Anthropic IPO, Buying Bonds, the AI Bubble | Finimize Pro Round Up | October 2026

Finimize Pro Monthly Roundup: Key Takeaways

Performance Update

Since the research ideas launched nearly two years ago, the team reported:

• 28% average annualized absolute return
• 10 percentage points of average annualized return above the benchmark
• 65% hit rate (ideas that made money)

SharkNinja: Product-Led Growth and Expansion

SharkNinja designs and sells household appliances across roughly 40 product categories under its Shark and Ninja brands. The presenter described it as a roughly $25 billion company and the fifth-largest home and kitchen appliance maker.

Its playbook is to enter categories that appear mature, identify shortcomings in existing products, and use product innovation to improve them.

The company had achieved 13 consecutive quarters of double-digit growth. Market estimates anticipated growth slowing from about 19% this year to 13% next year and 11% the following year; Charlie Pullan viewed those expectations as too pessimistic.

Potential growth avenues include:

• Winning more share in existing categories
• Launching products in new categories
• Expanding internationally, particularly beyond its largely US-based business

Recent examples included an innovation-award feature, a partnership with McLaren Racing, and a limited-edition beauty collection with Sephora.

Pullan’s scenario analysis indicated about 20% upside from when he initiated the idea. The stock was up 10% after about three weeks; he said the next major test would be quarterly results in early November.

Bonds: Income and Portfolio Resilience

The presenter’s core point was that a bond’s starting yield is a useful guide to its long-term return if the issuer pays as promised and coupons can be reinvested at similar rates. For example, a 10-year bond bought at a 5% yield would produce an annualized return of roughly 5% under those assumptions.

Higher starting yields, he argued, provide more income to cushion price declines if yields rise. As an illustration, he cited 10-year US Treasuries yielding about 5.3%, estimating that yields could rise to roughly 6% over the next year before price losses erased a year’s income.

He also explained bond convexity: a given fall in yields may produce a larger price gain than the price loss from an equal-sized rise in yields.

His recommendation was to consider bonds as a portfolio addition, while using a diversified approach that does not leave investors excessively exposed to further yield increases.

Anthropic: Business Model, Capital Needs, and Valuation

Russell Burns described Anthropic, maker of Claude, as a leading large-language-model provider alongside OpenAI. The business uses chips, electricity, and data-center capacity to develop and serve its models.

Its commercial offerings include:

• Usage-based contracts, where customers pay according to usage
• Fixed-spend contracts, with customers committing to a monthly or annual amount
• Seat-based subscriptions, including a basic consumer offering cited at around $20 per month

Burns said token use had increased as token prices fell. Cheaper models—particularly Chinese open-weight models—may serve basic needs, while leading “frontier” models can command premium prices for more demanding tasks.

The main operating challenge is the capital required to train models and secure compute and energy. Burns described a form of circular financing: investors such as Amazon fund Anthropic, while Anthropic commits to buying services and chips from those providers. He cited Anthropic’s commitment to spend $100 billion over several years on Amazon-related services and chips.

The reported IPO discussion involved raising about $100 billion at a proposed $2 trillion valuation. Burns emphasized that the valuation was difficult to assess without the public filing and more detail on costs, margins, and cloud-provider economics.

A reverse discounted-cash-flow analysis was used to work backward from the proposed valuation to the performance it would require. Burns cited reported targets of $198 billion in revenue by 2028 and roughly $1.99 trillion by 2036, with a 25% operating margin, to support a $2 trillion valuation. That implies revenue growth of just over 30% annually from 2028 through 2036.

The discussion also described the required revenue as around “78%” of a reported $30 trillion total addressable market. That percentage appears inconsistent with the figures given in the subtitles.

Key questions for the IPO include:

• Can Anthropic sustain a product edge as competition intensifies?
• Will falling token prices pressure margins?
• How much does Anthropic pay cloud partners to distribute its models?

AI Infrastructure and Bubble-Risk Indicators

Burns also discussed Nvidia and Microsoft as potential beneficiaries of continued AI adoption:

• Nvidia: Positioned as a core supplier of AI infrastructure. Burns argued that more efficient hardware could lower the cost of generating tokens and support continued demand.
• Microsoft: Positioned to benefit from enterprise AI adoption through Copilot and its relationship with OpenAI. Burns cited Microsoft’s 27% stake in OpenAI and said its in-house models could reduce reliance on external providers for lower-end needs.

For monitoring whether AI investment is generating sustainable returns, the presenters grouped risks into three areas.

Monetization

Track token prices alongside total token consumption and spending. A price decline alone is not necessarily a warning if usage grows enough to offset it.

The speaker cited a Silicon Data token-price index moving from about $2 in late May to around 95 at the time of the discussion, a decline of more than 50%. The subtitles do not make the units fully clear.

Financing Costs

Monitor hyperscalers’ bond yields and, especially, credit spreads over government bonds. A sharp widening could signal rising financing costs and declining confidence that AI investments will earn adequate returns.

Physical Constraints

Track energy, water, grid transmission, data-center permitting, and local opposition.

The speaker cited 45 proposed US data-center projects worth $68 billion as blocked or delayed amid local opposition in the second quarter. Also watch for a slowdown in AI capital spending, which could reflect weak returns or limits on infrastructure buildout.

The speaker’s high-level hedge for a severe AI downturn was government bonds, on the view that weaker growth, lower inflation, rate cuts, and risk aversion could support them.

Q&A: Decision Frameworks and Portfolio Practices

Low-Beta Anomaly and Age-Based Risk-Taking

Theodora distinguished between choosing lower-volatility stocks and deciding how much of a portfolio to allocate to stocks overall. Lower-volatility shares may offer better risk-adjusted returns historically, but not necessarily higher total returns. She cautioned that they can become expensive, lag in strong bull markets, and may not repeat past performance.

Selling Discipline

The host summarized a framework from Lee Freeman-Shor’s The Art of Execution:

• Rabbits hold losing positions without reassessing the thesis.
• Assassins cut losses when the investment thesis breaks down. The example discussed was selling after a fall of roughly 20–30%.
• Hunters add to a losing position if the original thesis remains intact, while limiting the total portfolio loss—illustratively to 2–4%—and avoiding averaging down on highly risky businesses.
• For winners, raiders sell too quickly, while connoisseurs hold and may trim 10–20% after gains of roughly 30–40%.

Possible exit triggers include a broken thesis, an extreme valuation, or a pre-set fall from a peak. The host cited 30–35% as one possible trailing-stop range.

Core-and-Satellite Portfolio

Theodora suggested first establishing emergency cash and using available tax-advantaged accounts. A possible allocation was:

• 70–80% core in one or two diversified global tracker funds, with regular automatic contributions
• 20–30% satellite for higher-risk individual ideas or learning

She advised checking the top holdings and sector exposures across all funds, since apparently different investments may share exposure to large technology companies or AI spending.

Presenters and Sources

Presenters: Reda Farran (the host; rendered as “Rea” in the subtitles), Charlie Pullan, Theodora, and Russell Burns.

Sources and frameworks mentioned: Finimize research on SharkNinja, bonds, and Anthropic; Lee Freeman-Shor’s The Art of Execution; and the Silicon Data LLM token-expenditures index.