PSI Fund That Beat SOXX by Underweighting Nvidia Now Owns the Part of the Sector That Is Breaking

The Invesco Semiconductors ETF carries a 76.91% year-to-date return and a 131% twelve-month gain as of August 19, with approximately $2.6 billion in assets across 30 holdings. Total return over the trailing year, including dividends, reached 150.93% at one measurement point in July.

Those are the numbers that will appear in every month-end screen. They describe a fund that has already done its work.

The path was violent in both directions. PSI gained 102.37% from December 31, 2025 through July 6, 2026 — a $10,000 position on the last trading day of 2025 was worth approximately $20,496 by the May 26 close. Then the fund gave back a substantial portion of that, with the year-to-date figure falling from 102.37% in early July to 76.91% by August 19. It fell 9.40% in a single session on July 2, closing at $158.01.

That is a 25-percentage-point swing in the year-to-date number across six weeks, in a fund that holds nothing but U.S. semiconductor stocks.

The sector context into month-end is deteriorating. The VanEck Semiconductor ETF fell more than 3% during the final full week of August and dropped more than 3% on Friday alone. Marvell Technology lost 10.3% on Friday to close at $216.62. Applied Materials — PSI's largest holding at 6.87% — fell more than 5% across the week. Nvidia declined 4.45% on Friday to roughly $217.55.

Meanwhile the Nasdaq Composite lost 0.52% Friday to 26,402.42, the S&P 500 fell 0.25% to 7,711.76, and Federal Reserve Chair Kevin Warsh pushed September hike odds from 35.4% to 59.7%.

Monday's premarket offers a marginal stabilisation. Intel climbed 1.3% and ranked as the most actively traded name before the bell. Nvidia advanced 0.6%. South Korea's Kospi fell 1.2% overnight with Samsung Electronics and SK Hynix dragging.

The thesis for this forecast: PSI's outperformance came from owning equipment and memory rather than the GPU franchise, and that is exactly the part of the semiconductor complex now repricing hardest. A quantitative momentum strategy at 76.91% year to date into a sector that just broke is a fund whose methodology is about to be tested.

The Semis Broke Last Week and Nvidia Beat Anyway

The most instructive event of the month for any semiconductor fund was a blowout earnings report that the sector could not hold.

Nvidia reported fiscal second-quarter revenue of $96.2 billion, up 106% year over year, against a $92.3 billion consensus. Adjusted earnings came in at $2.22 per share, up 111.4% from $1.05, versus $2.09 expected. Data center led, edge computing generated $7.2 billion up 27%, and gaming grew 46% to $7.92 billion. The company guided fiscal 2028 revenue growth toward 70% against the 44% analysts had modeled.

The stock fell 1.2% after hours to $207.06, then jumped 8.7% on Thursday, then gave back 4.45% on Friday.

The VanEck Semiconductor ETF fell more than 3% on the week despite that. The iShares Semiconductor ETF dropped more than 3% on Friday alone.

The drag came from the second tier — which is where PSI's weight sits. Marvell Technology fell 10.3% to $216.62 after reporting revenue of $2.74 billion up 37%, earnings of 94 cents against a 93-cent consensus, and a raised fiscal 2028 outlook of roughly $18 billion representing about 50% growth, up from a prior $16.5 billion forecast. It entered the print up 184% on the year and lost a tenth of its market value on a beat-and-raise.

Applied Materials lost more than 5% across the week. Synopsys and Cadence Design Systems ran the other way, rising more than 11% and 6%.

Micron closed near $932.86 after a 13% August advance. It is PSI's fourth-largest holding at 5.57%.

The lesson embedded in that week is the one that matters for a momentum-selected fund. Beating expectations is no longer sufficient when the whole complex is being repriced on the discount rate. With the thirty-year Treasury at 5.21%, the ten-year at 4.72% and September hike odds at 58%, semiconductor multiples are compressing regardless of what the companies report.

PSI holds 30 names selected substantially on price and earnings momentum. Both inputs are now working against it.

Nvidia Is Only 5.89% and That Was the Whole Advantage

The structural feature that produced PSI's outperformance is its refusal to concentrate in the sector's largest company.

Nvidia is weighted at 5.89% of the fund. In market-cap-weighted semiconductor ETFs, that position typically runs three to four times larger.

The top holdings as of July 1 were Applied Materials at 6.87%, KLA Corporation at 6.48%, Lam Research at 5.80%, Micron Technology at 5.57%, and Advanced Micro Devices at 4.98% — with Nvidia sitting below several of them.

Equipment manufacturers and memory suppliers together account for more than 30.3% of the fund's top ten holdings: Applied Materials, Lam Research, KLA, Cohu, Onto Innovation on the equipment side, and Micron on memory.

That composition is why PSI returned 131% over twelve months against the iShares Semiconductor ETF's 112%. The AI capital expenditure cycle spends its money on wafer fabrication equipment and high-bandwidth memory before it spends it on GPUs, and PSI captured the upstream leverage.

It is also why the fund is now more exposed than its peers, not less.

Equipment orders are the most cyclical line in the entire semiconductor supply chain. They lead revenue by four to six quarters and they collapse first when capital expenditure guidance softens. Memory is worse — it is a commodity with a boom-bust pricing cycle that Micron's 13% August gain and $932.86 share price reflect at the top of the range rather than the bottom.

Underweighting Nvidia was the right call while the upstream levered the downstream. If the AI capex cycle decelerates, the same underweight becomes a liability: Nvidia has contracted revenue visibility that Applied Materials and Lam Research do not.

The fund holds U.S. stocks exclusively, giving pure domestic exposure and no direct participation in TSMC, ASML, SK Hynix or Samsung. SK Hynix's U.S. listing, which raised $29.4 billion, slumped in its first full day of trading — a signal about memory sentiment that PSI feels through Micron rather than directly.

The Intellidex Rules: Momentum, Earnings, Quality, Management, Value

PSI is not a passive index fund and understanding the methodology is the whole analysis.

The fund tracks the Dynamic Semiconductor Intellidex Index, which employs a quantitative methodology evaluating companies across five criteria: price momentum, earnings momentum, quality, management action, and value. From that screen it selects 30 semiconductor-related equities judged to best represent the sector's potential for outperformance.

The fund invests at least 90% of total assets in the index constituents. It is non-diversified.

The weighting structure is the second differentiator. Larger stocks receive roughly 40% of total weight and smaller stocks 60% — the inverse of a market-cap approach. That is why the fund dedicates the majority of assets to mid and small caps and why it carries materially higher volatility than a traditional large-cap semiconductor fund.

Two of the five selection criteria are momentum-based. Price momentum and earnings momentum both look backward, which means the index systematically buys what has already worked and sells what has already broken.

In a sustained sector uptrend, that produces exactly what 2026 delivered: 131% against SOXX's 112%. In a sector inflection, it produces the opposite, buying at the top of the cycle and selling at the bottom.

Invesco has been explicit that the strategy cuts both ways. When semiconductor spending rises and earnings expectations improve, PSI benefits. When momentum fades, the same strategy can produce sharp losses.

The fund launched on June 23, 2005, which gives the methodology a much longer track record than the current AI boom and a documented history across multiple semiconductor cycles. Average annual return since inception is 18.18% — strong, but a fraction of the trailing twelve-month figure.

Net expense ratio runs 0.56% to 0.57%, materially above the 0.35% typical of market-cap semiconductor funds. Morningstar assigns a quantitatively derived Silver Medalist Rating, rated June 30.

The August Rebalance Just Reshuffled the Book

The fund and its underlying index rebalance and reconstitute on a quarterly schedule — February, May, August and November.

August's reconstitution has just occurred, and the timing is consequential.

The rebalance re-runs the five-factor screen and re-selects the 30 constituents. Names that ran hardest into the measurement window score highest on price and earnings momentum. Names that broke score lowest and get dropped.

That means the current portfolio is built on the momentum profile that existed as of the August screening date — before Nvidia's earnings, before Marvell's 10.3% single-day loss, and before Warsh repriced the entire long-duration complex on August 28.

A momentum-selected portfolio rebalanced into the top of a move is the definition of buying high. The next opportunity to correct it is November.

Holdings count moved from 30 to 31 at one measurement point, with the top ten representing 50.19% of assets. Half the fund sits in ten names, and the other half spreads across twenty-one.

The composition question for the November rebalance is whether equipment and memory retain their weight. Applied Materials down more than 5% on the week, Marvell down 10.3% on Friday, and a semiconductor ETF complex down 3%-plus all feed directly into the price momentum factor. If the sector stays under pressure through October, the November reconstitution rotates away from the exposures that produced the 131% return.

That is the methodology working as designed. It is also a guarantee that the fund sells the cycle low if one arrives before November.

For anyone holding PSI as a semiconductor allocation, the practical implication is that you own a portfolio built by a rules engine looking backward at a quarter that has already ended. In a trending market that is an advantage. In an inflection it is a three-month lag.

131% Against SOXX's 112% and the Cost of Getting There

The relative performance record is genuinely strong and it deserves a fair hearing.

As of August 19, PSI gained 131% over the trailing year against the iShares Semiconductor ETF's 112%. That is a 19-percentage-point spread from a fund charging roughly 0.56% versus a market-cap alternative charging less.

Total return including dividends reached 150.93% over one trailing-year measurement in mid-July. Year to date, PSI stood at 76.91% on August 19 after peaking at 102.37% on July 6.

Coverage of the sector has noted that three semiconductor funds quietly outperformed the VanEck Semiconductor ETF over the same period by betting on corners of the chip market most investors ignored. PSI is one of them, and the corners were equipment and high-bandwidth memory.

The cost of that outperformance is volatility, and the numbers are stark. PSI fell 9.40% in a single session on July 2. The year-to-date return dropped 25 percentage points between July 6 and August 19 without the fund's holdings changing materially.

A fund that can lose 9.4% in a day and 25 points of annual return in six weeks is not a core allocation. It is a trading vehicle with an ETF wrapper.

The mid and small-cap tilt drives it. Assigning 60% of weight to smaller companies in the most cyclical industry in technology produces higher beta in both directions, and the AI capex cycle has been the ideal environment for that structure.

The since-inception average annual return of 18.18% across more than twenty years is the honest long-run number. The trailing 131% is a cycle peak reading, and the fund's own literature describes semiconductor investing's cyclical nature explicitly.

Anyone underwriting 131% as a forward expectation is extrapolating the best twelve months in the fund's history.

$35.54 Million In While AUM Fell $169.9 Million

The flow data contains the single most useful observation about current positioning.

Over the trailing five days, PSI recorded $35.54 million of net inflows. Over the same five days, net assets under management fell $169.9 million.

Money came in. Assets fell. The difference — roughly $205 million — is pure mark-to-market loss.

That combination describes buyers stepping into a declining tape rather than fleeing it, which is constructive as a sentiment signal and irrelevant to the price. The fund's holders are adding on weakness and the weakness is winning.

The longer flow series shows genuine accumulation. One-month net flows reached $164.43 million, three-month $210.43 million, six-month $526.28 million, and one-year $748 million. Against three-year net flows of $608.68 million and five-year of $605.54 million, the past twelve months brought in more capital than the preceding four years combined.

The AUM series tells the same story with price included. One-month net AUM change was -$100.27 million and three-month +$75.94 million, but six-month +$1.07 billion and one-year +$1.78 billion.

So a fund that carried a few hundred million in assets historically — market capitalisation was $664.91 million in mid-2025 — now runs approximately $2.6 to $2.86 billion. That is roughly a fourfold increase in eighteen months, split between $748 million of new money and $1 billion-plus of appreciation.

Two implications follow. First, most current holders bought within the last twelve months at prices well above the fund's historical range, which means the average holder's cost basis sits close to current levels and there is little embedded profit cushioning further declines.

Second, an asset base that quadrupled on a momentum strategy in the most cyclical sector in technology is precisely the configuration that produces forced selling when the cycle turns.

A 65.96 P/E and What the Multiple Assumes

Valuation on a semiconductor fund is a cycle question rather than a level question, and this one is expensive on both readings.

PSI carried a price-to-earnings ratio of 65.96 as of July 1. That is a multiple on trailing earnings for a portfolio of companies whose earnings are at or near a cyclical peak.

The standard trap in semiconductor investing is that the sector looks cheapest at the top of the cycle, when earnings are maximised, and most expensive at the bottom. A 65.96 multiple on peak earnings is the opposite configuration — expensive on top of expensive — and it implies the market expects earnings to keep expanding from here.

The evidence for that expectation is real. Nvidia guided fiscal 2028 revenue growth toward 70% against a 44% consensus. Marvell raised its fiscal 2028 outlook to roughly $18 billion from $16.5 billion, about 50% growth. Micron gained 13% in August on memory pricing.

The evidence against it is equally real. Marvell fell 10.3% on that raised guidance. Applied Materials lost more than 5% on the week. The semiconductor complex declined more than 3% during a week in which its dominant company reported 106% revenue growth.

When a sector stops rallying on good news, the multiple is doing the work rather than the earnings.

The macro overlay compounds it. With the two-year Treasury at 4.36%, the ten-year at 4.72% and the thirty-year at 5.21% — and September Fed hike odds at 58% — the discount rate applied to 2028 semiconductor earnings has been rising all month. A 65.96 multiple is a duration bet, and duration is the exact factor being repriced.

The fee structure adds a further drag. At 0.56% to 0.57%, PSI costs roughly 20 basis points more annually than market-cap alternatives, which compounds against the strategy in any period where the quantitative screen does not add value.

Broadcom Reports Wednesday and It Is the Sector's Verdict

The single most consequential scheduled event for semiconductor funds this week arrives Wednesday, September 2.

Broadcom reports alongside Salesforce and Snowflake. It closed Friday down 0.7% near $368.79 after adding 4.5% during Thursday's Nvidia-driven rally.

Broadcom's custom silicon business is the most direct read available on whether hyperscaler AI capital expenditure is broadening beyond Nvidia's merchant GPU franchise or concentrating further into it. That distinction determines the earnings trajectory for every equipment and memory name in PSI's top ten.

The precedent is not encouraging. Marvell — the other major custom-silicon name — lost 10.3% on a beat-and-raise with revenue up 37% and a fiscal 2028 outlook lifted to approximately $18 billion. The market has demonstrated it will punish anything short of perfection in that specific segment.

The rest of the week fills in around it. Medtronic reports Tuesday before the open, with Palo Alto Networks and Zscaler after Tuesday's close. Dell Technologies and lululemon follow Thursday. Dell is the AI server read, and Fabrinet already dropped more than 9% earlier in August on a beat, purely on guided fiscal Q1 expense seasonality.

Roughly 89% of S&P 500 companies had reported as of August 10, which means this week's names carry disproportionate signalling weight with no earnings cushion behind them.

Layered on top: ISM Manufacturing PMI and July JOLTS on Tuesday, ADP on Wednesday, ISM Services and jobless claims on Thursday, and the August employment report on Friday. The Chicago Business Barometer already collapsed 10.5 points to 47.1 in August, the first contraction reading in four months.

A weak payrolls print below 40,000 cuts September hike odds under 45% and lifts every long-duration name including this fund. A print above 130,000 with firm wages pushes odds past 70% and semiconductor multiples compress further.

The Level Map and What the Chart Is Doing

The technical structure is defined by two documented reference points and a large gap between them.

The July 2 close of $158.01 came after a 9.40% single-session decline, and it is the clearest recent marker of what a momentum unwind looks like in this fund. On July 6, PSI stood at +102.37% year to date. By August 19 that figure had fallen to 76.91% — implying a level roughly 12% to 13% below the early-July peak.

That decline occurred while the fund took in $164.43 million over the trailing month, which is the same divergence visible in the five-day data: inflows against falling assets.

The sector proxies give the near-term direction. The VanEck Semiconductor ETF fell more than 3% on the week and more than 3% on Friday alone. The iShares Semiconductor ETF declined more than 3% Friday. Those two moves map almost directly onto PSI given the overlapping constituents, with PSI's mid-cap tilt implying a slightly larger decline.

Support sits at the level that held during the July drawdown. Resistance is the early-July peak, which requires roughly 13% to reclaim.

The volatility profile is what should govern position sizing. A fund that fell 9.40% in one session, gave back 25 percentage points of annual return in six weeks, and holds 50.19% of assets in ten mid-cap semiconductor names is not a fund that can be managed with tight stops.

The one structural support: five-day net inflows of $35.54 million into a declining tape means the marginal buyer is still present. Flows have been positive across every measurement window from five days to one year.

When those flows turn negative alongside price, the November rebalance will be selling into it.

The Sector Backdrop: Rates, Rotation, and a Broken Week

Everything in this fund is downstream of two variables it does not control.

The first is the discount rate. Semiconductor equities are the longest-duration cash flows in the equity market outside of pre-revenue biotech, and the entire curve moved against them last week. Warsh's Jackson Hole remarks pushed September hike odds from 35.4% to 59.7% and December odds to 80%. The two-year Treasury ripped 11.97 basis points to 4.352%, its highest since July 24. The dollar index rose 0.4% to 99.57.

Morgan Stanley's Mike Wilson has described investors rotating out of some of the year's biggest winning technology trades, expecting hyperscalers to stabilise while semiconductors bear the adjustment. JPMorgan took the other side, calling the chip sector's rough start to the second half a buying opportunity.

The second variable is the AI capital expenditure cycle itself, and the read there is genuinely mixed. Nvidia's 106% revenue growth and 70% forward guidance say spending accelerates. Marvell losing 10.3% on a raised outlook says the market has stopped paying for it.

The month-level performance shows where the money actually went. The S&P 500 technology sector rose almost 6% in August, with Nvidia up more than 8%, Microsoft 10% and Micron 13%. The iShares Expanded Tech-Software ETF gained 16%. Software outperformed semiconductors materially — a rotation within technology rather than out of it.

That rotation is bad for PSI specifically. The fund holds zero software. Its 30 constituents are all engaged principally in the manufacture of semiconductors.

Global signals point the same direction. South Korea's Kospi fell 1.2% overnight with Samsung Electronics and SK Hynix dragging, and SK Hynix slumped in its first full day of U.S. trading after a $29.4 billion raise. Memory sentiment is deteriorating at exactly the moment PSI carries 5.57% in Micron.

PSI ETF Forecast: Downside to the July Base, Upside Needs 13%

PSI enters September up 76.91% year to date with a momentum-selected portfolio rebalanced at the top of the move and a sector that just broke.

The bull case has five legs. The fund returned 131% over twelve months against the iShares Semiconductor ETF's 112%, proving the quantitative screen adds value in a trending sector. Underweighting Nvidia at 5.89% while holding equipment and memory at more than 30.3% of the top ten captured the upstream leverage in the AI capital expenditure cycle. Flows are positive across every window — $35.54 million over five days, $164.43 million over one month, $748 million over one year — with buyers adding into weakness. Nvidia guided fiscal 2028 revenue growth toward 70% against a 44% consensus, and Marvell raised its own outlook to roughly $18 billion. And Morningstar assigns a Silver Medalist Rating on a strategy with a track record dating to June 2005 and an 18.18% since-inception average annual return.

The bear case has five. The fund carries a 65.96 price-to-earnings ratio on peak-cycle earnings while the thirty-year Treasury sits at 5.21% and September hike odds run 58%. The semiconductor complex fell more than 3% in a week when Nvidia reported 106% revenue growth, and Marvell lost 10.3% on a beat-and-raise. Sixty percent of weight sits in mid and small caps in the most cyclical industry in technology, which produced a 9.40% single-session loss on July 2 and a 25-point year-to-date give-back in six weeks. The August rebalance rebuilt the portfolio on momentum measured before the sector broke, and the next correction comes in November. And software gained 16% in August against a technology sector at 6% — the rotation is happening inside tech and PSI holds no software.

The verdict is Hold with a bias to wait. Base case through the November rebalance: a range bounded by the July drawdown low on the downside and roughly 13% above current levels at the early-July peak on the upside. Reclaiming that peak requires a sub-40,000 payrolls print, a Broadcom beat on Wednesday, and Fed hike odds falling below 45%. Losing the July base means the November reconstitution sells equipment and memory at the bottom.

Own the strategy in an uptrend. This is not currently an uptrend.

That's TradingNEWS