Every three months, the investment world goes a bit mad for the quarterly corporate reporting season. It gives investors an up-to-date picture of how companies are faring, especially in volatile times. But it also encourages short-term thinking and vast duplication in previewing, collating, and instantly analysing the numbers.

For Nvidia, Bloomberg lists no fewer than 95 research analysts, of whom 95% are buyers. There is one lonely seller! And that’s before the myriad bloggers and podcasters chip in. Yes, the volume gets turned up to 11 for these few weeks.

Strong earnings momentum

The build-up to this earnings season has been encouraging. It’s rare to see earnings revisions – analysts’ upgrades to expected company profits – rise so strongly at this stage of a market or economic cycle. The current global positive balance of 20 (number of upgrades versus number of downgrades) has only been surpassed twice in the last decade: during Trump’s 2017–18 corporate tax cuts and the post-Covid recovery in 2021. This time, the main component is the once-in-a-generation AI-related investment boom. 

Earnings have grown strongly since the current bull market cycle began in late 2022. Indeed, they explain about three-quarters of global equity market returns since then. That compares reported 2023 ‘earnings’ for the MSCI All-Countries World index ($38.6) with the latest 2027 estimate ($68.1) – 75% earnings growth versus an index return of 104%. For all the concerns about markets being close to all-time highs, that’s a valid reason – and one that the gloomier prognosticators often overlook. 

As ever, there are plenty of arguments to counter the bulls. The key point is that AI-related capex – capital spending on data centres, chips, and other infrastructure – may prove unsustainable, as I discussed last week. The jury will remain out on that for a while: there’s an understandable lag between investment and widespread adoption, but the trends remain positive.

Meanwhile, the profits accruing to the recipients of the spending aren’t yet being recognised in the profit-and-loss accounts of the spenders, mainly the data-centre-building hyperscalers. Instead, the spending is capitalised on their balance sheets and then depreciated over time, under normal accounting conventions. Still, the scale is remarkable: more than $5tn by 2030, even on conservative forecasts.

Travelling and arriving

Then there is the huge windfall being gathered by makers of memory chips, which are in short supply. Last week, the South Korean company Samsung reported a 1,800% year-on-year increase in profits. Even so, it fell foul of the well-known financial market phenomenon of ‘travelling and arriving’ – when a share price rises ahead of expected good news, only to fall once that news arrives. The good results were so well anticipated, and its shares were so buoyed by a speculative mania in South Korea, that they fell 16% in two days (and have declined further since). Persistent demand and lengthy supply lags should grant the main chipmakers license to print money for at least a couple more years. But we also recognise that markets discount the future well in advance, and so will always be looking for a reason to call the top. It leads to greater volatility in markets that are already structurally prone to wilder short-term gyrations. The key for us is to back what we believe are sustainable, longer-term trends.

Gr-AI-t expectations

Looking at the US, where the tone will be set, we should expect some big share price movements. Speculative leveraged trading vehicles, herd behaviour, and easy access to products and platforms have amplified share price volatility on earnings days. Some of the intra-day moves may shock, but they shouldn’t surprise. We don’t want to be too influenced by short-term share price movements. 

S&P 500 EPS growth was 27% in Q1, or 17% when excluding some one-off private stake revaluations (OpenAI/Anthropic). Still five percentage points ahead of estimates. The consensus forecast expects 22% year-on-year growth in Q2. 

AI infrastructure stocks are expected to contribute nearly 60% of the growth this quarter, with Micron and NVIDIA accounting for more than 40% together. The top 10 account for almost three-quarters of Q2 growth and 58% of this calendar year. The non-AI balance in the top 10 comes from four Energy stocks. We should be grateful if, as currently forecast, the Energy stocks contribute nothing to EPS growth in 2027. If they do, something will have gone horribly wrong elsewhere.

Set fair? 

Returning to the bigger picture, earnings growth seems set fair. For the MSCI ACWI, forecasts are for 27.6% year-on-year in 2026, followed by 17.3% and 12.3% in 2027 and 2028, respectively. A curmudgeon would note the slowdown, but it’s still growth, nonetheless. The semiconductor-related earnings uplift may be sustainable, but it’s probably not repeatable. It’s hard to see equity markets falling if this is the outcome, although some derating – investors paying a lower multiple for the same earnings – wouldn’t come as a surprise. Based on these growth expectations, the price/earnings ratio falls from 19.3x in 2026 to 16.4x in 2027 and 14.6x in 2028.

Less boom, still room

MSCI ACWI earnings growth is forecast to slow after 2026

That begs the question, what might upset this rosy scenario? Some sort of AI shock is the biggest risk, at least in terms of reversing current trends and sentiment. This could be a function of underwhelming demand, maybe as consumers baulk at paying a high enough price to cover the delivery costs. Or it could be a supply shock if, for example, open-source models (mainly from China?) flood the market, or if there is a technology breakthrough that renders existing data centre capacity obsolete. We call these low-probability ‘tail risks’, but they still need monitoring. An optimist would counter that a cheaper and more abundant supply would invoke Jevons’ Paradox, whereby consumption expands to new areas as the ‘commodity’ is more affordable. 

The other main risk is a more conventional economic slowdown or even a recession. Such an event requires a trigger, typically a meaningful tightening of monetary policy. Polymarket’s betting probability for a US recession in 2026 is a lowly 11%. The Bloomberg consensus for the next twelve months has a 25% probability. I think we can safely say that a recession isn’t priced in. Interest rate rises are normally used to dampen inflation (and inflation expectations). Inflation is stickier than central bankers would like, and is being nudged by oil price swings as Middle East news breaks. But market-derived longer-term expectations remain relatively subdued, partly because central bankers keep talking tough on prices. Part of their art is to threaten the big stick but never use it. For now, we remain in the ‘no recession’ camp, but our opinions are constantly under review.