Franklin Templeton's CEO backs short-term AI financing, not long-term data centers

1. Adoption lags the technology — ▶ 3:08

   Productivity is running at 2%, but Johnson argues none of it comes from AI. She attributes it to technologies that have been around for decades, such as digitization and cloud services. Drawing on her time running technology at Franklin Templeton, she says the binding constraints are data quality and change management, not system integration, because employees resist tools that demand effort. Her conclusion is that today's AI will take a while to become meaningful to company earnings. Outside the hyperscalers, she sees broadening markets and strong earnings, with roughly 70% of companies beating projections.

2. Job-loss predictions keep failing — ▶ 5:55

   Johnson cites a 1997 Business Week cover predicting the internet would eliminate the broker. Today, she says, a financial adviser does more than ever, acting as a true financial planner with internet tools. Her pattern: companies first use new technology to make existing work more efficient, then employees start imagining uses no one anticipated. She offers her own example of asking her programmer son to build a golf-ball-finding app for Meta glasses. She still concedes the transition requires care, though she sees massive upside, and she names biosciences and materials science as parallel tailwinds.

3. Duration, not bubble, is the risk — ▶ 8:53

   Johnson describes a recurring curve of exuberance and disillusionment and says parts of AI feel frothy. Her concrete position concerns maturity. Hyperscalers are cash-flowing companies with significant earnings, unlike revenue-free internet-era firms, so she likes financing them short-term. She would avoid long-term financing of data centers, because it assumes energy and chip needs stay tied to today's technology for 20-30 years. She says Ken Griffin is probably right on this. She adds that revenue accounting is unclear and not comparable across companies, which is why she favors active management.

4. Late-stage venture inside mutual funds — ▶ 12:28

   As companies wait longer to go public, Franklin Templeton portfolio managers lost the IPO kickers that once lifted performance. The Silicon Valley growth equity team responded by buying late-stage venture. A mutual fund or ETF can hold up to 15% illiquid assets, but Johnson says funds typically run 6-8% so market moves do not cause a breach. Clients who saw the deals asked for a dedicated fund, and a venture group followed. She credits access to proximity: in Silicon Valley, executives' children go to school with the founders. The firm has held SpaceX and Anthropic.

5. Legg Mason rebalanced the client base — ▶ 18:58

   The Legg Mason deal, which the host put at $4.5 billion, was announced 5 days after Johnson became CEO. It added core fixed income and private markets, including Clarion Partners. Franklin had been 75% retail wealth and 25% institutional, and Legg Mason was the reverse, so the combined firm became roughly 50/50. COVID lockdowns began 3 weeks later and prevented in-person talent assessment. The firm closed 60 days early, helped by its existing habit of running global meetings over video. Johnson says acquirers often wrongly assume their own team is the best and should pick the best athletes instead.

6. Scale acquisitions and their results — ▶ 21:51

   Johnson says clients want fewer managers, so managers need broader capability and scale. Of 17 acquisitions in 6 years, some filled product gaps and some improved distribution. Canvas, a direct-indexing platform, grew from about $2.5 billion to $30 billion in roughly 4 years. Lexington and Clarion have more than doubled. Private credit rose from $26 billion at the BSP purchase to just under $100 billion, though she notes the firm also bought other managers along the way. She argues private credit is not one asset class, so each segment needs its own origination capability, as with its latest asset-backed team.

7. Carry creates built-in retention — ▶ 25:59

   Johnson says private-market compensation, where carry is embedded and paid out over years, gives acquirers natural retention that traditional asset managers lack. Lexington, she concedes, was fully priced but landed in the sweet spot for secondary private equity and outgrew the firm's models. Putnam was paid largely in Franklin stock because the seller, Great-West Life, wanted a diversified manager stake. Putnam had strong performance but was subscale for distribution platforms, and equity assets rose from about $136 billion to over $220 billion once its team reached those platforms.

8. Tokenization cuts servicing costs — ▶ 29:20

   Johnson expects blockchain to replace financial rails, slowly, because it threatens the business models of toll takers. She cites three features: a single immediately updated source of truth that removes reconciliation, smart contracts that enforce collateral terms automatically, and payments. In parallel runs of the SEC-approved on-chain money market fund against the legacy transfer agency system, she says the firm was astonished by the cost advantage. Minimum account size fell from $500 to $20, possibly $5. The fund accrues interest every second, so a hedge fund holding it as collateral for 4 hours 22 minutes 32 seconds is paid for exactly that period.

9. Three buckets of internal AI — ▶ 33:08

   Johnson sorts AI use by economics. In investing, the goal is better performance by freeing analysts from admin; cost savings there are insignificant by comparison. In distribution, a Microsoft-assisted multi-agent system pulls from product systems, adviser holdings feeds, social media, and rating agencies. Early results show a 23% uplift in clients one market leader visits and an 11% increase in sales, with no baseline or period given. In the back office, the aim is throughput rather than headcount cuts: AI writing code lets technology departments clear projects that were always limited by prioritization.

10. Governance may cap AI revenue — ▶ 35:43

    Johnson says two effects have not yet appeared. First, once employees are handed AI tools and build agents, costs start to bite, forcing questions about whether agents must run as often as configured. Second, many queries could likely be served by free models. A governance layer deciding which requests justify a top model would affect hyperscaler revenue, though she says the impact is unknown. The host reported a similar pattern among emerging companies, some of which went from frontier models with 14 trillion tokens to running 99% on open-weight models.

11. Democratizing access to illiquidity — ▶ 39:17

    Johnson traces the firm's origin to a subsidiary her father took over with $2.5 million in AUM and $10,000 in revenue. She recalls the first billion-dollar party around 1983 or 1984 and a $5,000 profit in 1973 or 1974. The founding logic was pooling money so ordinary investors could access equity returns with expert advice and diversification. She sees a similar opportunity in private markets, as companies stay private longer and debt shifts into private credit. She says access must be fair while recognizing these vehicles are illiquid, with excess returns coming from that illiquidity.

Johnson's argument is best read as a theory of time horizons. AI, in her view, is the biggest growth lever in the market, a technology she calls "such a massive disruptor." But she also believes its payoff inside companies will arrive slowly, while the technology itself keeps changing quickly. Those two beliefs sit uneasily together, and the gap between them shapes much of what she says: how she would finance AI infrastructure, how Franklin Templeton deploys AI internally, and why she is patient about tokenization. The common thread is the difference between what a technology can do and how long organizations take to absorb it.

The evidence for slow adoption is experiential rather than statistical. Johnson, who ran technology at Franklin Templeton, says the two hardest problems are the quality of an organization's data and change management, because people do not want to adopt new things that cost them effort. She notes that productivity running at 2% predates AI and reflects decades-old technologies such as digitization and cloud services. If she is right, the earnings effect of today's AI is still ahead. She expects divergence once it arrives, with leaders in affected sectors pulling away from the rest. Her optimism about jobs, illustrated by a 1997 Business Week cover predicting the internet would eliminate brokers, rests on the same timing logic: jobs vanish in forecasts and reappear as new work.

That framework explains her most concrete position on the bubble question. She does not say whether AI is a bubble. She describes a recurring curve of exuberance and disillusionment, linked to a Gartner model whose exact name she could not recall, and says it certainly feels frothy in places. Her answer is about duration. Hyperscalers are cash-generating companies with significant earnings, unlike revenue-free internet-era firms, so she would finance them at shorter maturities. Longer-term financing of data centers she would avoid, because it assumes energy and chip needs stay anchored to today's technology for 20-30 years. She says Ken Griffin is probably right on that point, and she adds that revenue accounting is unclear and not comparable across companies, which is why she favors active management.

A second uncertainty concerns demand rather than technology. Johnson says two things have not yet been seen. Companies first hand employees AI tools and let them build agents, then the bill arrives, prompting questions about whether those agents need to run as often as configured. And many employee queries could probably be served by free models. A governance layer deciding which requests justify a top model would affect hyperscaler revenue, she says, though she admits the impact is unknown. The host's conversations with emerging companies point the same way: some started on frontier models, reached 14 trillion tokens, and now run 99% on open-weight models. The inference, which neither speaker states, is that demand behind long-lived infrastructure may be less certain than headline usage suggests.

Inside Franklin Templeton, Johnson sorts AI into three buckets with different economics. In investing, the aim is better performance, with analysts spending less time on admin and more with companies; cost savings there are insignificant by comparison. In distribution, the aim is growth: a Microsoft-assisted system of coordinating agents draws on product systems, adviser holdings feeds, social media, and rating agencies. Early results, she says, show a 23% uplift in the number of clients one market leader visits and an 11% increase in sales. In the back office, the goal is throughput rather than savings, since AI-written code lets technology departments clear projects they could never previously prioritize. The sales figures come with no stated baseline, sample, or period, so they are suggestive rather than conclusive.

The firm's exposure to SpaceX and Anthropic shows the same pragmatism. Johnson says companies waiting longer to go public meant her portfolio managers stopped receiving IPO kickers, so the Silicon Valley growth equity team began buying late-stage venture. A mutual fund or ETF can hold up to 15% illiquid assets, though she says funds typically run 6-8% so market moves do not cause a breach. Clients who saw the deals asked for a dedicated fund, and a venture group followed. Access, she says, flows from proximity: in Silicon Valley, executives' children go to school with the founders. On SpaceX, she is enthusiastic about Starlink and Musk's record of delivering on bold goals, and names key-man risk as the main concern.

The acquisition record applies the same logic to the firm itself. Johnson's premise is that clients want fewer managers, so survival requires breadth and scale. The Legg Mason deal, announced 5 days after she became CEO and described by the host as $4.5 billion, moved the client mix from 75% retail wealth and 25% institutional to roughly 50/50, and added core fixed income and private markets. Lockdowns arrived 3 weeks later, yet the firm closed 60 days early, helped by already running global meetings over video. She reports Canvas growing from about $2.5 billion to $30 billion in roughly 4 years, and private credit from $26 billion to just under $100 billion. The private credit figure includes other acquisitions made along the way, so it overstates organic growth.

Why do acquired managers stay and grow? Johnson points to compensation structure. Private-market pay, with carry embedded and paid out over years, creates natural retention that traditional asset managers lack. She says she paid full price for Lexington at the time, but it hit the sweet spot for secondary private equity and outgrew the firm's models. The Putnam deal shows a different mechanism. Putnam had strong performance but was too small to reach distribution platforms, and equity assets rose from about $136 billion to over $220 billion once the team gained that access. Editorially, the pattern is that distribution, not investment skill, drove much of the value. That fits her scale thesis, though the transcript offers no organic growth figures to test it.

Tokenization is where her patience is most explicit. She expects blockchain to replace the rails of financial services but slowly, because it forces changes to the business models of toll takers who earn fees on transactions. She cites three features: a single, immediately updated source of truth that removes reconciliation between systems and counterparties, smart contracts that enforce collateral terms automatically, and payments. The evidence is the firm's own parallel run of its SEC-approved on-chain money market fund against its legacy transfer agency system, which she says astonished the team on cost. Minimum account size fell from $500 to $20, possibly $5. The transcript gives no fuller cost data, so the scale of the savings is hard to judge.

The same instinct frames her view of private markets. Her father took over the Franklin subsidiary at $2.5 million in AUM and $10,000 in revenue, the first billion-dollar party came around 1983 or 1984, and the firm eked out a $5,000 profit in 1973 or 1974. The founding idea was pooling money so ordinary investors get expert, diversified access to equity returns. She wants to repeat that in private markets, while acknowledging the vehicles are illiquid and earn excess returns because of it. The unresolved issue is how far retail access can extend when that illiquidity is the source of the return, a question the 15% cap on illiquid holdings only partly answers.