Steve Cohen made his fortune by being right when other people were comfortable being wrong. The Point72 founder runs one of the most data-intensive multi-manager platforms in the world.
So when he makes an 188% increase in one position and a 55% cut in another simultaneously, to me, that is not portfolio rebalancing — but a thesis change.
The Q2 2026 13F filing tells us the story. Cohen sold 1.24 million Broadcom shares, cutting his position by 55%, while purchasing 2.23 million additional Oracle shares, lifting his stake by 188% to 3.41 million shares worth approximately $499.82 million.
Point72’s portfolio sits at $88.13 billion across 3,859 holdings with a 27% turnover rate, according to GuruFocus.
This is Cohen’s answer to a question every AI investor is asking right now: Which company makes more money from the next phase of the build-out?
Why Cohen is trimming Broadcom: The ‘fell short’ problem
Let me be clear about what Broadcom is. It is one of the best-positioned semiconductor companies in the world. The Q2 fiscal 2026 results proved that.
Revenue hit $22.2 billion, up 48% year over year (YOY)
AI semiconductor sales of $10.8 billion, up 143%.
Free cash flow of $10.26 billion.
Q3 guidance for $29.4 billion in revenue, up 84% YOY
So why is Cohen selling?
The answer is in a single phrase from management’s Q3 AI revenue guidance. The forecast came in at more than $16 billion in Q3, representing more than 200% year-over-year growth.
But that number missed recent elevated investor expectations. When a stock is priced for perfection, and the guidance is merely extraordinary rather than extraordinary enough, the stock gets sold.
More Broadcom:
Broadcom stands to gain from new cloud deal
Morgan Stanley: Broadcom bears are wrong about Google TPU
Jim Cramer says one AI giant holds the key to the market’s next move
Broadcom also faces a new competitive pressure worth noting. Marvell recently expanded its relationship with Google across multiple TPU programs, including accelerators, storage controllers, and networking, according to TheStreet.
That raises the prospect that Google, historically a Broadcom anchor customer, could shift meaningful semiconductor work toward Marvell over time. For a company whose custom-silicon customer relationships drive the long-term revenue story, that dynamic matters.
Cohen is not calling Broadcom a broken business, but is saying the risk-reward at current prices has shifted.
Why is Cohen loading up on Oracle?
Oracle’s 12-month stock chart, down approximately 36% according to Yahoo Finance, tells you everything about why the stock is controversial and nothing about why it could work.
The business has one of the most extraordinary demand signals in enterprise technology right now.
Remaining performance obligations (RPO) grew by $85 billion in a single quarter, from $553 billion to $638 billion
Cloud infrastructure revenue grew 93% YOY to $5.8 billion in Q4
Full-year cloud revenue reached $34.0 billion, up 39%
Total fiscal 2026 revenue was $67.4 billion, up 17%
GAAP EPS grew 34% for the full year.Source: Oracle Q4 and FY 2026 Results
The $638 billion RPO figure is the number that really matters here. It represents contracted revenue that has yet to be recognized, giving investors insight into the company’s revenue visibility and the substantial amount of work already lined up for the years ahead.
The question investors are wrestling with is not whether Oracle has demand. It is whether the capital expenditure required to fulfill that demand will destroy shareholder value before the revenue recognition begins.
Related: Legendary fund manager makes aggressive SpaceX prediction
Free cash flow was -$23.7 billion in fiscal 2026. Oracle plans to raise approximately $40 billion through debt and equity in fiscal 2027 to fund the infrastructure buildout. Those are real costs with real dilution risk.
Cohen’s 188% increase in the position suggests he may be betting that the market is underestimating how quickly Oracle can reach positive cash flow — and how durable that $638 billion backlog could prove to be.
Oracle’s multicloud strategy strengthens that case. By allowing its databases and services to run across AWS, Microsoft Azure, and Google Cloud, Oracle can benefit from cloud growth without having to compete entirely on the same terms as the hyperscalers.
That gives the company a potentially durable position as enterprise workloads continue moving across multiple clouds.
Steve Cohen’s Point72 portfolio sits at $88.13 billion across 3,859 holdings with a 27% turnover rate.Bloomberg / Getty Images
What the switch tells investors about AI
Point72’s top five holdings in Q2, per GuruFocus statistics, were Credo Technology at 1.90%, ASML at 1.25%, Amazon at 1.20%, MKS Instruments at 1.01%, and Arista Networks at 0.99%. The Oracle position at $499 million approaches that concentration level. For a 3,859-holding portfolio, that is meaningful sizing.
The pattern I see in Cohen’s switch is a rotation from AI infrastructure that has already been priced to perfection toward AI infrastructure that has been punished for its capital intensity.
Broadcom’s 48% revenue growth looks impressive on the surface, but with the stock broadly flat after guidance failed to clear already-elevated expectations, the valuation leaves little room for disappointment.
On a risk-adjusted basis, investors are still paying a hefty price for that growth. Oracle down 39% over 12 months with $638 billion in backlog and 93% cloud infrastructure growth is, in Cohen’s read, mispriced.
My read of the trade is directionally coherent. After rewarding AI chips and custom silicon in 2025, the market is now asking who will capture the infrastructure boom in the long term.
Oracle’s $638 billion backlog is a powerful demand signal — and Cohen is betting the path to free cash flow won’t be as painful as bears expect.
Related: Billionaire investor makes Amazon his biggest stock bet