly signed agreements with Nscale described as worth about $44.6 billion over multiple years, contingent on Nscale building the relevant data centers.
• The host cited a Reuters report about Anthropic’s reported $518 billion in AI spending commitments, saying 80% were non-cancelable or required payment regardless of usage. The transcript notes that this information came from a purported review of Anthropic’s not-yet-public S-1.
• Anthropic’s stated rationale for large compute commitments is that computing capacity constrains AI development. The host disputed this, pointing to free credits and generous consumer-plan usage limits as evidence that Anthropic may have unused capacity.
• The host also argued that failed data-center projects could reduce Anthropic’s actual spending obligations if payments depend on facilities being completed.

Takeaways

• The transcript raises a financing and utilization question: large compute commitments could be burdensome if demand or revenue does not support them, while project delays could prevent some planned spending from taking effect.
• The claims about commitments, capacity constraints, and customer demand are presented through the host’s interpretation and reporting he cites; the transcript does not independently establish how Anthropic’s contracts will ultimately be treated.

Figure AI (Private)

• Nscale reportedly agreed to potentially deploy up to 100,000 NVIDIA GPUs for Figure AI beginning in the second half of 2027, under an agreement the host valued at about $3.5 billion.
• The host described Figure AI as pre-revenue or having minimal revenue and expressed the opinion that its demonstrations were misleading and its robots could not perform useful tasks. Those are the host’s allegations and assessment.

Takeaways

• The discussion questions whether a pre-revenue company could support a large future infrastructure commitment. The key issue raised is the customer’s ability to pay, not a stated investment recommendation on Figure AI.

OpenAI (Private)

• OpenAI was mentioned alongside Anthropic as an AI company that says it is constrained by computing capacity while also making large spending commitments.
• The host questioned whether the “compute constrained” explanation fully accounts for the companies’ activity, but did not provide specific OpenAI contract or financial figures.

Takeaways

• The transcript suggests scrutiny of whether AI companies’ expected demand and revenue can justify their infrastructure commitments. It offers no direct valuation or investment recommendation for OpenAI.

Bitcoin (BTC) and Bitcoin Mining

• Nscale reportedly began as a Bitcoin miner before pivoting toward AI data centers in 2023 or 2024.
• The host said the Glomfjord, Norway, facility had previously been a Bitcoin-mining operation and was being adapted for AI computing.

Takeaways

• Bitcoin is mentioned as part of Nscale’s history and the reuse of mining infrastructure, not as a direct investment thesis. The discussion does not express a view on BTC or its price.

AI Data Centers and Neocloud IPOs

• The episode’s central theme is that several NVIDIA-backed AI data-center IPOs may be struggling:
  • Firmus reportedly canceled its offering.
  • SB Energy and Nscale were described as facing weak investor demand, though neither was said to have officially canceled at the time discussed.
• The host interpreted these developments as possible signs that the AI investment boom is “starting to show cracks.” He also highlighted power access, construction delays, financing needs, and high interest costs as risks for data-center developers.
• The transcript describes Nscale’s UK announcement of a $2.5 billion infrastructure investment and a planned sovereign AI data center, while arguing that announced projects may not translate into completed facilities.

Takeaways

• The discussion supports a cautious view of early-stage AI infrastructure projects, especially where plans depend on major borrowing, future construction, and power availability.
• For evaluating similar companies, the transcript points to practical checks: distinguish operating capacity from planned capacity, verify permits and construction progress, and assess whether announced customer commitments depend on successful project completion.
• The host’s broader AI-bubble warning is an opinion, not a confirmed market conclusion. The transcript gives no specific market-wide price target or timeline.

Short-Selling Strategy Mentioned

• The host promoted a research service focused on companies it believes are fraudulent or overvalued and described a model portfolio of short positions.
• The host said the portfolio’s share-price performance had underperformed the S&P 500, which it presented as consistent with the portfolio’s stated goal. The transcript explicitly said this was not the hypothetical return from shorting the stocks.
• No individual short positions or specific short-sale recommendations were named.

Takeaways

• The discussion does not identify a particular stock to short. The portfolio performance claim is not the same as a realized short-selling return, so the distinction matters when assessing the service’s results.